drivers/net/bonding/bond_main.c

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  35
  36 #include <linux/kernel.h>
  37 #include <linux/module.h>
  38 #include <linux/types.h>
  39 #include <linux/fcntl.h>
  40 #include <linux/interrupt.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/ioport.h>
  43 #include <linux/in.h>
  44 #include <net/ip.h>
  45 #include <linux/ip.h>
  46 #include <linux/tcp.h>
  47 #include <linux/udp.h>
  48 #include <linux/slab.h>
  49 #include <linux/string.h>
  50 #include <linux/init.h>
  51 #include <linux/timer.h>
  52 #include <linux/socket.h>
  53 #include <linux/ctype.h>
  54 #include <linux/inet.h>
  55 #include <linux/bitops.h>
  56 #include <linux/io.h>
  57 #include <asm/system.h>
  58 #include <asm/dma.h>
  59 #include <linux/uaccess.h>
  60 #include <linux/errno.h>
  61 #include <linux/netdevice.h>
  62 #include <linux/inetdevice.h>
  63 #include <linux/igmp.h>
  64 #include <linux/etherdevice.h>
  65 #include <linux/skbuff.h>
  66 #include <net/sock.h>
  67 #include <linux/rtnetlink.h>
  68 #include <linux/smp.h>
  69 #include <linux/if_ether.h>
  70 #include <net/arp.h>
  71 #include <linux/mii.h>
  72 #include <linux/ethtool.h>
  73 #include <linux/if_vlan.h>
  74 #include <linux/if_bonding.h>
  75 #include <linux/jiffies.h>
  76 #include <linux/preempt.h>
  77 #include <net/route.h>
  78 #include <net/net_namespace.h>
  79 #include <net/netns/generic.h>
  80 #include "bonding.h"
  81 #include "bond_3ad.h"
  82 #include "bond_alb.h"
  83
  84 /*---------------------------- Module parameters ----------------------------*/
  85
  86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  87 #define BOND_LINK_MON_INTERV    0
  88 #define BOND_LINK_ARP_INTERV    0
  89
  90 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  91 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  92 static int num_grat_arp = 1;
  93 static int num_unsol_na = 1;
  94 static int miimon       = BOND_LINK_MON_INTERV;
  95 static int updelay;
  96 static int downdelay;
  97 static int use_carrier  = 1;
  98 static char *mode;
  99 static char *primary;
 100 static char *primary_reselect;
 101 static char *lacp_rate;
 102 static char *ad_select;
 103 static char *xmit_hash_policy;
 104 static int arp_interval = BOND_LINK_ARP_INTERV;
 105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 106 static char *arp_validate;
 107 static char *fail_over_mac;
 108 static int all_slaves_active = 0;
 109 static struct bond_params bonding_defaults;
 110 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 111
 112 module_param(max_bonds, int, 0);
 113 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 114 module_param(tx_queues, int, 0);
 115 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 116 module_param(num_grat_arp, int, 0644);
 117 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
 118 module_param(num_unsol_na, int, 0644);
 119 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
 120 module_param(miimon, int, 0);
 121 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 122 module_param(updelay, int, 0);
 123 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 124 module_param(downdelay, int, 0);
 125 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 126                             "in milliseconds");
 127 module_param(use_carrier, int, 0);
 128 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 129                               "0 for off, 1 for on (default)");
 130 module_param(mode, charp, 0);
 131 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
 132                        "1 for active-backup, 2 for balance-xor, "
 133                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 134                        "6 for balance-alb");
 135 module_param(primary, charp, 0);
 136 MODULE_PARM_DESC(primary, "Primary network device to use");
 137 module_param(primary_reselect, charp, 0);
 138 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 139                                    "once it comes up; "
 140                                    "0 for always (default), "
 141                                    "1 for only if speed of primary is "
 142                                    "better, "
 143                                    "2 for only on active slave "
 144                                    "failure");
 145 module_param(lacp_rate, charp, 0);
 146 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
 147                             "(slow/fast)");
 148 module_param(ad_select, charp, 0);
 149 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
 150 module_param(xmit_hash_policy, charp, 0);
 151 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
 152                                    ", 1 for layer 3+4");
 153 module_param(arp_interval, int, 0);
 154 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 155 module_param_array(arp_ip_target, charp, NULL, 0);
 156 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 157 module_param(arp_validate, charp, 0);
 158 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
 159 module_param(fail_over_mac, charp, 0);
 160 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
 161 module_param(all_slaves_active, int, 0);
 162 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 163                                      "by setting active flag for all slaves.  "
 164                                      "0 for never (default), 1 for always.");
 165 module_param(resend_igmp, int, 0);
 166 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on link failure");
 167
 168 /*----------------------------- Global variables ----------------------------*/
 169
 170 #ifdef CONFIG_NET_POLL_CONTROLLER
 171 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 172 #endif
 173
 174 int bond_net_id __read_mostly;
 175
 176 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 177 static int arp_ip_count;
 178 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 179 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 180 static int lacp_fast;
 181
 182 const struct bond_parm_tbl bond_lacp_tbl[] = {
 183 {       "slow",         AD_LACP_SLOW},
 184 {       "fast",         AD_LACP_FAST},
 185 {       NULL,           -1},
 186 };
 187
 188 const struct bond_parm_tbl bond_mode_tbl[] = {
 189 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 190 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 191 {       "balance-xor",          BOND_MODE_XOR},
 192 {       "broadcast",            BOND_MODE_BROADCAST},
 193 {       "802.3ad",              BOND_MODE_8023AD},
 194 {       "balance-tlb",          BOND_MODE_TLB},
 195 {       "balance-alb",          BOND_MODE_ALB},
 196 {       NULL,                   -1},
 197 };
 198
 199 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 200 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 201 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 202 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 203 {       NULL,                   -1},
 204 };
 205
 206 const struct bond_parm_tbl arp_validate_tbl[] = {
 207 {       "none",                 BOND_ARP_VALIDATE_NONE},
 208 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 209 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 210 {       "all",                  BOND_ARP_VALIDATE_ALL},
 211 {       NULL,                   -1},
 212 };
 213
 214 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 215 {       "none",                 BOND_FOM_NONE},
 216 {       "active",               BOND_FOM_ACTIVE},
 217 {       "follow",               BOND_FOM_FOLLOW},
 218 {       NULL,                   -1},
 219 };
 220
 221 const struct bond_parm_tbl pri_reselect_tbl[] = {
 222 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 223 {       "better",               BOND_PRI_RESELECT_BETTER},
 224 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 225 {       NULL,                   -1},
 226 };
 227
 228 struct bond_parm_tbl ad_select_tbl[] = {
 229 {       "stable",       BOND_AD_STABLE},
 230 {       "bandwidth",    BOND_AD_BANDWIDTH},
 231 {       "count",        BOND_AD_COUNT},
 232 {       NULL,           -1},
 233 };
 234
 235 /*-------------------------- Forward declarations ---------------------------*/
 236
 237 static void bond_send_gratuitous_arp(struct bonding *bond);
 238 static int bond_init(struct net_device *bond_dev);
 239 static void bond_uninit(struct net_device *bond_dev);
 240
 241 /*---------------------------- General routines -----------------------------*/
 242
 243 const char *bond_mode_name(int mode)
 244 {
 245         static const char *names[] = {
 246                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 247                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 248                 [BOND_MODE_XOR] = "load balancing (xor)",
 249                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 250                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 251                 [BOND_MODE_TLB] = "transmit load balancing",
 252                 [BOND_MODE_ALB] = "adaptive load balancing",
 253         };
 254
 255         if (mode < 0 || mode > BOND_MODE_ALB)
 256                 return "unknown";
 257
 258         return names[mode];
 259 }
 260
 261 /*---------------------------------- VLAN -----------------------------------*/
 262
 263 /**
 264  * bond_add_vlan - add a new vlan id on bond
 265  * @bond: bond that got the notification
 266  * @vlan_id: the vlan id to add
 267  *
 268  * Returns -ENOMEM if allocation failed.
 269  */
 270 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 271 {
 272         struct vlan_entry *vlan;
 273
 274         pr_debug("bond: %s, vlan id %d\n",
 275                  (bond ? bond->dev->name : "None"), vlan_id);
 276
 277         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 278         if (!vlan)
 279                 return -ENOMEM;
 280
 281         INIT_LIST_HEAD(&vlan->vlan_list);
 282         vlan->vlan_id = vlan_id;
 283
 284         write_lock_bh(&bond->lock);
 285
 286         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 287
 288         write_unlock_bh(&bond->lock);
 289
 290         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 291
 292         return 0;
 293 }
 294
 295 /**
 296  * bond_del_vlan - delete a vlan id from bond
 297  * @bond: bond that got the notification
 298  * @vlan_id: the vlan id to delete
 299  *
 300  * returns -ENODEV if @vlan_id was not found in @bond.
 301  */
 302 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 303 {
 304         struct vlan_entry *vlan;
 305         int res = -ENODEV;
 306
 307         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 308
 309         block_netpoll_tx();
 310         write_lock_bh(&bond->lock);
 311
 312         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 313                 if (vlan->vlan_id == vlan_id) {
 314                         list_del(&vlan->vlan_list);
 315
 316                         if (bond_is_lb(bond))
 317                                 bond_alb_clear_vlan(bond, vlan_id);
 318
 319                         pr_debug("removed VLAN ID %d from bond %s\n",
 320                                  vlan_id, bond->dev->name);
 321
 322                         kfree(vlan);
 323
 324                         if (list_empty(&bond->vlan_list) &&
 325                             (bond->slave_cnt == 0)) {
 326                                 /* Last VLAN removed and no slaves, so
 327                                  * restore block on adding VLANs. This will
 328                                  * be removed once new slaves that are not
 329                                  * VLAN challenged will be added.
 330                                  */
 331                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
 332                         }
 333
 334                         res = 0;
 335                         goto out;
 336                 }
 337         }
 338
 339         pr_debug("couldn't find VLAN ID %d in bond %s\n",
 340                  vlan_id, bond->dev->name);
 341
 342 out:
 343         write_unlock_bh(&bond->lock);
 344         unblock_netpoll_tx();
 345         return res;
 346 }
 347
 348 /**
 349  * bond_has_challenged_slaves
 350  * @bond: the bond we're working on
 351  *
 352  * Searches the slave list. Returns 1 if a vlan challenged slave
 353  * was found, 0 otherwise.
 354  *
 355  * Assumes bond->lock is held.
 356  */
 357 static int bond_has_challenged_slaves(struct bonding *bond)
 358 {
 359         struct slave *slave;
 360         int i;
 361
 362         bond_for_each_slave(bond, slave, i) {
 363                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
 364                         pr_debug("found VLAN challenged slave - %s\n",
 365                                  slave->dev->name);
 366                         return 1;
 367                 }
 368         }
 369
 370         pr_debug("no VLAN challenged slaves found\n");
 371         return 0;
 372 }
 373
 374 /**
 375  * bond_next_vlan - safely skip to the next item in the vlans list.
 376  * @bond: the bond we're working on
 377  * @curr: item we're advancing from
 378  *
 379  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 380  * or @curr->next otherwise (even if it is @curr itself again).
 381  *
 382  * Caller must hold bond->lock
 383  */
 384 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 385 {
 386         struct vlan_entry *next, *last;
 387
 388         if (list_empty(&bond->vlan_list))
 389                 return NULL;
 390
 391         if (!curr) {
 392                 next = list_entry(bond->vlan_list.next,
 393                                   struct vlan_entry, vlan_list);
 394         } else {
 395                 last = list_entry(bond->vlan_list.prev,
 396                                   struct vlan_entry, vlan_list);
 397                 if (last == curr) {
 398                         next = list_entry(bond->vlan_list.next,
 399                                           struct vlan_entry, vlan_list);
 400                 } else {
 401                         next = list_entry(curr->vlan_list.next,
 402                                           struct vlan_entry, vlan_list);
 403                 }
 404         }
 405
 406         return next;
 407 }
 408
 409 /**
 410  * bond_dev_queue_xmit - Prepare skb for xmit.
 411  *
 412  * @bond: bond device that got this skb for tx.
 413  * @skb: hw accel VLAN tagged skb to transmit
 414  * @slave_dev: slave that is supposed to xmit this skbuff
 415  */
 416 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 417                         struct net_device *slave_dev)
 418 {
 419         skb->dev = slave_dev;
 420         skb->priority = 1;
 421         if (unlikely(netpoll_tx_running(slave_dev)))
 422                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 423         else
 424                 dev_queue_xmit(skb);
 425
 426         return 0;
 427 }
 428
 429 /*
 430  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
 431  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
 432  * lock because:
 433  * a. This operation is performed in IOCTL context,
 434  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 435  * c. Holding a lock with BH disabled while directly calling a base driver
 436  *    entry point is generally a BAD idea.
 437  *
 438  * The design of synchronization/protection for this operation in the 8021q
 439  * module is good for one or more VLAN devices over a single physical device
 440  * and cannot be extended for a teaming solution like bonding, so there is a
 441  * potential race condition here where a net device from the vlan group might
 442  * be referenced (either by a base driver or the 8021q code) while it is being
 443  * removed from the system. However, it turns out we're not making matters
 444  * worse, and if it works for regular VLAN usage it will work here too.
 445 */
 446
 447 /**
 448  * bond_vlan_rx_register - Propagates registration to slaves
 449  * @bond_dev: bonding net device that got called
 450  * @grp: vlan group being registered
 451  */
 452 static void bond_vlan_rx_register(struct net_device *bond_dev,
 453                                   struct vlan_group *grp)
 454 {
 455         struct bonding *bond = netdev_priv(bond_dev);
 456         struct slave *slave;
 457         int i;
 458
 459         write_lock_bh(&bond->lock);
 460         bond->vlgrp = grp;
 461         write_unlock_bh(&bond->lock);
 462
 463         bond_for_each_slave(bond, slave, i) {
 464                 struct net_device *slave_dev = slave->dev;
 465                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 466
 467                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 468                     slave_ops->ndo_vlan_rx_register) {
 469                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
 470                 }
 471         }
 472 }
 473
 474 /**
 475  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 476  * @bond_dev: bonding net device that got called
 477  * @vid: vlan id being added
 478  */
 479 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 480 {
 481         struct bonding *bond = netdev_priv(bond_dev);
 482         struct slave *slave;
 483         int i, res;
 484
 485         bond_for_each_slave(bond, slave, i) {
 486                 struct net_device *slave_dev = slave->dev;
 487                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 488
 489                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 490                     slave_ops->ndo_vlan_rx_add_vid) {
 491                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
 492                 }
 493         }
 494
 495         res = bond_add_vlan(bond, vid);
 496         if (res) {
 497                 pr_err("%s: Error: Failed to add vlan id %d\n",
 498                        bond_dev->name, vid);
 499         }
 500 }
 501
 502 /**
 503  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 504  * @bond_dev: bonding net device that got called
 505  * @vid: vlan id being removed
 506  */
 507 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 508 {
 509         struct bonding *bond = netdev_priv(bond_dev);
 510         struct slave *slave;
 511         struct net_device *vlan_dev;
 512         int i, res;
 513
 514         bond_for_each_slave(bond, slave, i) {
 515                 struct net_device *slave_dev = slave->dev;
 516                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 517
 518                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 519                     slave_ops->ndo_vlan_rx_kill_vid) {
 520                         /* Save and then restore vlan_dev in the grp array,
 521                          * since the slave's driver might clear it.
 522                          */
 523                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
 524                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
 525                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
 526                 }
 527         }
 528
 529         res = bond_del_vlan(bond, vid);
 530         if (res) {
 531                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 532                        bond_dev->name, vid);
 533         }
 534 }
 535
 536 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 537 {
 538         struct vlan_entry *vlan;
 539         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 540
 541         if (!bond->vlgrp)
 542                 return;
 543
 544         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 545             slave_ops->ndo_vlan_rx_register)
 546                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
 547
 548         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 549             !(slave_ops->ndo_vlan_rx_add_vid))
 550                 return;
 551
 552         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
 553                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 554 }
 555
 556 static void bond_del_vlans_from_slave(struct bonding *bond,
 557                                       struct net_device *slave_dev)
 558 {
 559         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 560         struct vlan_entry *vlan;
 561         struct net_device *vlan_dev;
 562
 563         if (!bond->vlgrp)
 564                 return;
 565
 566         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 567             !(slave_ops->ndo_vlan_rx_kill_vid))
 568                 goto unreg;
 569
 570         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 571                 if (!vlan->vlan_id)
 572                         continue;
 573                 /* Save and then restore vlan_dev in the grp array,
 574                  * since the slave's driver might clear it.
 575                  */
 576                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
 577                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 578                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
 579         }
 580
 581 unreg:
 582         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 583             slave_ops->ndo_vlan_rx_register)
 584                 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
 585 }
 586
 587 /*------------------------------- Link status -------------------------------*/
 588
 589 /*
 590  * Set the carrier state for the master according to the state of its
 591  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 592  * do special 802.3ad magic.
 593  *
 594  * Returns zero if carrier state does not change, nonzero if it does.
 595  */
 596 static int bond_set_carrier(struct bonding *bond)
 597 {
 598         struct slave *slave;
 599         int i;
 600
 601         if (bond->slave_cnt == 0)
 602                 goto down;
 603
 604         if (bond->params.mode == BOND_MODE_8023AD)
 605                 return bond_3ad_set_carrier(bond);
 606
 607         bond_for_each_slave(bond, slave, i) {
 608                 if (slave->link == BOND_LINK_UP) {
 609                         if (!netif_carrier_ok(bond->dev)) {
 610                                 netif_carrier_on(bond->dev);
 611                                 return 1;
 612                         }
 613                         return 0;
 614                 }
 615         }
 616
 617 down:
 618         if (netif_carrier_ok(bond->dev)) {
 619                 netif_carrier_off(bond->dev);
 620                 return 1;
 621         }
 622         return 0;
 623 }
 624
 625 /*
 626  * Get link speed and duplex from the slave's base driver
 627  * using ethtool. If for some reason the call fails or the
 628  * values are invalid, fake speed and duplex to 100/Full
 629  * and return error.
 630  */
 631 static int bond_update_speed_duplex(struct slave *slave)
 632 {
 633         struct net_device *slave_dev = slave->dev;
 634         struct ethtool_cmd etool;
 635         int res;
 636
 637         /* Fake speed and duplex */
 638         slave->speed = SPEED_100;
 639         slave->duplex = DUPLEX_FULL;
 640
 641         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
 642                 return -1;
 643
 644         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 645         if (res < 0)
 646                 return -1;
 647
 648         switch (etool.speed) {
 649         case SPEED_10:
 650         case SPEED_100:
 651         case SPEED_1000:
 652         case SPEED_10000:
 653                 break;
 654         default:
 655                 return -1;
 656         }
 657
 658         switch (etool.duplex) {
 659         case DUPLEX_FULL:
 660         case DUPLEX_HALF:
 661                 break;
 662         default:
 663                 return -1;
 664         }
 665
 666         slave->speed = etool.speed;
 667         slave->duplex = etool.duplex;
 668
 669         return 0;
 670 }
 671
 672 /*
 673  * if <dev> supports MII link status reporting, check its link status.
 674  *
 675  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 676  * depending upon the setting of the use_carrier parameter.
 677  *
 678  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 679  * can't tell and just pretend it is), or 0, meaning that the link is
 680  * down.
 681  *
 682  * If reporting is non-zero, instead of faking link up, return -1 if
 683  * both ETHTOOL and MII ioctls fail (meaning the device does not
 684  * support them).  If use_carrier is set, return whatever it says.
 685  * It'd be nice if there was a good way to tell if a driver supports
 686  * netif_carrier, but there really isn't.
 687  */
 688 static int bond_check_dev_link(struct bonding *bond,
 689                                struct net_device *slave_dev, int reporting)
 690 {
 691         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 692         int (*ioctl)(struct net_device *, struct ifreq *, int);
 693         struct ifreq ifr;
 694         struct mii_ioctl_data *mii;
 695
 696         if (!reporting && !netif_running(slave_dev))
 697                 return 0;
 698
 699         if (bond->params.use_carrier)
 700                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 701
 702         /* Try to get link status using Ethtool first. */
 703         if (slave_dev->ethtool_ops) {
 704                 if (slave_dev->ethtool_ops->get_link) {
 705                         u32 link;
 706
 707                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 708
 709                         return link ? BMSR_LSTATUS : 0;
 710                 }
 711         }
 712
 713         /* Ethtool can't be used, fallback to MII ioctls. */
 714         ioctl = slave_ops->ndo_do_ioctl;
 715         if (ioctl) {
 716                 /* TODO: set pointer to correct ioctl on a per team member */
 717                 /*       bases to make this more efficient. that is, once  */
 718                 /*       we determine the correct ioctl, we will always    */
 719                 /*       call it and not the others for that team          */
 720                 /*       member.                                           */
 721
 722                 /*
 723                  * We cannot assume that SIOCGMIIPHY will also read a
 724                  * register; not all network drivers (e.g., e100)
 725                  * support that.
 726                  */
 727
 728                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 729                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 730                 mii = if_mii(&ifr);
 731                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 732                         mii->reg_num = MII_BMSR;
 733                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 734                                 return mii->val_out & BMSR_LSTATUS;
 735                 }
 736         }
 737
 738         /*
 739          * If reporting, report that either there's no dev->do_ioctl,
 740          * or both SIOCGMIIREG and get_link failed (meaning that we
 741          * cannot report link status).  If not reporting, pretend
 742          * we're ok.
 743          */
 744         return reporting ? -1 : BMSR_LSTATUS;
 745 }
 746
 747 /*----------------------------- Multicast list ------------------------------*/
 748
 749 /*
 750  * Push the promiscuity flag down to appropriate slaves
 751  */
 752 static int bond_set_promiscuity(struct bonding *bond, int inc)
 753 {
 754         int err = 0;
 755         if (USES_PRIMARY(bond->params.mode)) {
 756                 /* write lock already acquired */
 757                 if (bond->curr_active_slave) {
 758                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 759                                                   inc);
 760                 }
 761         } else {
 762                 struct slave *slave;
 763                 int i;
 764                 bond_for_each_slave(bond, slave, i) {
 765                         err = dev_set_promiscuity(slave->dev, inc);
 766                         if (err)
 767                                 return err;
 768                 }
 769         }
 770         return err;
 771 }
 772
 773 /*
 774  * Push the allmulti flag down to all slaves
 775  */
 776 static int bond_set_allmulti(struct bonding *bond, int inc)
 777 {
 778         int err = 0;
 779         if (USES_PRIMARY(bond->params.mode)) {
 780                 /* write lock already acquired */
 781                 if (bond->curr_active_slave) {
 782                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 783                                                inc);
 784                 }
 785         } else {
 786                 struct slave *slave;
 787                 int i;
 788                 bond_for_each_slave(bond, slave, i) {
 789                         err = dev_set_allmulti(slave->dev, inc);
 790                         if (err)
 791                                 return err;
 792                 }
 793         }
 794         return err;
 795 }
 796
 797 /*
 798  * Add a Multicast address to slaves
 799  * according to mode
 800  */
 801 static void bond_mc_add(struct bonding *bond, void *addr)
 802 {
 803         if (USES_PRIMARY(bond->params.mode)) {
 804                 /* write lock already acquired */
 805                 if (bond->curr_active_slave)
 806                         dev_mc_add(bond->curr_active_slave->dev, addr);
 807         } else {
 808                 struct slave *slave;
 809                 int i;
 810
 811                 bond_for_each_slave(bond, slave, i)
 812                         dev_mc_add(slave->dev, addr);
 813         }
 814 }
 815
 816 /*
 817  * Remove a multicast address from slave
 818  * according to mode
 819  */
 820 static void bond_mc_del(struct bonding *bond, void *addr)
 821 {
 822         if (USES_PRIMARY(bond->params.mode)) {
 823                 /* write lock already acquired */
 824                 if (bond->curr_active_slave)
 825                         dev_mc_del(bond->curr_active_slave->dev, addr);
 826         } else {
 827                 struct slave *slave;
 828                 int i;
 829                 bond_for_each_slave(bond, slave, i) {
 830                         dev_mc_del(slave->dev, addr);
 831                 }
 832         }
 833 }
 834
 835
 836 static void __bond_resend_igmp_join_requests(struct net_device *dev)
 837 {
 838         struct in_device *in_dev;
 839
 840         rcu_read_lock();
 841         in_dev = __in_dev_get_rcu(dev);
 842         if (in_dev)
 843                 ip_mc_rejoin_groups(in_dev);
 844         rcu_read_unlock();
 845 }
 846
 847 /*
 848  * Retrieve the list of registered multicast addresses for the bonding
 849  * device and retransmit an IGMP JOIN request to the current active
 850  * slave.
 851  */
 852 static void bond_resend_igmp_join_requests(struct bonding *bond)
 853 {
 854         struct net_device *vlan_dev;
 855         struct vlan_entry *vlan;
 856
 857         read_lock(&bond->lock);
 858
 859         /* rejoin all groups on bond device */
 860         __bond_resend_igmp_join_requests(bond->dev);
 861
 862         /* rejoin all groups on vlan devices */
 863         if (bond->vlgrp) {
 864                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 865                         vlan_dev = vlan_group_get_device(bond->vlgrp,
 866                                                          vlan->vlan_id);
 867                         if (vlan_dev)
 868                                 __bond_resend_igmp_join_requests(vlan_dev);
 869                 }
 870         }
 871
 872         if (--bond->igmp_retrans > 0)
 873                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 874
 875         read_unlock(&bond->lock);
 876 }
 877
 878 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 879 {
 880         struct bonding *bond = container_of(work, struct bonding,
 881                                                         mcast_work.work);
 882         bond_resend_igmp_join_requests(bond);
 883 }
 884
 885 /*
 886  * flush all members of flush->mc_list from device dev->mc_list
 887  */
 888 static void bond_mc_list_flush(struct net_device *bond_dev,
 889                                struct net_device *slave_dev)
 890 {
 891         struct bonding *bond = netdev_priv(bond_dev);
 892         struct netdev_hw_addr *ha;
 893
 894         netdev_for_each_mc_addr(ha, bond_dev)
 895                 dev_mc_del(slave_dev, ha->addr);
 896
 897         if (bond->params.mode == BOND_MODE_8023AD) {
 898                 /* del lacpdu mc addr from mc list */
 899                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 900
 901                 dev_mc_del(slave_dev, lacpdu_multicast);
 902         }
 903 }
 904
 905 /*--------------------------- Active slave change ---------------------------*/
 906
 907 /*
 908  * Update the mc list and multicast-related flags for the new and
 909  * old active slaves (if any) according to the multicast mode, and
 910  * promiscuous flags unconditionally.
 911  */
 912 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 913                          struct slave *old_active)
 914 {
 915         struct netdev_hw_addr *ha;
 916
 917         if (!USES_PRIMARY(bond->params.mode))
 918                 /* nothing to do -  mc list is already up-to-date on
 919                  * all slaves
 920                  */
 921                 return;
 922
 923         if (old_active) {
 924                 if (bond->dev->flags & IFF_PROMISC)
 925                         dev_set_promiscuity(old_active->dev, -1);
 926
 927                 if (bond->dev->flags & IFF_ALLMULTI)
 928                         dev_set_allmulti(old_active->dev, -1);
 929
 930                 netdev_for_each_mc_addr(ha, bond->dev)
 931                         dev_mc_del(old_active->dev, ha->addr);
 932         }
 933
 934         if (new_active) {
 935                 /* FIXME: Signal errors upstream. */
 936                 if (bond->dev->flags & IFF_PROMISC)
 937                         dev_set_promiscuity(new_active->dev, 1);
 938
 939                 if (bond->dev->flags & IFF_ALLMULTI)
 940                         dev_set_allmulti(new_active->dev, 1);
 941
 942                 netdev_for_each_mc_addr(ha, bond->dev)
 943                         dev_mc_add(new_active->dev, ha->addr);
 944         }
 945 }
 946
 947 /*
 948  * bond_do_fail_over_mac
 949  *
 950  * Perform special MAC address swapping for fail_over_mac settings
 951  *
 952  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 953  */
 954 static void bond_do_fail_over_mac(struct bonding *bond,
 955                                   struct slave *new_active,
 956                                   struct slave *old_active)
 957         __releases(&bond->curr_slave_lock)
 958         __releases(&bond->lock)
 959         __acquires(&bond->lock)
 960         __acquires(&bond->curr_slave_lock)
 961 {
 962         u8 tmp_mac[ETH_ALEN];
 963         struct sockaddr saddr;
 964         int rv;
 965
 966         switch (bond->params.fail_over_mac) {
 967         case BOND_FOM_ACTIVE:
 968                 if (new_active)
 969                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 970                                new_active->dev->addr_len);
 971                 break;
 972         case BOND_FOM_FOLLOW:
 973                 /*
 974                  * if new_active && old_active, swap them
 975                  * if just old_active, do nothing (going to no active slave)
 976                  * if just new_active, set new_active to bond's MAC
 977                  */
 978                 if (!new_active)
 979                         return;
 980
 981                 write_unlock_bh(&bond->curr_slave_lock);
 982                 read_unlock(&bond->lock);
 983
 984                 if (old_active) {
 985                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 986                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 987                                ETH_ALEN);
 988                         saddr.sa_family = new_active->dev->type;
 989                 } else {
 990                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 991                         saddr.sa_family = bond->dev->type;
 992                 }
 993
 994                 rv = dev_set_mac_address(new_active->dev, &saddr);
 995                 if (rv) {
 996                         pr_err("%s: Error %d setting MAC of slave %s\n",
 997                                bond->dev->name, -rv, new_active->dev->name);
 998                         goto out;
 999                 }
1000
1001                 if (!old_active)
1002                         goto out;
1003
1004                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1005                 saddr.sa_family = old_active->dev->type;
1006
1007                 rv = dev_set_mac_address(old_active->dev, &saddr);
1008                 if (rv)
1009                         pr_err("%s: Error %d setting MAC of slave %s\n",
1010                                bond->dev->name, -rv, new_active->dev->name);
1011 out:
1012                 read_lock(&bond->lock);
1013                 write_lock_bh(&bond->curr_slave_lock);
1014                 break;
1015         default:
1016                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1017                        bond->dev->name, bond->params.fail_over_mac);
1018                 break;
1019         }
1020
1021 }
1022
1023 static bool bond_should_change_active(struct bonding *bond)
1024 {
1025         struct slave *prim = bond->primary_slave;
1026         struct slave *curr = bond->curr_active_slave;
1027
1028         if (!prim || !curr || curr->link != BOND_LINK_UP)
1029                 return true;
1030         if (bond->force_primary) {
1031                 bond->force_primary = false;
1032                 return true;
1033         }
1034         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1035             (prim->speed < curr->speed ||
1036              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1037                 return false;
1038         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1039                 return false;
1040         return true;
1041 }
1042
1043 /**
1044  * find_best_interface - select the best available slave to be the active one
1045  * @bond: our bonding struct
1046  *
1047  * Warning: Caller must hold curr_slave_lock for writing.
1048  */
1049 static struct slave *bond_find_best_slave(struct bonding *bond)
1050 {
1051         struct slave *new_active, *old_active;
1052         struct slave *bestslave = NULL;
1053         int mintime = bond->params.updelay;
1054         int i;
1055
1056         new_active = bond->curr_active_slave;
1057
1058         if (!new_active) { /* there were no active slaves left */
1059                 if (bond->slave_cnt > 0)   /* found one slave */
1060                         new_active = bond->first_slave;
1061                 else
1062                         return NULL; /* still no slave, return NULL */
1063         }
1064
1065         if ((bond->primary_slave) &&
1066             bond->primary_slave->link == BOND_LINK_UP &&
1067             bond_should_change_active(bond)) {
1068                 new_active = bond->primary_slave;
1069         }
1070
1071         /* remember where to stop iterating over the slaves */
1072         old_active = new_active;
1073
1074         bond_for_each_slave_from(bond, new_active, i, old_active) {
1075                 if (new_active->link == BOND_LINK_UP) {
1076                         return new_active;
1077                 } else if (new_active->link == BOND_LINK_BACK &&
1078                            IS_UP(new_active->dev)) {
1079                         /* link up, but waiting for stabilization */
1080                         if (new_active->delay < mintime) {
1081                                 mintime = new_active->delay;
1082                                 bestslave = new_active;
1083                         }
1084                 }
1085         }
1086
1087         return bestslave;
1088 }
1089
1090 /**
1091  * change_active_interface - change the active slave into the specified one
1092  * @bond: our bonding struct
1093  * @new: the new slave to make the active one
1094  *
1095  * Set the new slave to the bond's settings and unset them on the old
1096  * curr_active_slave.
1097  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1098  *
1099  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1100  * because it is apparently the best available slave we have, even though its
1101  * updelay hasn't timed out yet.
1102  *
1103  * If new_active is not NULL, caller must hold bond->lock for read and
1104  * curr_slave_lock for write_bh.
1105  */
1106 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1107 {
1108         struct slave *old_active = bond->curr_active_slave;
1109
1110         if (old_active == new_active)
1111                 return;
1112
1113         if (new_active) {
1114                 new_active->jiffies = jiffies;
1115
1116                 if (new_active->link == BOND_LINK_BACK) {
1117                         if (USES_PRIMARY(bond->params.mode)) {
1118                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1119                                         bond->dev->name, new_active->dev->name,
1120                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1121                         }
1122
1123                         new_active->delay = 0;
1124                         new_active->link = BOND_LINK_UP;
1125
1126                         if (bond->params.mode == BOND_MODE_8023AD)
1127                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1128
1129                         if (bond_is_lb(bond))
1130                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1131                 } else {
1132                         if (USES_PRIMARY(bond->params.mode)) {
1133                                 pr_info("%s: making interface %s the new active one.\n",
1134                                         bond->dev->name, new_active->dev->name);
1135                         }
1136                 }
1137         }
1138
1139         if (USES_PRIMARY(bond->params.mode))
1140                 bond_mc_swap(bond, new_active, old_active);
1141
1142         if (bond_is_lb(bond)) {
1143                 bond_alb_handle_active_change(bond, new_active);
1144                 if (old_active)
1145                         bond_set_slave_inactive_flags(old_active);
1146                 if (new_active)
1147                         bond_set_slave_active_flags(new_active);
1148         } else {
1149                 bond->curr_active_slave = new_active;
1150         }
1151
1152         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1153                 if (old_active)
1154                         bond_set_slave_inactive_flags(old_active);
1155
1156                 if (new_active) {
1157                         bond_set_slave_active_flags(new_active);
1158
1159                         if (bond->params.fail_over_mac)
1160                                 bond_do_fail_over_mac(bond, new_active,
1161                                                       old_active);
1162
1163                         if (netif_running(bond->dev)) {
1164                                 bond->send_grat_arp = bond->params.num_grat_arp;
1165                                 bond_send_gratuitous_arp(bond);
1166
1167                                 bond->send_unsol_na = bond->params.num_unsol_na;
1168                                 bond_send_unsolicited_na(bond);
1169                         }
1170
1171                         write_unlock_bh(&bond->curr_slave_lock);
1172                         read_unlock(&bond->lock);
1173
1174                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1175
1176                         read_lock(&bond->lock);
1177                         write_lock_bh(&bond->curr_slave_lock);
1178                 }
1179         }
1180
1181         /* resend IGMP joins since active slave has changed or
1182          * all were sent on curr_active_slave */
1183         if (((USES_PRIMARY(bond->params.mode) && new_active) ||
1184              bond->params.mode == BOND_MODE_ROUNDROBIN) &&
1185             netif_running(bond->dev)) {
1186                 bond->igmp_retrans = bond->params.resend_igmp;
1187                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1188         }
1189 }
1190
1191 /**
1192  * bond_select_active_slave - select a new active slave, if needed
1193  * @bond: our bonding struct
1194  *
1195  * This functions should be called when one of the following occurs:
1196  * - The old curr_active_slave has been released or lost its link.
1197  * - The primary_slave has got its link back.
1198  * - A slave has got its link back and there's no old curr_active_slave.
1199  *
1200  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1201  */
1202 void bond_select_active_slave(struct bonding *bond)
1203 {
1204         struct slave *best_slave;
1205         int rv;
1206
1207         best_slave = bond_find_best_slave(bond);
1208         if (best_slave != bond->curr_active_slave) {
1209                 bond_change_active_slave(bond, best_slave);
1210                 rv = bond_set_carrier(bond);
1211                 if (!rv)
1212                         return;
1213
1214                 if (netif_carrier_ok(bond->dev)) {
1215                         pr_info("%s: first active interface up!\n",
1216                                 bond->dev->name);
1217                 } else {
1218                         pr_info("%s: now running without any active interface !\n",
1219                                 bond->dev->name);
1220                 }
1221         }
1222 }
1223
1224 /*--------------------------- slave list handling ---------------------------*/
1225
1226 /*
1227  * This function attaches the slave to the end of list.
1228  *
1229  * bond->lock held for writing by caller.
1230  */
1231 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1232 {
1233         if (bond->first_slave == NULL) { /* attaching the first slave */
1234                 new_slave->next = new_slave;
1235                 new_slave->prev = new_slave;
1236                 bond->first_slave = new_slave;
1237         } else {
1238                 new_slave->next = bond->first_slave;
1239                 new_slave->prev = bond->first_slave->prev;
1240                 new_slave->next->prev = new_slave;
1241                 new_slave->prev->next = new_slave;
1242         }
1243
1244         bond->slave_cnt++;
1245 }
1246
1247 /*
1248  * This function detaches the slave from the list.
1249  * WARNING: no check is made to verify if the slave effectively
1250  * belongs to <bond>.
1251  * Nothing is freed on return, structures are just unchained.
1252  * If any slave pointer in bond was pointing to <slave>,
1253  * it should be changed by the calling function.
1254  *
1255  * bond->lock held for writing by caller.
1256  */
1257 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1258 {
1259         if (slave->next)
1260                 slave->next->prev = slave->prev;
1261
1262         if (slave->prev)
1263                 slave->prev->next = slave->next;
1264
1265         if (bond->first_slave == slave) { /* slave is the first slave */
1266                 if (bond->slave_cnt > 1) { /* there are more slave */
1267                         bond->first_slave = slave->next;
1268                 } else {
1269                         bond->first_slave = NULL; /* slave was the last one */
1270                 }
1271         }
1272
1273         slave->next = NULL;
1274         slave->prev = NULL;
1275         bond->slave_cnt--;
1276 }
1277
1278 #ifdef CONFIG_NET_POLL_CONTROLLER
1279 static inline int slave_enable_netpoll(struct slave *slave)
1280 {
1281         struct netpoll *np;
1282         int err = 0;
1283
1284         np = kzalloc(sizeof(*np), GFP_KERNEL);
1285         err = -ENOMEM;
1286         if (!np)
1287                 goto out;
1288
1289         np->dev = slave->dev;
1290         err = __netpoll_setup(np);
1291         if (err) {
1292                 kfree(np);
1293                 goto out;
1294         }
1295         slave->np = np;
1296 out:
1297         return err;
1298 }
1299 static inline void slave_disable_netpoll(struct slave *slave)
1300 {
1301         struct netpoll *np = slave->np;
1302
1303         if (!np)
1304                 return;
1305
1306         slave->np = NULL;
1307         synchronize_rcu_bh();
1308         __netpoll_cleanup(np);
1309         kfree(np);
1310 }
1311 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1312 {
1313         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1314                 return false;
1315         if (!slave_dev->netdev_ops->ndo_poll_controller)
1316                 return false;
1317         return true;
1318 }
1319
1320 static void bond_poll_controller(struct net_device *bond_dev)
1321 {
1322 }
1323
1324 static void __bond_netpoll_cleanup(struct bonding *bond)
1325 {
1326         struct slave *slave;
1327         int i;
1328
1329         bond_for_each_slave(bond, slave, i)
1330                 if (IS_UP(slave->dev))
1331                         slave_disable_netpoll(slave);
1332 }
1333 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1334 {
1335         struct bonding *bond = netdev_priv(bond_dev);
1336
1337         read_lock(&bond->lock);
1338         __bond_netpoll_cleanup(bond);
1339         read_unlock(&bond->lock);
1340 }
1341
1342 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1343 {
1344         struct bonding *bond = netdev_priv(dev);
1345         struct slave *slave;
1346         int i, err = 0;
1347
1348         read_lock(&bond->lock);
1349         bond_for_each_slave(bond, slave, i) {
1350                 if (!IS_UP(slave->dev))
1351                         continue;
1352                 err = slave_enable_netpoll(slave);
1353                 if (err) {
1354                         __bond_netpoll_cleanup(bond);
1355                         break;
1356                 }
1357         }
1358         read_unlock(&bond->lock);
1359         return err;
1360 }
1361
1362 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1363 {
1364         return bond->dev->npinfo;
1365 }
1366
1367 #else
1368 static inline int slave_enable_netpoll(struct slave *slave)
1369 {
1370         return 0;
1371 }
1372 static inline void slave_disable_netpoll(struct slave *slave)
1373 {
1374 }
1375 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1376 {
1377 }
1378 #endif
1379
1380 /*---------------------------------- IOCTL ----------------------------------*/
1381
1382 static int bond_sethwaddr(struct net_device *bond_dev,
1383                           struct net_device *slave_dev)
1384 {
1385         pr_debug("bond_dev=%p\n", bond_dev);
1386         pr_debug("slave_dev=%p\n", slave_dev);
1387         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1388         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1389         return 0;
1390 }
1391
1392 #define BOND_VLAN_FEATURES \
1393         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1394          NETIF_F_HW_VLAN_FILTER)
1395
1396 /*
1397  * Compute the common dev->feature set available to all slaves.  Some
1398  * feature bits are managed elsewhere, so preserve those feature bits
1399  * on the master device.
1400  */
1401 static int bond_compute_features(struct bonding *bond)
1402 {
1403         struct slave *slave;
1404         struct net_device *bond_dev = bond->dev;
1405         u32 features = bond_dev->features;
1406         u32 vlan_features = 0;
1407         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1408                                                 bond_dev->hard_header_len);
1409         int i;
1410
1411         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1412         features |=  NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1413
1414         if (!bond->first_slave)
1415                 goto done;
1416
1417         features &= ~NETIF_F_ONE_FOR_ALL;
1418
1419         vlan_features = bond->first_slave->dev->vlan_features;
1420         bond_for_each_slave(bond, slave, i) {
1421                 features = netdev_increment_features(features,
1422                                                      slave->dev->features,
1423                                                      NETIF_F_ONE_FOR_ALL);
1424                 vlan_features = netdev_increment_features(vlan_features,
1425                                                         slave->dev->vlan_features,
1426                                                         NETIF_F_ONE_FOR_ALL);
1427                 if (slave->dev->hard_header_len > max_hard_header_len)
1428                         max_hard_header_len = slave->dev->hard_header_len;
1429         }
1430
1431 done:
1432         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1433         bond_dev->features = netdev_fix_features(bond_dev, features);
1434         bond_dev->vlan_features = netdev_fix_features(bond_dev, vlan_features);
1435         bond_dev->hard_header_len = max_hard_header_len;
1436
1437         return 0;
1438 }
1439
1440 static void bond_setup_by_slave(struct net_device *bond_dev,
1441                                 struct net_device *slave_dev)
1442 {
1443         struct bonding *bond = netdev_priv(bond_dev);
1444
1445         bond_dev->header_ops        = slave_dev->header_ops;
1446
1447         bond_dev->type              = slave_dev->type;
1448         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1449         bond_dev->addr_len          = slave_dev->addr_len;
1450
1451         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1452                 slave_dev->addr_len);
1453         bond->setup_by_slave = 1;
1454 }
1455
1456 /* On bonding slaves other than the currently active slave, suppress
1457  * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
1458  * ARP on active-backup slaves with arp_validate enabled.
1459  */
1460 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1461                                             struct slave *slave,
1462                                             struct bonding *bond)
1463 {
1464         if (bond_is_slave_inactive(slave)) {
1465                 if (slave_do_arp_validate(bond, slave) &&
1466                     skb->protocol == __cpu_to_be16(ETH_P_ARP))
1467                         return false;
1468
1469                 if (bond->params.mode == BOND_MODE_ALB &&
1470                     skb->pkt_type != PACKET_BROADCAST &&
1471                     skb->pkt_type != PACKET_MULTICAST)
1472                                 return false;
1473
1474                 if (bond->params.mode == BOND_MODE_8023AD &&
1475                     skb->protocol == __cpu_to_be16(ETH_P_SLOW))
1476                         return false;
1477
1478                 return true;
1479         }
1480         return false;
1481 }
1482
1483 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1484 {
1485         struct sk_buff *skb = *pskb;
1486         struct slave *slave;
1487         struct net_device *bond_dev;
1488         struct bonding *bond;
1489
1490         slave = bond_slave_get_rcu(skb->dev);
1491         bond_dev = ACCESS_ONCE(slave->dev->master);
1492         if (unlikely(!bond_dev))
1493                 return RX_HANDLER_PASS;
1494
1495         skb = skb_share_check(skb, GFP_ATOMIC);
1496         if (unlikely(!skb))
1497                 return RX_HANDLER_CONSUMED;
1498
1499         *pskb = skb;
1500
1501         bond = netdev_priv(bond_dev);
1502
1503         if (bond->params.arp_interval)
1504                 slave->dev->last_rx = jiffies;
1505
1506         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1507                 return RX_HANDLER_EXACT;
1508         }
1509
1510         skb->dev = bond_dev;
1511
1512         if (bond->params.mode == BOND_MODE_ALB &&
1513             bond_dev->priv_flags & IFF_BRIDGE_PORT &&
1514             skb->pkt_type == PACKET_HOST) {
1515
1516                 if (unlikely(skb_cow_head(skb,
1517                                           skb->data - skb_mac_header(skb)))) {
1518                         kfree_skb(skb);
1519                         return RX_HANDLER_CONSUMED;
1520                 }
1521                 memcpy(eth_hdr(skb)->h_dest, bond_dev->dev_addr, ETH_ALEN);
1522         }
1523
1524         return RX_HANDLER_ANOTHER;
1525 }
1526
1527 /* enslave device <slave> to bond device <master> */
1528 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1529 {
1530         struct bonding *bond = netdev_priv(bond_dev);
1531         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1532         struct slave *new_slave = NULL;
1533         struct netdev_hw_addr *ha;
1534         struct sockaddr addr;
1535         int link_reporting;
1536         int old_features = bond_dev->features;
1537         int res = 0;
1538
1539         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1540                 slave_ops->ndo_do_ioctl == NULL) {
1541                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1542                            bond_dev->name, slave_dev->name);
1543         }
1544
1545         /* bond must be initialized by bond_open() before enslaving */
1546         if (!(bond_dev->flags & IFF_UP)) {
1547                 pr_warning("%s: master_dev is not up in bond_enslave\n",
1548                            bond_dev->name);
1549         }
1550
1551         /* already enslaved */
1552         if (slave_dev->flags & IFF_SLAVE) {
1553                 pr_debug("Error, Device was already enslaved\n");
1554                 return -EBUSY;
1555         }
1556
1557         /* vlan challenged mutual exclusion */
1558         /* no need to lock since we're protected by rtnl_lock */
1559         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1560                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1561                 if (bond->vlgrp) {
1562                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1563                                bond_dev->name, slave_dev->name, bond_dev->name);
1564                         return -EPERM;
1565                 } else {
1566                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1567                                    bond_dev->name, slave_dev->name,
1568                                    slave_dev->name, bond_dev->name);
1569                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1570                 }
1571         } else {
1572                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1573                 if (bond->slave_cnt == 0) {
1574                         /* First slave, and it is not VLAN challenged,
1575                          * so remove the block of adding VLANs over the bond.
1576                          */
1577                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1578                 }
1579         }
1580
1581         /*
1582          * Old ifenslave binaries are no longer supported.  These can
1583          * be identified with moderate accuracy by the state of the slave:
1584          * the current ifenslave will set the interface down prior to
1585          * enslaving it; the old ifenslave will not.
1586          */
1587         if ((slave_dev->flags & IFF_UP)) {
1588                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1589                        slave_dev->name);
1590                 res = -EPERM;
1591                 goto err_undo_flags;
1592         }
1593
1594         /* set bonding device ether type by slave - bonding netdevices are
1595          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1596          * there is a need to override some of the type dependent attribs/funcs.
1597          *
1598          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1599          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1600          */
1601         if (bond->slave_cnt == 0) {
1602                 if (bond_dev->type != slave_dev->type) {
1603                         pr_debug("%s: change device type from %d to %d\n",
1604                                  bond_dev->name,
1605                                  bond_dev->type, slave_dev->type);
1606
1607                         res = netdev_bonding_change(bond_dev,
1608                                                     NETDEV_PRE_TYPE_CHANGE);
1609                         res = notifier_to_errno(res);
1610                         if (res) {
1611                                 pr_err("%s: refused to change device type\n",
1612                                        bond_dev->name);
1613                                 res = -EBUSY;
1614                                 goto err_undo_flags;
1615                         }
1616
1617                         /* Flush unicast and multicast addresses */
1618                         dev_uc_flush(bond_dev);
1619                         dev_mc_flush(bond_dev);
1620
1621                         if (slave_dev->type != ARPHRD_ETHER)
1622                                 bond_setup_by_slave(bond_dev, slave_dev);
1623                         else
1624                                 ether_setup(bond_dev);
1625
1626                         netdev_bonding_change(bond_dev,
1627                                               NETDEV_POST_TYPE_CHANGE);
1628                 }
1629         } else if (bond_dev->type != slave_dev->type) {
1630                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1631                        slave_dev->name,
1632                        slave_dev->type, bond_dev->type);
1633                 res = -EINVAL;
1634                 goto err_undo_flags;
1635         }
1636
1637         if (slave_ops->ndo_set_mac_address == NULL) {
1638                 if (bond->slave_cnt == 0) {
1639                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1640                                    bond_dev->name);
1641                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1642                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1643                         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",
1644                                bond_dev->name);
1645                         res = -EOPNOTSUPP;
1646                         goto err_undo_flags;
1647                 }
1648         }
1649
1650         /* If this is the first slave, then we need to set the master's hardware
1651          * address to be the same as the slave's. */
1652         if (is_zero_ether_addr(bond->dev->dev_addr))
1653                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1654                        slave_dev->addr_len);
1655
1656
1657         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1658         if (!new_slave) {
1659                 res = -ENOMEM;
1660                 goto err_undo_flags;
1661         }
1662
1663         /*
1664          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1665          * is set via sysfs or module option if desired.
1666          */
1667         new_slave->queue_id = 0;
1668
1669         /* Save slave's original mtu and then set it to match the bond */
1670         new_slave->original_mtu = slave_dev->mtu;
1671         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1672         if (res) {
1673                 pr_debug("Error %d calling dev_set_mtu\n", res);
1674                 goto err_free;
1675         }
1676
1677         /*
1678          * Save slave's original ("permanent") mac address for modes
1679          * that need it, and for restoring it upon release, and then
1680          * set it to the master's address
1681          */
1682         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1683
1684         if (!bond->params.fail_over_mac) {
1685                 /*
1686                  * Set slave to master's mac address.  The application already
1687                  * set the master's mac address to that of the first slave
1688                  */
1689                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1690                 addr.sa_family = slave_dev->type;
1691                 res = dev_set_mac_address(slave_dev, &addr);
1692                 if (res) {
1693                         pr_debug("Error %d calling set_mac_address\n", res);
1694                         goto err_restore_mtu;
1695                 }
1696         }
1697
1698         res = netdev_set_bond_master(slave_dev, bond_dev);
1699         if (res) {
1700                 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1701                 goto err_restore_mac;
1702         }
1703         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1704                                          new_slave);
1705         if (res) {
1706                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1707                 goto err_unset_master;
1708         }
1709
1710         /* open the slave since the application closed it */
1711         res = dev_open(slave_dev);
1712         if (res) {
1713                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1714                 goto err_unreg_rxhandler;
1715         }
1716
1717         new_slave->dev = slave_dev;
1718         slave_dev->priv_flags |= IFF_BONDING;
1719
1720         if (bond_is_lb(bond)) {
1721                 /* bond_alb_init_slave() must be called before all other stages since
1722                  * it might fail and we do not want to have to undo everything
1723                  */
1724                 res = bond_alb_init_slave(bond, new_slave);
1725                 if (res)
1726                         goto err_close;
1727         }
1728
1729         /* If the mode USES_PRIMARY, then the new slave gets the
1730          * master's promisc (and mc) settings only if it becomes the
1731          * curr_active_slave, and that is taken care of later when calling
1732          * bond_change_active()
1733          */
1734         if (!USES_PRIMARY(bond->params.mode)) {
1735                 /* set promiscuity level to new slave */
1736                 if (bond_dev->flags & IFF_PROMISC) {
1737                         res = dev_set_promiscuity(slave_dev, 1);
1738                         if (res)
1739                                 goto err_close;
1740                 }
1741
1742                 /* set allmulti level to new slave */
1743                 if (bond_dev->flags & IFF_ALLMULTI) {
1744                         res = dev_set_allmulti(slave_dev, 1);
1745                         if (res)
1746                                 goto err_close;
1747                 }
1748
1749                 netif_addr_lock_bh(bond_dev);
1750                 /* upload master's mc_list to new slave */
1751                 netdev_for_each_mc_addr(ha, bond_dev)
1752                         dev_mc_add(slave_dev, ha->addr);
1753                 netif_addr_unlock_bh(bond_dev);
1754         }
1755
1756         if (bond->params.mode == BOND_MODE_8023AD) {
1757                 /* add lacpdu mc addr to mc list */
1758                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1759
1760                 dev_mc_add(slave_dev, lacpdu_multicast);
1761         }
1762
1763         bond_add_vlans_on_slave(bond, slave_dev);
1764
1765         write_lock_bh(&bond->lock);
1766
1767         bond_attach_slave(bond, new_slave);
1768
1769         new_slave->delay = 0;
1770         new_slave->link_failure_count = 0;
1771
1772         bond_compute_features(bond);
1773
1774         write_unlock_bh(&bond->lock);
1775
1776         read_lock(&bond->lock);
1777
1778         new_slave->last_arp_rx = jiffies;
1779
1780         if (bond->params.miimon && !bond->params.use_carrier) {
1781                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1782
1783                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1784                         /*
1785                          * miimon is set but a bonded network driver
1786                          * does not support ETHTOOL/MII and
1787                          * arp_interval is not set.  Note: if
1788                          * use_carrier is enabled, we will never go
1789                          * here (because netif_carrier is always
1790                          * supported); thus, we don't need to change
1791                          * the messages for netif_carrier.
1792                          */
1793                         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",
1794                                bond_dev->name, slave_dev->name);
1795                 } else if (link_reporting == -1) {
1796                         /* unable get link status using mii/ethtool */
1797                         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",
1798                                    bond_dev->name, slave_dev->name);
1799                 }
1800         }
1801
1802         /* check for initial state */
1803         if (!bond->params.miimon ||
1804             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1805                 if (bond->params.updelay) {
1806                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1807                         new_slave->link  = BOND_LINK_BACK;
1808                         new_slave->delay = bond->params.updelay;
1809                 } else {
1810                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1811                         new_slave->link  = BOND_LINK_UP;
1812                 }
1813                 new_slave->jiffies = jiffies;
1814         } else {
1815                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1816                 new_slave->link  = BOND_LINK_DOWN;
1817         }
1818
1819         if (bond_update_speed_duplex(new_slave) &&
1820             (new_slave->link != BOND_LINK_DOWN)) {
1821                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1822                            bond_dev->name, new_slave->dev->name);
1823
1824                 if (bond->params.mode == BOND_MODE_8023AD) {
1825                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1826                                    bond_dev->name);
1827                 }
1828         }
1829
1830         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1831                 /* if there is a primary slave, remember it */
1832                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1833                         bond->primary_slave = new_slave;
1834                         bond->force_primary = true;
1835                 }
1836         }
1837
1838         write_lock_bh(&bond->curr_slave_lock);
1839
1840         switch (bond->params.mode) {
1841         case BOND_MODE_ACTIVEBACKUP:
1842                 bond_set_slave_inactive_flags(new_slave);
1843                 bond_select_active_slave(bond);
1844                 break;
1845         case BOND_MODE_8023AD:
1846                 /* in 802.3ad mode, the internal mechanism
1847                  * will activate the slaves in the selected
1848                  * aggregator
1849                  */
1850                 bond_set_slave_inactive_flags(new_slave);
1851                 /* if this is the first slave */
1852                 if (bond->slave_cnt == 1) {
1853                         SLAVE_AD_INFO(new_slave).id = 1;
1854                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1855                          * can be called only after the mac address of the bond is set
1856                          */
1857                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1858                                             bond->params.lacp_fast);
1859                 } else {
1860                         SLAVE_AD_INFO(new_slave).id =
1861                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1862                 }
1863
1864                 bond_3ad_bind_slave(new_slave);
1865                 break;
1866         case BOND_MODE_TLB:
1867         case BOND_MODE_ALB:
1868                 bond_set_active_slave(new_slave);
1869                 bond_set_slave_inactive_flags(new_slave);
1870                 bond_select_active_slave(bond);
1871                 break;
1872         default:
1873                 pr_debug("This slave is always active in trunk mode\n");
1874
1875                 /* always active in trunk mode */
1876                 bond_set_active_slave(new_slave);
1877
1878                 /* In trunking mode there is little meaning to curr_active_slave
1879                  * anyway (it holds no special properties of the bond device),
1880                  * so we can change it without calling change_active_interface()
1881                  */
1882                 if (!bond->curr_active_slave)
1883                         bond->curr_active_slave = new_slave;
1884
1885                 break;
1886         } /* switch(bond_mode) */
1887
1888         write_unlock_bh(&bond->curr_slave_lock);
1889
1890         bond_set_carrier(bond);
1891
1892 #ifdef CONFIG_NET_POLL_CONTROLLER
1893         slave_dev->npinfo = bond_netpoll_info(bond);
1894         if (slave_dev->npinfo) {
1895                 if (slave_enable_netpoll(new_slave)) {
1896                         read_unlock(&bond->lock);
1897                         pr_info("Error, %s: master_dev is using netpoll, "
1898                                  "but new slave device does not support netpoll.\n",
1899                                  bond_dev->name);
1900                         res = -EBUSY;
1901                         goto err_close;
1902                 }
1903         }
1904 #endif
1905
1906         read_unlock(&bond->lock);
1907
1908         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1909         if (res)
1910                 goto err_close;
1911
1912         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1913                 bond_dev->name, slave_dev->name,
1914                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1915                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1916
1917         /* enslave is successful */
1918         return 0;
1919
1920 /* Undo stages on error */
1921 err_close:
1922         dev_close(slave_dev);
1923
1924 err_unreg_rxhandler:
1925         netdev_rx_handler_unregister(slave_dev);
1926         synchronize_net();
1927
1928 err_unset_master:
1929         netdev_set_bond_master(slave_dev, NULL);
1930
1931 err_restore_mac:
1932         if (!bond->params.fail_over_mac) {
1933                 /* XXX TODO - fom follow mode needs to change master's
1934                  * MAC if this slave's MAC is in use by the bond, or at
1935                  * least print a warning.
1936                  */
1937                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1938                 addr.sa_family = slave_dev->type;
1939                 dev_set_mac_address(slave_dev, &addr);
1940         }
1941
1942 err_restore_mtu:
1943         dev_set_mtu(slave_dev, new_slave->original_mtu);
1944
1945 err_free:
1946         kfree(new_slave);
1947
1948 err_undo_flags:
1949         bond_dev->features = old_features;
1950
1951         return res;
1952 }
1953
1954 /*
1955  * Try to release the slave device <slave> from the bond device <master>
1956  * It is legal to access curr_active_slave without a lock because all the function
1957  * is write-locked.
1958  *
1959  * The rules for slave state should be:
1960  *   for Active/Backup:
1961  *     Active stays on all backups go down
1962  *   for Bonded connections:
1963  *     The first up interface should be left on and all others downed.
1964  */
1965 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1966 {
1967         struct bonding *bond = netdev_priv(bond_dev);
1968         struct slave *slave, *oldcurrent;
1969         struct sockaddr addr;
1970
1971         /* slave is not a slave or master is not master of this slave */
1972         if (!(slave_dev->flags & IFF_SLAVE) ||
1973             (slave_dev->master != bond_dev)) {
1974                 pr_err("%s: Error: cannot release %s.\n",
1975                        bond_dev->name, slave_dev->name);
1976                 return -EINVAL;
1977         }
1978
1979         block_netpoll_tx();
1980         netdev_bonding_change(bond_dev, NETDEV_BONDING_DESLAVE);
1981         write_lock_bh(&bond->lock);
1982
1983         slave = bond_get_slave_by_dev(bond, slave_dev);
1984         if (!slave) {
1985                 /* not a slave of this bond */
1986                 pr_info("%s: %s not enslaved\n",
1987                         bond_dev->name, slave_dev->name);
1988                 write_unlock_bh(&bond->lock);
1989                 unblock_netpoll_tx();
1990                 return -EINVAL;
1991         }
1992
1993         if (!bond->params.fail_over_mac) {
1994                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1995                     bond->slave_cnt > 1)
1996                         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",
1997                                    bond_dev->name, slave_dev->name,
1998                                    slave->perm_hwaddr,
1999                                    bond_dev->name, slave_dev->name);
2000         }
2001
2002         /* Inform AD package of unbinding of slave. */
2003         if (bond->params.mode == BOND_MODE_8023AD) {
2004                 /* must be called before the slave is
2005                  * detached from the list
2006                  */
2007                 bond_3ad_unbind_slave(slave);
2008         }
2009
2010         pr_info("%s: releasing %s interface %s\n",
2011                 bond_dev->name,
2012                 bond_is_active_slave(slave) ? "active" : "backup",
2013                 slave_dev->name);
2014
2015         oldcurrent = bond->curr_active_slave;
2016
2017         bond->current_arp_slave = NULL;
2018
2019         /* release the slave from its bond */
2020         bond_detach_slave(bond, slave);
2021
2022         bond_compute_features(bond);
2023
2024         if (bond->primary_slave == slave)
2025                 bond->primary_slave = NULL;
2026
2027         if (oldcurrent == slave)
2028                 bond_change_active_slave(bond, NULL);
2029
2030         if (bond_is_lb(bond)) {
2031                 /* Must be called only after the slave has been
2032                  * detached from the list and the curr_active_slave
2033                  * has been cleared (if our_slave == old_current),
2034                  * but before a new active slave is selected.
2035                  */
2036                 write_unlock_bh(&bond->lock);
2037                 bond_alb_deinit_slave(bond, slave);
2038                 write_lock_bh(&bond->lock);
2039         }
2040
2041         if (oldcurrent == slave) {
2042                 /*
2043                  * Note that we hold RTNL over this sequence, so there
2044                  * is no concern that another slave add/remove event
2045                  * will interfere.
2046                  */
2047                 write_unlock_bh(&bond->lock);
2048                 read_lock(&bond->lock);
2049                 write_lock_bh(&bond->curr_slave_lock);
2050
2051                 bond_select_active_slave(bond);
2052
2053                 write_unlock_bh(&bond->curr_slave_lock);
2054                 read_unlock(&bond->lock);
2055                 write_lock_bh(&bond->lock);
2056         }
2057
2058         if (bond->slave_cnt == 0) {
2059                 bond_set_carrier(bond);
2060
2061                 /* if the last slave was removed, zero the mac address
2062                  * of the master so it will be set by the application
2063                  * to the mac address of the first slave
2064                  */
2065                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2066
2067                 if (!bond->vlgrp) {
2068                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2069                 } else {
2070                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2071                                    bond_dev->name, bond_dev->name);
2072                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2073                                    bond_dev->name);
2074                 }
2075         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2076                    !bond_has_challenged_slaves(bond)) {
2077                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2078                         bond_dev->name, slave_dev->name, bond_dev->name);
2079                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
2080         }
2081
2082         write_unlock_bh(&bond->lock);
2083         unblock_netpoll_tx();
2084
2085         /* must do this from outside any spinlocks */
2086         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2087
2088         bond_del_vlans_from_slave(bond, slave_dev);
2089
2090         /* If the mode USES_PRIMARY, then we should only remove its
2091          * promisc and mc settings if it was the curr_active_slave, but that was
2092          * already taken care of above when we detached the slave
2093          */
2094         if (!USES_PRIMARY(bond->params.mode)) {
2095                 /* unset promiscuity level from slave */
2096                 if (bond_dev->flags & IFF_PROMISC)
2097                         dev_set_promiscuity(slave_dev, -1);
2098
2099                 /* unset allmulti level from slave */
2100                 if (bond_dev->flags & IFF_ALLMULTI)
2101                         dev_set_allmulti(slave_dev, -1);
2102
2103                 /* flush master's mc_list from slave */
2104                 netif_addr_lock_bh(bond_dev);
2105                 bond_mc_list_flush(bond_dev, slave_dev);
2106                 netif_addr_unlock_bh(bond_dev);
2107         }
2108
2109         netdev_rx_handler_unregister(slave_dev);
2110         synchronize_net();
2111         netdev_set_bond_master(slave_dev, NULL);
2112
2113         slave_disable_netpoll(slave);
2114
2115         /* close slave before restoring its mac address */
2116         dev_close(slave_dev);
2117
2118         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2119                 /* restore original ("permanent") mac address */
2120                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2121                 addr.sa_family = slave_dev->type;
2122                 dev_set_mac_address(slave_dev, &addr);
2123         }
2124
2125         dev_set_mtu(slave_dev, slave->original_mtu);
2126
2127         slave_dev->priv_flags &= ~IFF_BONDING;
2128
2129         kfree(slave);
2130
2131         return 0;  /* deletion OK */
2132 }
2133
2134 /*
2135 * First release a slave and than destroy the bond if no more slaves are left.
2136 * Must be under rtnl_lock when this function is called.
2137 */
2138 static int  bond_release_and_destroy(struct net_device *bond_dev,
2139                                      struct net_device *slave_dev)
2140 {
2141         struct bonding *bond = netdev_priv(bond_dev);
2142         int ret;
2143
2144         ret = bond_release(bond_dev, slave_dev);
2145         if ((ret == 0) && (bond->slave_cnt == 0)) {
2146                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2147                 pr_info("%s: destroying bond %s.\n",
2148                         bond_dev->name, bond_dev->name);
2149                 unregister_netdevice(bond_dev);
2150         }
2151         return ret;
2152 }
2153
2154 /*
2155  * This function releases all slaves.
2156  */
2157 static int bond_release_all(struct net_device *bond_dev)
2158 {
2159         struct bonding *bond = netdev_priv(bond_dev);
2160         struct slave *slave;
2161         struct net_device *slave_dev;
2162         struct sockaddr addr;
2163
2164         write_lock_bh(&bond->lock);
2165
2166         netif_carrier_off(bond_dev);
2167
2168         if (bond->slave_cnt == 0)
2169                 goto out;
2170
2171         bond->current_arp_slave = NULL;
2172         bond->primary_slave = NULL;
2173         bond_change_active_slave(bond, NULL);
2174
2175         while ((slave = bond->first_slave) != NULL) {
2176                 /* Inform AD package of unbinding of slave
2177                  * before slave is detached from the list.
2178                  */
2179                 if (bond->params.mode == BOND_MODE_8023AD)
2180                         bond_3ad_unbind_slave(slave);
2181
2182                 slave_dev = slave->dev;
2183                 bond_detach_slave(bond, slave);
2184
2185                 /* now that the slave is detached, unlock and perform
2186                  * all the undo steps that should not be called from
2187                  * within a lock.
2188                  */
2189                 write_unlock_bh(&bond->lock);
2190
2191                 if (bond_is_lb(bond)) {
2192                         /* must be called only after the slave
2193                          * has been detached from the list
2194                          */
2195                         bond_alb_deinit_slave(bond, slave);
2196                 }
2197
2198                 bond_compute_features(bond);
2199
2200                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2201                 bond_del_vlans_from_slave(bond, slave_dev);
2202
2203                 /* If the mode USES_PRIMARY, then we should only remove its
2204                  * promisc and mc settings if it was the curr_active_slave, but that was
2205                  * already taken care of above when we detached the slave
2206                  */
2207                 if (!USES_PRIMARY(bond->params.mode)) {
2208                         /* unset promiscuity level from slave */
2209                         if (bond_dev->flags & IFF_PROMISC)
2210                                 dev_set_promiscuity(slave_dev, -1);
2211
2212                         /* unset allmulti level from slave */
2213                         if (bond_dev->flags & IFF_ALLMULTI)
2214                                 dev_set_allmulti(slave_dev, -1);
2215
2216                         /* flush master's mc_list from slave */
2217                         netif_addr_lock_bh(bond_dev);
2218                         bond_mc_list_flush(bond_dev, slave_dev);
2219                         netif_addr_unlock_bh(bond_dev);
2220                 }
2221
2222                 netdev_rx_handler_unregister(slave_dev);
2223                 synchronize_net();
2224                 netdev_set_bond_master(slave_dev, NULL);
2225
2226                 slave_disable_netpoll(slave);
2227
2228                 /* close slave before restoring its mac address */
2229                 dev_close(slave_dev);
2230
2231                 if (!bond->params.fail_over_mac) {
2232                         /* restore original ("permanent") mac address*/
2233                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2234                         addr.sa_family = slave_dev->type;
2235                         dev_set_mac_address(slave_dev, &addr);
2236                 }
2237
2238                 kfree(slave);
2239
2240                 /* re-acquire the lock before getting the next slave */
2241                 write_lock_bh(&bond->lock);
2242         }
2243
2244         /* zero the mac address of the master so it will be
2245          * set by the application to the mac address of the
2246          * first slave
2247          */
2248         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2249
2250         if (!bond->vlgrp) {
2251                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2252         } else {
2253                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2254                            bond_dev->name, bond_dev->name);
2255                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2256                            bond_dev->name);
2257         }
2258
2259         pr_info("%s: released all slaves\n", bond_dev->name);
2260
2261 out:
2262         write_unlock_bh(&bond->lock);
2263         return 0;
2264 }
2265
2266 /*
2267  * This function changes the active slave to slave <slave_dev>.
2268  * It returns -EINVAL in the following cases.
2269  *  - <slave_dev> is not found in the list.
2270  *  - There is not active slave now.
2271  *  - <slave_dev> is already active.
2272  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2273  *  - <slave_dev> is not running.
2274  * In these cases, this function does nothing.
2275  * In the other cases, current_slave pointer is changed and 0 is returned.
2276  */
2277 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2278 {
2279         struct bonding *bond = netdev_priv(bond_dev);
2280         struct slave *old_active = NULL;
2281         struct slave *new_active = NULL;
2282         int res = 0;
2283
2284         if (!USES_PRIMARY(bond->params.mode))
2285                 return -EINVAL;
2286
2287         /* Verify that master_dev is indeed the master of slave_dev */
2288         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2289                 return -EINVAL;
2290
2291         read_lock(&bond->lock);
2292
2293         read_lock(&bond->curr_slave_lock);
2294         old_active = bond->curr_active_slave;
2295         read_unlock(&bond->curr_slave_lock);
2296
2297         new_active = bond_get_slave_by_dev(bond, slave_dev);
2298
2299         /*
2300          * Changing to the current active: do nothing; return success.
2301          */
2302         if (new_active && (new_active == old_active)) {
2303                 read_unlock(&bond->lock);
2304                 return 0;
2305         }
2306
2307         if ((new_active) &&
2308             (old_active) &&
2309             (new_active->link == BOND_LINK_UP) &&
2310             IS_UP(new_active->dev)) {
2311                 block_netpoll_tx();
2312                 write_lock_bh(&bond->curr_slave_lock);
2313                 bond_change_active_slave(bond, new_active);
2314                 write_unlock_bh(&bond->curr_slave_lock);
2315                 unblock_netpoll_tx();
2316         } else
2317                 res = -EINVAL;
2318
2319         read_unlock(&bond->lock);
2320
2321         return res;
2322 }
2323
2324 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2325 {
2326         struct bonding *bond = netdev_priv(bond_dev);
2327
2328         info->bond_mode = bond->params.mode;
2329         info->miimon = bond->params.miimon;
2330
2331         read_lock(&bond->lock);
2332         info->num_slaves = bond->slave_cnt;
2333         read_unlock(&bond->lock);
2334
2335         return 0;
2336 }
2337
2338 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2339 {
2340         struct bonding *bond = netdev_priv(bond_dev);
2341         struct slave *slave;
2342         int i, res = -ENODEV;
2343
2344         read_lock(&bond->lock);
2345
2346         bond_for_each_slave(bond, slave, i) {
2347                 if (i == (int)info->slave_id) {
2348                         res = 0;
2349                         strcpy(info->slave_name, slave->dev->name);
2350                         info->link = slave->link;
2351                         info->state = bond_slave_state(slave);
2352                         info->link_failure_count = slave->link_failure_count;
2353                         break;
2354                 }
2355         }
2356
2357         read_unlock(&bond->lock);
2358
2359         return res;
2360 }
2361
2362 /*-------------------------------- Monitoring -------------------------------*/
2363
2364
2365 static int bond_miimon_inspect(struct bonding *bond)
2366 {
2367         struct slave *slave;
2368         int i, link_state, commit = 0;
2369         bool ignore_updelay;
2370
2371         ignore_updelay = !bond->curr_active_slave ? true : false;
2372
2373         bond_for_each_slave(bond, slave, i) {
2374                 slave->new_link = BOND_LINK_NOCHANGE;
2375
2376                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2377
2378                 switch (slave->link) {
2379                 case BOND_LINK_UP:
2380                         if (link_state)
2381                                 continue;
2382
2383                         slave->link = BOND_LINK_FAIL;
2384                         slave->delay = bond->params.downdelay;
2385                         if (slave->delay) {
2386                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2387                                         bond->dev->name,
2388                                         (bond->params.mode ==
2389                                          BOND_MODE_ACTIVEBACKUP) ?
2390                                         (bond_is_active_slave(slave) ?
2391                                          "active " : "backup ") : "",
2392                                         slave->dev->name,
2393                                         bond->params.downdelay * bond->params.miimon);
2394                         }
2395                         /*FALLTHRU*/
2396                 case BOND_LINK_FAIL:
2397                         if (link_state) {
2398                                 /*
2399                                  * recovered before downdelay expired
2400                                  */
2401                                 slave->link = BOND_LINK_UP;
2402                                 slave->jiffies = jiffies;
2403                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2404                                         bond->dev->name,
2405                                         (bond->params.downdelay - slave->delay) *
2406                                         bond->params.miimon,
2407                                         slave->dev->name);
2408                                 continue;
2409                         }
2410
2411                         if (slave->delay <= 0) {
2412                                 slave->new_link = BOND_LINK_DOWN;
2413                                 commit++;
2414                                 continue;
2415                         }
2416
2417                         slave->delay--;
2418                         break;
2419
2420                 case BOND_LINK_DOWN:
2421                         if (!link_state)
2422                                 continue;
2423
2424                         slave->link = BOND_LINK_BACK;
2425                         slave->delay = bond->params.updelay;
2426
2427                         if (slave->delay) {
2428                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2429                                         bond->dev->name, slave->dev->name,
2430                                         ignore_updelay ? 0 :
2431                                         bond->params.updelay *
2432                                         bond->params.miimon);
2433                         }
2434                         /*FALLTHRU*/
2435                 case BOND_LINK_BACK:
2436                         if (!link_state) {
2437                                 slave->link = BOND_LINK_DOWN;
2438                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2439                                         bond->dev->name,
2440                                         (bond->params.updelay - slave->delay) *
2441                                         bond->params.miimon,
2442                                         slave->dev->name);
2443
2444                                 continue;
2445                         }
2446
2447                         if (ignore_updelay)
2448                                 slave->delay = 0;
2449
2450                         if (slave->delay <= 0) {
2451                                 slave->new_link = BOND_LINK_UP;
2452                                 commit++;
2453                                 ignore_updelay = false;
2454                                 continue;
2455                         }
2456
2457                         slave->delay--;
2458                         break;
2459                 }
2460         }
2461
2462         return commit;
2463 }
2464
2465 static void bond_miimon_commit(struct bonding *bond)
2466 {
2467         struct slave *slave;
2468         int i;
2469
2470         bond_for_each_slave(bond, slave, i) {
2471                 switch (slave->new_link) {
2472                 case BOND_LINK_NOCHANGE:
2473                         continue;
2474
2475                 case BOND_LINK_UP:
2476                         slave->link = BOND_LINK_UP;
2477                         slave->jiffies = jiffies;
2478
2479                         if (bond->params.mode == BOND_MODE_8023AD) {
2480                                 /* prevent it from being the active one */
2481                                 bond_set_backup_slave(slave);
2482                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2483                                 /* make it immediately active */
2484                                 bond_set_active_slave(slave);
2485                         } else if (slave != bond->primary_slave) {
2486                                 /* prevent it from being the active one */
2487                                 bond_set_backup_slave(slave);
2488                         }
2489
2490                         bond_update_speed_duplex(slave);
2491
2492                         pr_info("%s: link status definitely up for interface %s, %d Mbps %s duplex.\n",
2493                                 bond->dev->name, slave->dev->name,
2494                                 slave->speed, slave->duplex ? "full" : "half");
2495
2496                         /* notify ad that the link status has changed */
2497                         if (bond->params.mode == BOND_MODE_8023AD)
2498                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2499
2500                         if (bond_is_lb(bond))
2501                                 bond_alb_handle_link_change(bond, slave,
2502                                                             BOND_LINK_UP);
2503
2504                         if (!bond->curr_active_slave ||
2505                             (slave == bond->primary_slave))
2506                                 goto do_failover;
2507
2508                         continue;
2509
2510                 case BOND_LINK_DOWN:
2511                         if (slave->link_failure_count < UINT_MAX)
2512                                 slave->link_failure_count++;
2513
2514                         slave->link = BOND_LINK_DOWN;
2515
2516                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2517                             bond->params.mode == BOND_MODE_8023AD)
2518                                 bond_set_slave_inactive_flags(slave);
2519
2520                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2521                                 bond->dev->name, slave->dev->name);
2522
2523                         if (bond->params.mode == BOND_MODE_8023AD)
2524                                 bond_3ad_handle_link_change(slave,
2525                                                             BOND_LINK_DOWN);
2526
2527                         if (bond_is_lb(bond))
2528                                 bond_alb_handle_link_change(bond, slave,
2529                                                             BOND_LINK_DOWN);
2530
2531                         if (slave == bond->curr_active_slave)
2532                                 goto do_failover;
2533
2534                         continue;
2535
2536                 default:
2537                         pr_err("%s: invalid new link %d on slave %s\n",
2538                                bond->dev->name, slave->new_link,
2539                                slave->dev->name);
2540                         slave->new_link = BOND_LINK_NOCHANGE;
2541
2542                         continue;
2543                 }
2544
2545 do_failover:
2546                 ASSERT_RTNL();
2547                 block_netpoll_tx();
2548                 write_lock_bh(&bond->curr_slave_lock);
2549                 bond_select_active_slave(bond);
2550                 write_unlock_bh(&bond->curr_slave_lock);
2551                 unblock_netpoll_tx();
2552         }
2553
2554         bond_set_carrier(bond);
2555 }
2556
2557 /*
2558  * bond_mii_monitor
2559  *
2560  * Really a wrapper that splits the mii monitor into two phases: an
2561  * inspection, then (if inspection indicates something needs to be done)
2562  * an acquisition of appropriate locks followed by a commit phase to
2563  * implement whatever link state changes are indicated.
2564  */
2565 void bond_mii_monitor(struct work_struct *work)
2566 {
2567         struct bonding *bond = container_of(work, struct bonding,
2568                                             mii_work.work);
2569
2570         read_lock(&bond->lock);
2571         if (bond->kill_timers)
2572                 goto out;
2573
2574         if (bond->slave_cnt == 0)
2575                 goto re_arm;
2576
2577         if (bond->send_grat_arp) {
2578                 read_lock(&bond->curr_slave_lock);
2579                 bond_send_gratuitous_arp(bond);
2580                 read_unlock(&bond->curr_slave_lock);
2581         }
2582
2583         if (bond->send_unsol_na) {
2584                 read_lock(&bond->curr_slave_lock);
2585                 bond_send_unsolicited_na(bond);
2586                 read_unlock(&bond->curr_slave_lock);
2587         }
2588
2589         if (bond_miimon_inspect(bond)) {
2590                 read_unlock(&bond->lock);
2591                 rtnl_lock();
2592                 read_lock(&bond->lock);
2593
2594                 bond_miimon_commit(bond);
2595
2596                 read_unlock(&bond->lock);
2597                 rtnl_unlock();  /* might sleep, hold no other locks */
2598                 read_lock(&bond->lock);
2599         }
2600
2601 re_arm:
2602         if (bond->params.miimon)
2603                 queue_delayed_work(bond->wq, &bond->mii_work,
2604                                    msecs_to_jiffies(bond->params.miimon));
2605 out:
2606         read_unlock(&bond->lock);
2607 }
2608
2609 static __be32 bond_glean_dev_ip(struct net_device *dev)
2610 {
2611         struct in_device *idev;
2612         struct in_ifaddr *ifa;
2613         __be32 addr = 0;
2614
2615         if (!dev)
2616                 return 0;
2617
2618         rcu_read_lock();
2619         idev = __in_dev_get_rcu(dev);
2620         if (!idev)
2621                 goto out;
2622
2623         ifa = idev->ifa_list;
2624         if (!ifa)
2625                 goto out;
2626
2627         addr = ifa->ifa_local;
2628 out:
2629         rcu_read_unlock();
2630         return addr;
2631 }
2632
2633 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2634 {
2635         struct vlan_entry *vlan;
2636
2637         if (ip == bond->master_ip)
2638                 return 1;
2639
2640         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2641                 if (ip == vlan->vlan_ip)
2642                         return 1;
2643         }
2644
2645         return 0;
2646 }
2647
2648 /*
2649  * We go to the (large) trouble of VLAN tagging ARP frames because
2650  * switches in VLAN mode (especially if ports are configured as
2651  * "native" to a VLAN) might not pass non-tagged frames.
2652  */
2653 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2654 {
2655         struct sk_buff *skb;
2656
2657         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2658                  slave_dev->name, dest_ip, src_ip, vlan_id);
2659
2660         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2661                          NULL, slave_dev->dev_addr, NULL);
2662
2663         if (!skb) {
2664                 pr_err("ARP packet allocation failed\n");
2665                 return;
2666         }
2667         if (vlan_id) {
2668                 skb = vlan_put_tag(skb, vlan_id);
2669                 if (!skb) {
2670                         pr_err("failed to insert VLAN tag\n");
2671                         return;
2672                 }
2673         }
2674         arp_xmit(skb);
2675 }
2676
2677
2678 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2679 {
2680         int i, vlan_id;
2681         __be32 *targets = bond->params.arp_targets;
2682         struct vlan_entry *vlan;
2683         struct net_device *vlan_dev;
2684         struct rtable *rt;
2685
2686         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2687                 if (!targets[i])
2688                         break;
2689                 pr_debug("basa: target %x\n", targets[i]);
2690                 if (!bond->vlgrp) {
2691                         pr_debug("basa: empty vlan: arp_send\n");
2692                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2693                                       bond->master_ip, 0);
2694                         continue;
2695                 }
2696
2697                 /*
2698                  * If VLANs are configured, we do a route lookup to
2699                  * determine which VLAN interface would be used, so we
2700                  * can tag the ARP with the proper VLAN tag.
2701                  */
2702                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2703                                      RTO_ONLINK, 0);
2704                 if (IS_ERR(rt)) {
2705                         if (net_ratelimit()) {
2706                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2707                                            bond->dev->name, &targets[i]);
2708                         }
2709                         continue;
2710                 }
2711
2712                 /*
2713                  * This target is not on a VLAN
2714                  */
2715                 if (rt->dst.dev == bond->dev) {
2716                         ip_rt_put(rt);
2717                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2718                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2719                                       bond->master_ip, 0);
2720                         continue;
2721                 }
2722
2723                 vlan_id = 0;
2724                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2725                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2726                         if (vlan_dev == rt->dst.dev) {
2727                                 vlan_id = vlan->vlan_id;
2728                                 pr_debug("basa: vlan match on %s %d\n",
2729                                        vlan_dev->name, vlan_id);
2730                                 break;
2731                         }
2732                 }
2733
2734                 if (vlan_id) {
2735                         ip_rt_put(rt);
2736                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2737                                       vlan->vlan_ip, vlan_id);
2738                         continue;
2739                 }
2740
2741                 if (net_ratelimit()) {
2742                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2743                                    bond->dev->name, &targets[i],
2744                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2745                 }
2746                 ip_rt_put(rt);
2747         }
2748 }
2749
2750 /*
2751  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2752  * for each VLAN above us.
2753  *
2754  * Caller must hold curr_slave_lock for read or better
2755  */
2756 static void bond_send_gratuitous_arp(struct bonding *bond)
2757 {
2758         struct slave *slave = bond->curr_active_slave;
2759         struct vlan_entry *vlan;
2760         struct net_device *vlan_dev;
2761
2762         pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2763                  bond->dev->name, slave ? slave->dev->name : "NULL");
2764
2765         if (!slave || !bond->send_grat_arp ||
2766             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2767                 return;
2768
2769         bond->send_grat_arp--;
2770
2771         if (bond->master_ip) {
2772                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2773                                 bond->master_ip, 0);
2774         }
2775
2776         if (!bond->vlgrp)
2777                 return;
2778
2779         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2780                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2781                 if (vlan->vlan_ip) {
2782                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2783                                       vlan->vlan_ip, vlan->vlan_id);
2784                 }
2785         }
2786 }
2787
2788 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2789 {
2790         int i;
2791         __be32 *targets = bond->params.arp_targets;
2792
2793         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2794                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2795                          &sip, &tip, i, &targets[i],
2796                          bond_has_this_ip(bond, tip));
2797                 if (sip == targets[i]) {
2798                         if (bond_has_this_ip(bond, tip))
2799                                 slave->last_arp_rx = jiffies;
2800                         return;
2801                 }
2802         }
2803 }
2804
2805 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2806 {
2807         struct arphdr *arp;
2808         struct slave *slave;
2809         struct bonding *bond;
2810         unsigned char *arp_ptr;
2811         __be32 sip, tip;
2812
2813         if (dev->priv_flags & IFF_802_1Q_VLAN) {
2814                 /*
2815                  * When using VLANS and bonding, dev and oriv_dev may be
2816                  * incorrect if the physical interface supports VLAN
2817                  * acceleration.  With this change ARP validation now
2818                  * works for hosts only reachable on the VLAN interface.
2819                  */
2820                 dev = vlan_dev_real_dev(dev);
2821                 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2822         }
2823
2824         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2825                 goto out;
2826
2827         bond = netdev_priv(dev);
2828         read_lock(&bond->lock);
2829
2830         pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2831                  bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2832                  orig_dev ? orig_dev->name : "NULL");
2833
2834         slave = bond_get_slave_by_dev(bond, orig_dev);
2835         if (!slave || !slave_do_arp_validate(bond, slave))
2836                 goto out_unlock;
2837
2838         skb = skb_share_check(skb, GFP_ATOMIC);
2839         if (!skb)
2840                 goto out_unlock;
2841
2842         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2843                 goto out_unlock;
2844
2845         arp = arp_hdr(skb);
2846         if (arp->ar_hln != dev->addr_len ||
2847             skb->pkt_type == PACKET_OTHERHOST ||
2848             skb->pkt_type == PACKET_LOOPBACK ||
2849             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2850             arp->ar_pro != htons(ETH_P_IP) ||
2851             arp->ar_pln != 4)
2852                 goto out_unlock;
2853
2854         arp_ptr = (unsigned char *)(arp + 1);
2855         arp_ptr += dev->addr_len;
2856         memcpy(&sip, arp_ptr, 4);
2857         arp_ptr += 4 + dev->addr_len;
2858         memcpy(&tip, arp_ptr, 4);
2859
2860         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2861                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2862                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2863                  &sip, &tip);
2864
2865         /*
2866          * Backup slaves won't see the ARP reply, but do come through
2867          * here for each ARP probe (so we swap the sip/tip to validate
2868          * the probe).  In a "redundant switch, common router" type of
2869          * configuration, the ARP probe will (hopefully) travel from
2870          * the active, through one switch, the router, then the other
2871          * switch before reaching the backup.
2872          */
2873         if (bond_is_active_slave(slave))
2874                 bond_validate_arp(bond, slave, sip, tip);
2875         else
2876                 bond_validate_arp(bond, slave, tip, sip);
2877
2878 out_unlock:
2879         read_unlock(&bond->lock);
2880 out:
2881         dev_kfree_skb(skb);
2882         return NET_RX_SUCCESS;
2883 }
2884
2885 /*
2886  * this function is called regularly to monitor each slave's link
2887  * ensuring that traffic is being sent and received when arp monitoring
2888  * is used in load-balancing mode. if the adapter has been dormant, then an
2889  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2890  * arp monitoring in active backup mode.
2891  */
2892 void bond_loadbalance_arp_mon(struct work_struct *work)
2893 {
2894         struct bonding *bond = container_of(work, struct bonding,
2895                                             arp_work.work);
2896         struct slave *slave, *oldcurrent;
2897         int do_failover = 0;
2898         int delta_in_ticks;
2899         int i;
2900
2901         read_lock(&bond->lock);
2902
2903         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2904
2905         if (bond->kill_timers)
2906                 goto out;
2907
2908         if (bond->slave_cnt == 0)
2909                 goto re_arm;
2910
2911         read_lock(&bond->curr_slave_lock);
2912         oldcurrent = bond->curr_active_slave;
2913         read_unlock(&bond->curr_slave_lock);
2914
2915         /* see if any of the previous devices are up now (i.e. they have
2916          * xmt and rcv traffic). the curr_active_slave does not come into
2917          * the picture unless it is null. also, slave->jiffies is not needed
2918          * here because we send an arp on each slave and give a slave as
2919          * long as it needs to get the tx/rx within the delta.
2920          * TODO: what about up/down delay in arp mode? it wasn't here before
2921          *       so it can wait
2922          */
2923         bond_for_each_slave(bond, slave, i) {
2924                 unsigned long trans_start = dev_trans_start(slave->dev);
2925
2926                 if (slave->link != BOND_LINK_UP) {
2927                         if (time_in_range(jiffies,
2928                                 trans_start - delta_in_ticks,
2929                                 trans_start + delta_in_ticks) &&
2930                             time_in_range(jiffies,
2931                                 slave->dev->last_rx - delta_in_ticks,
2932                                 slave->dev->last_rx + delta_in_ticks)) {
2933
2934                                 slave->link  = BOND_LINK_UP;
2935                                 bond_set_active_slave(slave);
2936
2937                                 /* primary_slave has no meaning in round-robin
2938                                  * mode. the window of a slave being up and
2939                                  * curr_active_slave being null after enslaving
2940                                  * is closed.
2941                                  */
2942                                 if (!oldcurrent) {
2943                                         pr_info("%s: link status definitely up for interface %s, ",
2944                                                 bond->dev->name,
2945                                                 slave->dev->name);
2946                                         do_failover = 1;
2947                                 } else {
2948                                         pr_info("%s: interface %s is now up\n",
2949                                                 bond->dev->name,
2950                                                 slave->dev->name);
2951                                 }
2952                         }
2953                 } else {
2954                         /* slave->link == BOND_LINK_UP */
2955
2956                         /* not all switches will respond to an arp request
2957                          * when the source ip is 0, so don't take the link down
2958                          * if we don't know our ip yet
2959                          */
2960                         if (!time_in_range(jiffies,
2961                                 trans_start - delta_in_ticks,
2962                                 trans_start + 2 * delta_in_ticks) ||
2963                             !time_in_range(jiffies,
2964                                 slave->dev->last_rx - delta_in_ticks,
2965                                 slave->dev->last_rx + 2 * delta_in_ticks)) {
2966
2967                                 slave->link  = BOND_LINK_DOWN;
2968                                 bond_set_backup_slave(slave);
2969
2970                                 if (slave->link_failure_count < UINT_MAX)
2971                                         slave->link_failure_count++;
2972
2973                                 pr_info("%s: interface %s is now down.\n",
2974                                         bond->dev->name,
2975                                         slave->dev->name);
2976
2977                                 if (slave == oldcurrent)
2978                                         do_failover = 1;
2979                         }
2980                 }
2981
2982                 /* note: if switch is in round-robin mode, all links
2983                  * must tx arp to ensure all links rx an arp - otherwise
2984                  * links may oscillate or not come up at all; if switch is
2985                  * in something like xor mode, there is nothing we can
2986                  * do - all replies will be rx'ed on same link causing slaves
2987                  * to be unstable during low/no traffic periods
2988                  */
2989                 if (IS_UP(slave->dev))
2990                         bond_arp_send_all(bond, slave);
2991         }
2992
2993         if (do_failover) {
2994                 block_netpoll_tx();
2995                 write_lock_bh(&bond->curr_slave_lock);
2996
2997                 bond_select_active_slave(bond);
2998
2999                 write_unlock_bh(&bond->curr_slave_lock);
3000                 unblock_netpoll_tx();
3001         }
3002
3003 re_arm:
3004         if (bond->params.arp_interval)
3005                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3006 out:
3007         read_unlock(&bond->lock);
3008 }
3009
3010 /*
3011  * Called to inspect slaves for active-backup mode ARP monitor link state
3012  * changes.  Sets new_link in slaves to specify what action should take
3013  * place for the slave.  Returns 0 if no changes are found, >0 if changes
3014  * to link states must be committed.
3015  *
3016  * Called with bond->lock held for read.
3017  */
3018 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
3019 {
3020         struct slave *slave;
3021         int i, commit = 0;
3022         unsigned long trans_start;
3023
3024         bond_for_each_slave(bond, slave, i) {
3025                 slave->new_link = BOND_LINK_NOCHANGE;
3026
3027                 if (slave->link != BOND_LINK_UP) {
3028                         if (time_in_range(jiffies,
3029                                 slave_last_rx(bond, slave) - delta_in_ticks,
3030                                 slave_last_rx(bond, slave) + delta_in_ticks)) {
3031
3032                                 slave->new_link = BOND_LINK_UP;
3033                                 commit++;
3034                         }
3035
3036                         continue;
3037                 }
3038
3039                 /*
3040                  * Give slaves 2*delta after being enslaved or made
3041                  * active.  This avoids bouncing, as the last receive
3042                  * times need a full ARP monitor cycle to be updated.
3043                  */
3044                 if (time_in_range(jiffies,
3045                                   slave->jiffies - delta_in_ticks,
3046                                   slave->jiffies + 2 * delta_in_ticks))
3047                         continue;
3048
3049                 /*
3050                  * Backup slave is down if:
3051                  * - No current_arp_slave AND
3052                  * - more than 3*delta since last receive AND
3053                  * - the bond has an IP address
3054                  *
3055                  * Note: a non-null current_arp_slave indicates
3056                  * the curr_active_slave went down and we are
3057                  * searching for a new one; under this condition
3058                  * we only take the curr_active_slave down - this
3059                  * gives each slave a chance to tx/rx traffic
3060                  * before being taken out
3061                  */
3062                 if (!bond_is_active_slave(slave) &&
3063                     !bond->current_arp_slave &&
3064                     !time_in_range(jiffies,
3065                         slave_last_rx(bond, slave) - delta_in_ticks,
3066                         slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
3067
3068                         slave->new_link = BOND_LINK_DOWN;
3069                         commit++;
3070                 }
3071
3072                 /*
3073                  * Active slave is down if:
3074                  * - more than 2*delta since transmitting OR
3075                  * - (more than 2*delta since receive AND
3076                  *    the bond has an IP address)
3077                  */
3078                 trans_start = dev_trans_start(slave->dev);
3079                 if (bond_is_active_slave(slave) &&
3080                     (!time_in_range(jiffies,
3081                         trans_start - delta_in_ticks,
3082                         trans_start + 2 * delta_in_ticks) ||
3083                      !time_in_range(jiffies,
3084                         slave_last_rx(bond, slave) - delta_in_ticks,
3085                         slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
3086
3087                         slave->new_link = BOND_LINK_DOWN;
3088                         commit++;
3089                 }
3090         }
3091
3092         return commit;
3093 }
3094
3095 /*
3096  * Called to commit link state changes noted by inspection step of
3097  * active-backup mode ARP monitor.
3098  *
3099  * Called with RTNL and bond->lock for read.
3100  */
3101 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3102 {
3103         struct slave *slave;
3104         int i;
3105         unsigned long trans_start;
3106
3107         bond_for_each_slave(bond, slave, i) {
3108                 switch (slave->new_link) {
3109                 case BOND_LINK_NOCHANGE:
3110                         continue;
3111
3112                 case BOND_LINK_UP:
3113                         trans_start = dev_trans_start(slave->dev);
3114                         if ((!bond->curr_active_slave &&
3115                              time_in_range(jiffies,
3116                                            trans_start - delta_in_ticks,
3117                                            trans_start + delta_in_ticks)) ||
3118                             bond->curr_active_slave != slave) {
3119                                 slave->link = BOND_LINK_UP;
3120                                 bond->current_arp_slave = NULL;
3121
3122                                 pr_info("%s: link status definitely up for interface %s.\n",
3123                                         bond->dev->name, slave->dev->name);
3124
3125                                 if (!bond->curr_active_slave ||
3126                                     (slave == bond->primary_slave))
3127                                         goto do_failover;
3128
3129                         }
3130
3131                         continue;
3132
3133                 case BOND_LINK_DOWN:
3134                         if (slave->link_failure_count < UINT_MAX)
3135                                 slave->link_failure_count++;
3136
3137                         slave->link = BOND_LINK_DOWN;
3138                         bond_set_slave_inactive_flags(slave);
3139
3140                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3141                                 bond->dev->name, slave->dev->name);
3142
3143                         if (slave == bond->curr_active_slave) {
3144                                 bond->current_arp_slave = NULL;
3145                                 goto do_failover;
3146                         }
3147
3148                         continue;
3149
3150                 default:
3151                         pr_err("%s: impossible: new_link %d on slave %s\n",
3152                                bond->dev->name, slave->new_link,
3153                                slave->dev->name);
3154                         continue;
3155                 }
3156
3157 do_failover:
3158                 ASSERT_RTNL();
3159                 block_netpoll_tx();
3160                 write_lock_bh(&bond->curr_slave_lock);
3161                 bond_select_active_slave(bond);
3162                 write_unlock_bh(&bond->curr_slave_lock);
3163                 unblock_netpoll_tx();
3164         }
3165
3166         bond_set_carrier(bond);
3167 }
3168
3169 /*
3170  * Send ARP probes for active-backup mode ARP monitor.
3171  *
3172  * Called with bond->lock held for read.
3173  */
3174 static void bond_ab_arp_probe(struct bonding *bond)
3175 {
3176         struct slave *slave;
3177         int i;
3178
3179         read_lock(&bond->curr_slave_lock);
3180
3181         if (bond->current_arp_slave && bond->curr_active_slave)
3182                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3183                         bond->current_arp_slave->dev->name,
3184                         bond->curr_active_slave->dev->name);
3185
3186         if (bond->curr_active_slave) {
3187                 bond_arp_send_all(bond, bond->curr_active_slave);
3188                 read_unlock(&bond->curr_slave_lock);
3189                 return;
3190         }
3191
3192         read_unlock(&bond->curr_slave_lock);
3193
3194         /* if we don't have a curr_active_slave, search for the next available
3195          * backup slave from the current_arp_slave and make it the candidate
3196          * for becoming the curr_active_slave
3197          */
3198
3199         if (!bond->current_arp_slave) {
3200                 bond->current_arp_slave = bond->first_slave;
3201                 if (!bond->current_arp_slave)
3202                         return;
3203         }
3204
3205         bond_set_slave_inactive_flags(bond->current_arp_slave);
3206
3207         /* search for next candidate */
3208         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3209                 if (IS_UP(slave->dev)) {
3210                         slave->link = BOND_LINK_BACK;
3211                         bond_set_slave_active_flags(slave);
3212                         bond_arp_send_all(bond, slave);
3213                         slave->jiffies = jiffies;
3214                         bond->current_arp_slave = slave;
3215                         break;
3216                 }
3217
3218                 /* if the link state is up at this point, we
3219                  * mark it down - this can happen if we have
3220                  * simultaneous link failures and
3221                  * reselect_active_interface doesn't make this
3222                  * one the current slave so it is still marked
3223                  * up when it is actually down
3224                  */
3225                 if (slave->link == BOND_LINK_UP) {
3226                         slave->link = BOND_LINK_DOWN;
3227                         if (slave->link_failure_count < UINT_MAX)
3228                                 slave->link_failure_count++;
3229
3230                         bond_set_slave_inactive_flags(slave);
3231
3232                         pr_info("%s: backup interface %s is now down.\n",
3233                                 bond->dev->name, slave->dev->name);
3234                 }
3235         }
3236 }
3237
3238 void bond_activebackup_arp_mon(struct work_struct *work)
3239 {
3240         struct bonding *bond = container_of(work, struct bonding,
3241                                             arp_work.work);
3242         int delta_in_ticks;
3243
3244         read_lock(&bond->lock);
3245
3246         if (bond->kill_timers)
3247                 goto out;
3248
3249         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3250
3251         if (bond->slave_cnt == 0)
3252                 goto re_arm;
3253
3254         if (bond->send_grat_arp) {
3255                 read_lock(&bond->curr_slave_lock);
3256                 bond_send_gratuitous_arp(bond);
3257                 read_unlock(&bond->curr_slave_lock);
3258         }
3259
3260         if (bond->send_unsol_na) {
3261                 read_lock(&bond->curr_slave_lock);
3262                 bond_send_unsolicited_na(bond);
3263                 read_unlock(&bond->curr_slave_lock);
3264         }
3265
3266         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3267                 read_unlock(&bond->lock);
3268                 rtnl_lock();
3269                 read_lock(&bond->lock);
3270
3271                 bond_ab_arp_commit(bond, delta_in_ticks);
3272
3273                 read_unlock(&bond->lock);
3274                 rtnl_unlock();
3275                 read_lock(&bond->lock);
3276         }
3277
3278         bond_ab_arp_probe(bond);
3279
3280 re_arm:
3281         if (bond->params.arp_interval)
3282                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3283 out:
3284         read_unlock(&bond->lock);
3285 }
3286
3287 /*-------------------------- netdev event handling --------------------------*/
3288
3289 /*
3290  * Change device name
3291  */
3292 static int bond_event_changename(struct bonding *bond)
3293 {
3294         bond_remove_proc_entry(bond);
3295         bond_create_proc_entry(bond);
3296
3297         bond_debug_reregister(bond);
3298
3299         return NOTIFY_DONE;
3300 }
3301
3302 static int bond_master_netdev_event(unsigned long event,
3303                                     struct net_device *bond_dev)
3304 {
3305         struct bonding *event_bond = netdev_priv(bond_dev);
3306
3307         switch (event) {
3308         case NETDEV_CHANGENAME:
3309                 return bond_event_changename(event_bond);
3310         default:
3311                 break;
3312         }
3313
3314         return NOTIFY_DONE;
3315 }
3316
3317 static int bond_slave_netdev_event(unsigned long event,
3318                                    struct net_device *slave_dev)
3319 {
3320         struct net_device *bond_dev = slave_dev->master;
3321         struct bonding *bond = netdev_priv(bond_dev);
3322
3323         switch (event) {
3324         case NETDEV_UNREGISTER:
3325                 if (bond_dev) {
3326                         if (bond->setup_by_slave)
3327                                 bond_release_and_destroy(bond_dev, slave_dev);
3328                         else
3329                                 bond_release(bond_dev, slave_dev);
3330                 }
3331                 break;
3332         case NETDEV_CHANGE:
3333                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3334                         struct slave *slave;
3335
3336                         slave = bond_get_slave_by_dev(bond, slave_dev);
3337                         if (slave) {
3338                                 u16 old_speed = slave->speed;
3339                                 u16 old_duplex = slave->duplex;
3340
3341                                 bond_update_speed_duplex(slave);
3342
3343                                 if (bond_is_lb(bond))
3344                                         break;
3345
3346                                 if (old_speed != slave->speed)
3347                                         bond_3ad_adapter_speed_changed(slave);
3348                                 if (old_duplex != slave->duplex)
3349                                         bond_3ad_adapter_duplex_changed(slave);
3350                         }
3351                 }
3352
3353                 break;
3354         case NETDEV_DOWN:
3355                 /*
3356                  * ... Or is it this?
3357                  */
3358                 break;
3359         case NETDEV_CHANGEMTU:
3360                 /*
3361                  * TODO: Should slaves be allowed to
3362                  * independently alter their MTU?  For
3363                  * an active-backup bond, slaves need
3364                  * not be the same type of device, so
3365                  * MTUs may vary.  For other modes,
3366                  * slaves arguably should have the
3367                  * same MTUs. To do this, we'd need to
3368                  * take over the slave's change_mtu
3369                  * function for the duration of their
3370                  * servitude.
3371                  */
3372                 break;
3373         case NETDEV_CHANGENAME:
3374                 /*
3375                  * TODO: handle changing the primary's name
3376                  */
3377                 break;
3378         case NETDEV_FEAT_CHANGE:
3379                 bond_compute_features(bond);
3380                 break;
3381         default:
3382                 break;
3383         }
3384
3385         return NOTIFY_DONE;
3386 }
3387
3388 /*
3389  * bond_netdev_event: handle netdev notifier chain events.
3390  *
3391  * This function receives events for the netdev chain.  The caller (an
3392  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3393  * locks for us to safely manipulate the slave devices (RTNL lock,
3394  * dev_probe_lock).
3395  */
3396 static int bond_netdev_event(struct notifier_block *this,
3397                              unsigned long event, void *ptr)
3398 {
3399         struct net_device *event_dev = (struct net_device *)ptr;
3400
3401         pr_debug("event_dev: %s, event: %lx\n",
3402                  event_dev ? event_dev->name : "None",
3403                  event);
3404
3405         if (!(event_dev->priv_flags & IFF_BONDING))
3406                 return NOTIFY_DONE;
3407
3408         if (event_dev->flags & IFF_MASTER) {
3409                 pr_debug("IFF_MASTER\n");
3410                 return bond_master_netdev_event(event, event_dev);
3411         }
3412
3413         if (event_dev->flags & IFF_SLAVE) {
3414                 pr_debug("IFF_SLAVE\n");
3415                 return bond_slave_netdev_event(event, event_dev);
3416         }
3417
3418         return NOTIFY_DONE;
3419 }
3420
3421 /*
3422  * bond_inetaddr_event: handle inetaddr notifier chain events.
3423  *
3424  * We keep track of device IPs primarily to use as source addresses in
3425  * ARP monitor probes (rather than spewing out broadcasts all the time).
3426  *
3427  * We track one IP for the main device (if it has one), plus one per VLAN.
3428  */
3429 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3430 {
3431         struct in_ifaddr *ifa = ptr;
3432         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3433         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3434         struct bonding *bond;
3435         struct vlan_entry *vlan;
3436
3437         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3438                 if (bond->dev == event_dev) {
3439                         switch (event) {
3440                         case NETDEV_UP:
3441                                 bond->master_ip = ifa->ifa_local;
3442                                 return NOTIFY_OK;
3443                         case NETDEV_DOWN:
3444                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3445                                 return NOTIFY_OK;
3446                         default:
3447                                 return NOTIFY_DONE;
3448                         }
3449                 }
3450
3451                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3452                         if (!bond->vlgrp)
3453                                 continue;
3454                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3455                         if (vlan_dev == event_dev) {
3456                                 switch (event) {
3457                                 case NETDEV_UP:
3458                                         vlan->vlan_ip = ifa->ifa_local;
3459                                         return NOTIFY_OK;
3460                                 case NETDEV_DOWN:
3461                                         vlan->vlan_ip =
3462                                                 bond_glean_dev_ip(vlan_dev);
3463                                         return NOTIFY_OK;
3464                                 default:
3465                                         return NOTIFY_DONE;
3466                                 }
3467                         }
3468                 }
3469         }
3470         return NOTIFY_DONE;
3471 }
3472
3473 static struct notifier_block bond_netdev_notifier = {
3474         .notifier_call = bond_netdev_event,
3475 };
3476
3477 static struct notifier_block bond_inetaddr_notifier = {
3478         .notifier_call = bond_inetaddr_event,
3479 };
3480
3481 /*-------------------------- Packet type handling ---------------------------*/
3482
3483 /* register to receive lacpdus on a bond */
3484 static void bond_register_lacpdu(struct bonding *bond)
3485 {
3486         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3487
3488         /* initialize packet type */
3489         pk_type->type = PKT_TYPE_LACPDU;
3490         pk_type->dev = bond->dev;
3491         pk_type->func = bond_3ad_lacpdu_recv;
3492
3493         dev_add_pack(pk_type);
3494 }
3495
3496 /* unregister to receive lacpdus on a bond */
3497 static void bond_unregister_lacpdu(struct bonding *bond)
3498 {
3499         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3500 }
3501
3502 void bond_register_arp(struct bonding *bond)
3503 {
3504         struct packet_type *pt = &bond->arp_mon_pt;
3505
3506         if (pt->type)
3507                 return;
3508
3509         pt->type = htons(ETH_P_ARP);
3510         pt->dev = bond->dev;
3511         pt->func = bond_arp_rcv;
3512         dev_add_pack(pt);
3513 }
3514
3515 void bond_unregister_arp(struct bonding *bond)
3516 {
3517         struct packet_type *pt = &bond->arp_mon_pt;
3518
3519         dev_remove_pack(pt);
3520         pt->type = 0;
3521 }
3522
3523 /*---------------------------- Hashing Policies -----------------------------*/
3524
3525 /*
3526  * Hash for the output device based upon layer 2 and layer 3 data. If
3527  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3528  */
3529 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3530 {
3531         struct ethhdr *data = (struct ethhdr *)skb->data;
3532         struct iphdr *iph = ip_hdr(skb);
3533
3534         if (skb->protocol == htons(ETH_P_IP)) {
3535                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3536                         (data->h_dest[5] ^ data->h_source[5])) % count;
3537         }
3538
3539         return (data->h_dest[5] ^ data->h_source[5]) % count;
3540 }
3541
3542 /*
3543  * Hash for the output device based upon layer 3 and layer 4 data. If
3544  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3545  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3546  */
3547 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3548 {
3549         struct ethhdr *data = (struct ethhdr *)skb->data;
3550         struct iphdr *iph = ip_hdr(skb);
3551         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3552         int layer4_xor = 0;
3553
3554         if (skb->protocol == htons(ETH_P_IP)) {
3555                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3556                     (iph->protocol == IPPROTO_TCP ||
3557                      iph->protocol == IPPROTO_UDP)) {
3558                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3559                 }
3560                 return (layer4_xor ^
3561                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3562
3563         }
3564
3565         return (data->h_dest[5] ^ data->h_source[5]) % count;
3566 }
3567
3568 /*
3569  * Hash for the output device based upon layer 2 data
3570  */
3571 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3572 {
3573         struct ethhdr *data = (struct ethhdr *)skb->data;
3574
3575         return (data->h_dest[5] ^ data->h_source[5]) % count;
3576 }
3577
3578 /*-------------------------- Device entry points ----------------------------*/
3579
3580 static int bond_open(struct net_device *bond_dev)
3581 {
3582         struct bonding *bond = netdev_priv(bond_dev);
3583
3584         bond->kill_timers = 0;
3585
3586         INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3587
3588         if (bond_is_lb(bond)) {
3589                 /* bond_alb_initialize must be called before the timer
3590                  * is started.
3591                  */
3592                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3593                         /* something went wrong - fail the open operation */
3594                         return -ENOMEM;
3595                 }
3596
3597                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3598                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3599         }
3600
3601         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3602                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3603                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3604         }
3605
3606         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3607                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3608                         INIT_DELAYED_WORK(&bond->arp_work,
3609                                           bond_activebackup_arp_mon);
3610                 else
3611                         INIT_DELAYED_WORK(&bond->arp_work,
3612                                           bond_loadbalance_arp_mon);
3613
3614                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3615                 if (bond->params.arp_validate)
3616                         bond_register_arp(bond);
3617         }
3618
3619         if (bond->params.mode == BOND_MODE_8023AD) {
3620                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3621                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3622                 /* register to receive LACPDUs */
3623                 bond_register_lacpdu(bond);
3624                 bond_3ad_initiate_agg_selection(bond, 1);
3625         }
3626
3627         return 0;
3628 }
3629
3630 static int bond_close(struct net_device *bond_dev)
3631 {
3632         struct bonding *bond = netdev_priv(bond_dev);
3633
3634         if (bond->params.mode == BOND_MODE_8023AD) {
3635                 /* Unregister the receive of LACPDUs */
3636                 bond_unregister_lacpdu(bond);
3637         }
3638
3639         if (bond->params.arp_validate)
3640                 bond_unregister_arp(bond);
3641
3642         write_lock_bh(&bond->lock);
3643
3644         bond->send_grat_arp = 0;
3645         bond->send_unsol_na = 0;
3646
3647         /* signal timers not to re-arm */
3648         bond->kill_timers = 1;
3649
3650         write_unlock_bh(&bond->lock);
3651
3652         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3653                 cancel_delayed_work(&bond->mii_work);
3654         }
3655
3656         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3657                 cancel_delayed_work(&bond->arp_work);
3658         }
3659
3660         switch (bond->params.mode) {
3661         case BOND_MODE_8023AD:
3662                 cancel_delayed_work(&bond->ad_work);
3663                 break;
3664         case BOND_MODE_TLB:
3665         case BOND_MODE_ALB:
3666                 cancel_delayed_work(&bond->alb_work);
3667                 break;
3668         default:
3669                 break;
3670         }
3671
3672         if (delayed_work_pending(&bond->mcast_work))
3673                 cancel_delayed_work(&bond->mcast_work);
3674
3675         if (bond_is_lb(bond)) {
3676                 /* Must be called only after all
3677                  * slaves have been released
3678                  */
3679                 bond_alb_deinitialize(bond);
3680         }
3681
3682         return 0;
3683 }
3684
3685 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3686                                                 struct rtnl_link_stats64 *stats)
3687 {
3688         struct bonding *bond = netdev_priv(bond_dev);
3689         struct rtnl_link_stats64 temp;
3690         struct slave *slave;
3691         int i;
3692
3693         memset(stats, 0, sizeof(*stats));
3694
3695         read_lock_bh(&bond->lock);
3696
3697         bond_for_each_slave(bond, slave, i) {
3698                 const struct rtnl_link_stats64 *sstats =
3699                         dev_get_stats(slave->dev, &temp);
3700
3701                 stats->rx_packets += sstats->rx_packets;
3702                 stats->rx_bytes += sstats->rx_bytes;
3703                 stats->rx_errors += sstats->rx_errors;
3704                 stats->rx_dropped += sstats->rx_dropped;
3705
3706                 stats->tx_packets += sstats->tx_packets;
3707                 stats->tx_bytes += sstats->tx_bytes;
3708                 stats->tx_errors += sstats->tx_errors;
3709                 stats->tx_dropped += sstats->tx_dropped;
3710
3711                 stats->multicast += sstats->multicast;
3712                 stats->collisions += sstats->collisions;
3713
3714                 stats->rx_length_errors += sstats->rx_length_errors;
3715                 stats->rx_over_errors += sstats->rx_over_errors;
3716                 stats->rx_crc_errors += sstats->rx_crc_errors;
3717                 stats->rx_frame_errors += sstats->rx_frame_errors;
3718                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3719                 stats->rx_missed_errors += sstats->rx_missed_errors;
3720
3721                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3722                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3723                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3724                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3725                 stats->tx_window_errors += sstats->tx_window_errors;
3726         }
3727
3728         read_unlock_bh(&bond->lock);
3729
3730         return stats;
3731 }
3732
3733 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3734 {
3735         struct net_device *slave_dev = NULL;
3736         struct ifbond k_binfo;
3737         struct ifbond __user *u_binfo = NULL;
3738         struct ifslave k_sinfo;
3739         struct ifslave __user *u_sinfo = NULL;
3740         struct mii_ioctl_data *mii = NULL;
3741         int res = 0;
3742
3743         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3744
3745         switch (cmd) {
3746         case SIOCGMIIPHY:
3747                 mii = if_mii(ifr);
3748                 if (!mii)
3749                         return -EINVAL;
3750
3751                 mii->phy_id = 0;
3752                 /* Fall Through */
3753         case SIOCGMIIREG:
3754                 /*
3755                  * We do this again just in case we were called by SIOCGMIIREG
3756                  * instead of SIOCGMIIPHY.
3757                  */
3758                 mii = if_mii(ifr);
3759                 if (!mii)
3760                         return -EINVAL;
3761
3762
3763                 if (mii->reg_num == 1) {
3764                         struct bonding *bond = netdev_priv(bond_dev);
3765                         mii->val_out = 0;
3766                         read_lock(&bond->lock);
3767                         read_lock(&bond->curr_slave_lock);
3768                         if (netif_carrier_ok(bond->dev))
3769                                 mii->val_out = BMSR_LSTATUS;
3770
3771                         read_unlock(&bond->curr_slave_lock);
3772                         read_unlock(&bond->lock);
3773                 }
3774
3775                 return 0;
3776         case BOND_INFO_QUERY_OLD:
3777         case SIOCBONDINFOQUERY:
3778                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3779
3780                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3781                         return -EFAULT;
3782
3783                 res = bond_info_query(bond_dev, &k_binfo);
3784                 if (res == 0 &&
3785                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3786                         return -EFAULT;
3787
3788                 return res;
3789         case BOND_SLAVE_INFO_QUERY_OLD:
3790         case SIOCBONDSLAVEINFOQUERY:
3791                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3792
3793                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3794                         return -EFAULT;
3795
3796                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3797                 if (res == 0 &&
3798                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3799                         return -EFAULT;
3800
3801                 return res;
3802         default:
3803                 /* Go on */
3804                 break;
3805         }
3806
3807         if (!capable(CAP_NET_ADMIN))
3808                 return -EPERM;
3809
3810         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3811
3812         pr_debug("slave_dev=%p:\n", slave_dev);
3813
3814         if (!slave_dev)
3815                 res = -ENODEV;
3816         else {
3817                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3818                 switch (cmd) {
3819                 case BOND_ENSLAVE_OLD:
3820                 case SIOCBONDENSLAVE:
3821                         res = bond_enslave(bond_dev, slave_dev);
3822                         break;
3823                 case BOND_RELEASE_OLD:
3824                 case SIOCBONDRELEASE:
3825                         res = bond_release(bond_dev, slave_dev);
3826                         break;
3827                 case BOND_SETHWADDR_OLD:
3828                 case SIOCBONDSETHWADDR:
3829                         res = bond_sethwaddr(bond_dev, slave_dev);
3830                         break;
3831                 case BOND_CHANGE_ACTIVE_OLD:
3832                 case SIOCBONDCHANGEACTIVE:
3833                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3834                         break;
3835                 default:
3836                         res = -EOPNOTSUPP;
3837                 }
3838
3839                 dev_put(slave_dev);
3840         }
3841
3842         return res;
3843 }
3844
3845 static bool bond_addr_in_mc_list(unsigned char *addr,
3846                                  struct netdev_hw_addr_list *list,
3847                                  int addrlen)
3848 {
3849         struct netdev_hw_addr *ha;
3850
3851         netdev_hw_addr_list_for_each(ha, list)
3852                 if (!memcmp(ha->addr, addr, addrlen))
3853                         return true;
3854
3855         return false;
3856 }
3857
3858 static void bond_set_multicast_list(struct net_device *bond_dev)
3859 {
3860         struct bonding *bond = netdev_priv(bond_dev);
3861         struct netdev_hw_addr *ha;
3862         bool found;
3863
3864         /*
3865          * Do promisc before checking multicast_mode
3866          */
3867         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
3868                 /*
3869                  * FIXME: Need to handle the error when one of the multi-slaves
3870                  * encounters error.
3871                  */
3872                 bond_set_promiscuity(bond, 1);
3873
3874
3875         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
3876                 bond_set_promiscuity(bond, -1);
3877
3878
3879         /* set allmulti flag to slaves */
3880         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
3881                 /*
3882                  * FIXME: Need to handle the error when one of the multi-slaves
3883                  * encounters error.
3884                  */
3885                 bond_set_allmulti(bond, 1);
3886
3887
3888         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
3889                 bond_set_allmulti(bond, -1);
3890
3891
3892         read_lock(&bond->lock);
3893
3894         bond->flags = bond_dev->flags;
3895
3896         /* looking for addresses to add to slaves' mc list */
3897         netdev_for_each_mc_addr(ha, bond_dev) {
3898                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3899                                              bond_dev->addr_len);
3900                 if (!found)
3901                         bond_mc_add(bond, ha->addr);
3902         }
3903
3904         /* looking for addresses to delete from slaves' list */
3905         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3906                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3907                                              bond_dev->addr_len);
3908                 if (!found)
3909                         bond_mc_del(bond, ha->addr);
3910         }
3911
3912         /* save master's multicast list */
3913         __hw_addr_flush(&bond->mc_list);
3914         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3915                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3916
3917         read_unlock(&bond->lock);
3918 }
3919
3920 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3921 {
3922         struct bonding *bond = netdev_priv(dev);
3923         struct slave *slave = bond->first_slave;
3924
3925         if (slave) {
3926                 const struct net_device_ops *slave_ops
3927                         = slave->dev->netdev_ops;
3928                 if (slave_ops->ndo_neigh_setup)
3929                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
3930         }
3931         return 0;
3932 }
3933
3934 /*
3935  * Change the MTU of all of a master's slaves to match the master
3936  */
3937 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3938 {
3939         struct bonding *bond = netdev_priv(bond_dev);
3940         struct slave *slave, *stop_at;
3941         int res = 0;
3942         int i;
3943
3944         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3945                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3946
3947         /* Can't hold bond->lock with bh disabled here since
3948          * some base drivers panic. On the other hand we can't
3949          * hold bond->lock without bh disabled because we'll
3950          * deadlock. The only solution is to rely on the fact
3951          * that we're under rtnl_lock here, and the slaves
3952          * list won't change. This doesn't solve the problem
3953          * of setting the slave's MTU while it is
3954          * transmitting, but the assumption is that the base
3955          * driver can handle that.
3956          *
3957          * TODO: figure out a way to safely iterate the slaves
3958          * list, but without holding a lock around the actual
3959          * call to the base driver.
3960          */
3961
3962         bond_for_each_slave(bond, slave, i) {
3963                 pr_debug("s %p s->p %p c_m %p\n",
3964                          slave,
3965                          slave->prev,
3966                          slave->dev->netdev_ops->ndo_change_mtu);
3967
3968                 res = dev_set_mtu(slave->dev, new_mtu);
3969
3970                 if (res) {
3971                         /* If we failed to set the slave's mtu to the new value
3972                          * we must abort the operation even in ACTIVE_BACKUP
3973                          * mode, because if we allow the backup slaves to have
3974                          * different mtu values than the active slave we'll
3975                          * need to change their mtu when doing a failover. That
3976                          * means changing their mtu from timer context, which
3977                          * is probably not a good idea.
3978                          */
3979                         pr_debug("err %d %s\n", res, slave->dev->name);
3980                         goto unwind;
3981                 }
3982         }
3983
3984         bond_dev->mtu = new_mtu;
3985
3986         return 0;
3987
3988 unwind:
3989         /* unwind from head to the slave that failed */
3990         stop_at = slave;
3991         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3992                 int tmp_res;
3993
3994                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3995                 if (tmp_res) {
3996                         pr_debug("unwind err %d dev %s\n",
3997                                  tmp_res, slave->dev->name);
3998                 }
3999         }
4000
4001         return res;
4002 }
4003
4004 /*
4005  * Change HW address
4006  *
4007  * Note that many devices must be down to change the HW address, and
4008  * downing the master releases all slaves.  We can make bonds full of
4009  * bonding devices to test this, however.
4010  */
4011 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4012 {
4013         struct bonding *bond = netdev_priv(bond_dev);
4014         struct sockaddr *sa = addr, tmp_sa;
4015         struct slave *slave, *stop_at;
4016         int res = 0;
4017         int i;
4018
4019         if (bond->params.mode == BOND_MODE_ALB)
4020                 return bond_alb_set_mac_address(bond_dev, addr);
4021
4022
4023         pr_debug("bond=%p, name=%s\n",
4024                  bond, bond_dev ? bond_dev->name : "None");
4025
4026         /*
4027          * If fail_over_mac is set to active, do nothing and return
4028          * success.  Returning an error causes ifenslave to fail.
4029          */
4030         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4031                 return 0;
4032
4033         if (!is_valid_ether_addr(sa->sa_data))
4034                 return -EADDRNOTAVAIL;
4035
4036         /* Can't hold bond->lock with bh disabled here since
4037          * some base drivers panic. On the other hand we can't
4038          * hold bond->lock without bh disabled because we'll
4039          * deadlock. The only solution is to rely on the fact
4040          * that we're under rtnl_lock here, and the slaves
4041          * list won't change. This doesn't solve the problem
4042          * of setting the slave's hw address while it is
4043          * transmitting, but the assumption is that the base
4044          * driver can handle that.
4045          *
4046          * TODO: figure out a way to safely iterate the slaves
4047          * list, but without holding a lock around the actual
4048          * call to the base driver.
4049          */
4050
4051         bond_for_each_slave(bond, slave, i) {
4052                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4053                 pr_debug("slave %p %s\n", slave, slave->dev->name);
4054
4055                 if (slave_ops->ndo_set_mac_address == NULL) {
4056                         res = -EOPNOTSUPP;
4057                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4058                         goto unwind;
4059                 }
4060
4061                 res = dev_set_mac_address(slave->dev, addr);
4062                 if (res) {
4063                         /* TODO: consider downing the slave
4064                          * and retry ?
4065                          * User should expect communications
4066                          * breakage anyway until ARP finish
4067                          * updating, so...
4068                          */
4069                         pr_debug("err %d %s\n", res, slave->dev->name);
4070                         goto unwind;
4071                 }
4072         }
4073
4074         /* success */
4075         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4076         return 0;
4077
4078 unwind:
4079         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4080         tmp_sa.sa_family = bond_dev->type;
4081
4082         /* unwind from head to the slave that failed */
4083         stop_at = slave;
4084         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4085                 int tmp_res;
4086
4087                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4088                 if (tmp_res) {
4089                         pr_debug("unwind err %d dev %s\n",
4090                                  tmp_res, slave->dev->name);
4091                 }
4092         }
4093
4094         return res;
4095 }
4096
4097 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4098 {
4099         struct bonding *bond = netdev_priv(bond_dev);
4100         struct slave *slave, *start_at;
4101         int i, slave_no, res = 1;
4102         struct iphdr *iph = ip_hdr(skb);
4103
4104         read_lock(&bond->lock);
4105
4106         if (!BOND_IS_OK(bond))
4107                 goto out;
4108         /*
4109          * Start with the curr_active_slave that joined the bond as the
4110          * default for sending IGMP traffic.  For failover purposes one
4111          * needs to maintain some consistency for the interface that will
4112          * send the join/membership reports.  The curr_active_slave found
4113          * will send all of this type of traffic.
4114          */
4115         if ((iph->protocol == IPPROTO_IGMP) &&
4116             (skb->protocol == htons(ETH_P_IP))) {
4117
4118                 read_lock(&bond->curr_slave_lock);
4119                 slave = bond->curr_active_slave;
4120                 read_unlock(&bond->curr_slave_lock);
4121
4122                 if (!slave)
4123                         goto out;
4124         } else {
4125                 /*
4126                  * Concurrent TX may collide on rr_tx_counter; we accept
4127                  * that as being rare enough not to justify using an
4128                  * atomic op here.
4129                  */
4130                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4131
4132                 bond_for_each_slave(bond, slave, i) {
4133                         slave_no--;
4134                         if (slave_no < 0)
4135                                 break;
4136                 }
4137         }
4138
4139         start_at = slave;
4140         bond_for_each_slave_from(bond, slave, i, start_at) {
4141                 if (IS_UP(slave->dev) &&
4142                     (slave->link == BOND_LINK_UP) &&
4143                     bond_is_active_slave(slave)) {
4144                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4145                         break;
4146                 }
4147         }
4148
4149 out:
4150         if (res) {
4151                 /* no suitable interface, frame not sent */
4152                 dev_kfree_skb(skb);
4153         }
4154         read_unlock(&bond->lock);
4155         return NETDEV_TX_OK;
4156 }
4157
4158
4159 /*
4160  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4161  * the bond has a usable interface.
4162  */
4163 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4164 {
4165         struct bonding *bond = netdev_priv(bond_dev);
4166         int res = 1;
4167
4168         read_lock(&bond->lock);
4169         read_lock(&bond->curr_slave_lock);
4170
4171         if (!BOND_IS_OK(bond))
4172                 goto out;
4173
4174         if (!bond->curr_active_slave)
4175                 goto out;
4176
4177         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4178
4179 out:
4180         if (res)
4181                 /* no suitable interface, frame not sent */
4182                 dev_kfree_skb(skb);
4183
4184         read_unlock(&bond->curr_slave_lock);
4185         read_unlock(&bond->lock);
4186         return NETDEV_TX_OK;
4187 }
4188
4189 /*
4190  * In bond_xmit_xor() , we determine the output device by using a pre-
4191  * determined xmit_hash_policy(), If the selected device is not enabled,
4192  * find the next active slave.
4193  */
4194 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4195 {
4196         struct bonding *bond = netdev_priv(bond_dev);
4197         struct slave *slave, *start_at;
4198         int slave_no;
4199         int i;
4200         int res = 1;
4201
4202         read_lock(&bond->lock);
4203
4204         if (!BOND_IS_OK(bond))
4205                 goto out;
4206
4207         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4208
4209         bond_for_each_slave(bond, slave, i) {
4210                 slave_no--;
4211                 if (slave_no < 0)
4212                         break;
4213         }
4214
4215         start_at = slave;
4216
4217         bond_for_each_slave_from(bond, slave, i, start_at) {
4218                 if (IS_UP(slave->dev) &&
4219                     (slave->link == BOND_LINK_UP) &&
4220                     bond_is_active_slave(slave)) {
4221                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4222                         break;
4223                 }
4224         }
4225
4226 out:
4227         if (res) {
4228                 /* no suitable interface, frame not sent */
4229                 dev_kfree_skb(skb);
4230         }
4231         read_unlock(&bond->lock);
4232         return NETDEV_TX_OK;
4233 }
4234
4235 /*
4236  * in broadcast mode, we send everything to all usable interfaces.
4237  */
4238 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4239 {
4240         struct bonding *bond = netdev_priv(bond_dev);
4241         struct slave *slave, *start_at;
4242         struct net_device *tx_dev = NULL;
4243         int i;
4244         int res = 1;
4245
4246         read_lock(&bond->lock);
4247
4248         if (!BOND_IS_OK(bond))
4249                 goto out;
4250
4251         read_lock(&bond->curr_slave_lock);
4252         start_at = bond->curr_active_slave;
4253         read_unlock(&bond->curr_slave_lock);
4254
4255         if (!start_at)
4256                 goto out;
4257
4258         bond_for_each_slave_from(bond, slave, i, start_at) {
4259                 if (IS_UP(slave->dev) &&
4260                     (slave->link == BOND_LINK_UP) &&
4261                     bond_is_active_slave(slave)) {
4262                         if (tx_dev) {
4263                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4264                                 if (!skb2) {
4265                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4266                                                bond_dev->name);
4267                                         continue;
4268                                 }
4269
4270                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4271                                 if (res) {
4272                                         dev_kfree_skb(skb2);
4273                                         continue;
4274                                 }
4275                         }
4276                         tx_dev = slave->dev;
4277                 }
4278         }
4279
4280         if (tx_dev)
4281                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4282
4283 out:
4284         if (res)
4285                 /* no suitable interface, frame not sent */
4286                 dev_kfree_skb(skb);
4287
4288         /* frame sent to all suitable interfaces */
4289         read_unlock(&bond->lock);
4290         return NETDEV_TX_OK;
4291 }
4292
4293 /*------------------------- Device initialization ---------------------------*/
4294
4295 static void bond_set_xmit_hash_policy(struct bonding *bond)
4296 {
4297         switch (bond->params.xmit_policy) {
4298         case BOND_XMIT_POLICY_LAYER23:
4299                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4300                 break;
4301         case BOND_XMIT_POLICY_LAYER34:
4302                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4303                 break;
4304         case BOND_XMIT_POLICY_LAYER2:
4305         default:
4306                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4307                 break;
4308         }
4309 }
4310
4311 /*
4312  * Lookup the slave that corresponds to a qid
4313  */
4314 static inline int bond_slave_override(struct bonding *bond,
4315                                       struct sk_buff *skb)
4316 {
4317         int i, res = 1;
4318         struct slave *slave = NULL;
4319         struct slave *check_slave;
4320
4321         read_lock(&bond->lock);
4322
4323         if (!BOND_IS_OK(bond) || !skb->queue_mapping)
4324                 goto out;
4325
4326         /* Find out if any slaves have the same mapping as this skb. */
4327         bond_for_each_slave(bond, check_slave, i) {
4328                 if (check_slave->queue_id == skb->queue_mapping) {
4329                         slave = check_slave;
4330                         break;
4331                 }
4332         }
4333
4334         /* If the slave isn't UP, use default transmit policy. */
4335         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4336             (slave->link == BOND_LINK_UP)) {
4337                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4338         }
4339
4340 out:
4341         read_unlock(&bond->lock);
4342         return res;
4343 }
4344
4345 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4346 {
4347         /*
4348          * This helper function exists to help dev_pick_tx get the correct
4349          * destination queue.  Using a helper function skips a call to
4350          * skb_tx_hash and will put the skbs in the queue we expect on their
4351          * way down to the bonding driver.
4352          */
4353         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4354
4355         if (unlikely(txq >= dev->real_num_tx_queues)) {
4356                 do
4357                         txq -= dev->real_num_tx_queues;
4358                 while (txq >= dev->real_num_tx_queues);
4359         }
4360         return txq;
4361 }
4362
4363 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4364 {
4365         struct bonding *bond = netdev_priv(dev);
4366
4367         /*
4368          * If we risk deadlock from transmitting this in the
4369          * netpoll path, tell netpoll to queue the frame for later tx
4370          */
4371         if (is_netpoll_tx_blocked(dev))
4372                 return NETDEV_TX_BUSY;
4373
4374         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4375                 if (!bond_slave_override(bond, skb))
4376                         return NETDEV_TX_OK;
4377         }
4378
4379         switch (bond->params.mode) {
4380         case BOND_MODE_ROUNDROBIN:
4381                 return bond_xmit_roundrobin(skb, dev);
4382         case BOND_MODE_ACTIVEBACKUP:
4383                 return bond_xmit_activebackup(skb, dev);
4384         case BOND_MODE_XOR:
4385                 return bond_xmit_xor(skb, dev);
4386         case BOND_MODE_BROADCAST:
4387                 return bond_xmit_broadcast(skb, dev);
4388         case BOND_MODE_8023AD:
4389                 return bond_3ad_xmit_xor(skb, dev);
4390         case BOND_MODE_ALB:
4391         case BOND_MODE_TLB:
4392                 return bond_alb_xmit(skb, dev);
4393         default:
4394                 /* Should never happen, mode already checked */
4395                 pr_err("%s: Error: Unknown bonding mode %d\n",
4396                        dev->name, bond->params.mode);
4397                 WARN_ON_ONCE(1);
4398                 dev_kfree_skb(skb);
4399                 return NETDEV_TX_OK;
4400         }
4401 }
4402
4403
4404 /*
4405  * set bond mode specific net device operations
4406  */
4407 void bond_set_mode_ops(struct bonding *bond, int mode)
4408 {
4409         struct net_device *bond_dev = bond->dev;
4410
4411         switch (mode) {
4412         case BOND_MODE_ROUNDROBIN:
4413                 break;
4414         case BOND_MODE_ACTIVEBACKUP:
4415                 break;
4416         case BOND_MODE_XOR:
4417                 bond_set_xmit_hash_policy(bond);
4418                 break;
4419         case BOND_MODE_BROADCAST:
4420                 break;
4421         case BOND_MODE_8023AD:
4422                 bond_set_xmit_hash_policy(bond);
4423                 break;
4424         case BOND_MODE_ALB:
4425                 /* FALLTHRU */
4426         case BOND_MODE_TLB:
4427                 break;
4428         default:
4429                 /* Should never happen, mode already checked */
4430                 pr_err("%s: Error: Unknown bonding mode %d\n",
4431                        bond_dev->name, mode);
4432                 break;
4433         }
4434 }
4435
4436 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4437                                     struct ethtool_drvinfo *drvinfo)
4438 {
4439         strncpy(drvinfo->driver, DRV_NAME, 32);
4440         strncpy(drvinfo->version, DRV_VERSION, 32);
4441         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4442 }
4443
4444 static const struct ethtool_ops bond_ethtool_ops = {
4445         .get_drvinfo            = bond_ethtool_get_drvinfo,
4446         .get_link               = ethtool_op_get_link,
4447         .get_tx_csum            = ethtool_op_get_tx_csum,
4448         .get_sg                 = ethtool_op_get_sg,
4449         .get_tso                = ethtool_op_get_tso,
4450         .get_ufo                = ethtool_op_get_ufo,
4451         .get_flags              = ethtool_op_get_flags,
4452 };
4453
4454 static const struct net_device_ops bond_netdev_ops = {
4455         .ndo_init               = bond_init,
4456         .ndo_uninit             = bond_uninit,
4457         .ndo_open               = bond_open,
4458         .ndo_stop               = bond_close,
4459         .ndo_start_xmit         = bond_start_xmit,
4460         .ndo_select_queue       = bond_select_queue,
4461         .ndo_get_stats64        = bond_get_stats,
4462         .ndo_do_ioctl           = bond_do_ioctl,
4463         .ndo_set_multicast_list = bond_set_multicast_list,
4464         .ndo_change_mtu         = bond_change_mtu,
4465         .ndo_set_mac_address    = bond_set_mac_address,
4466         .ndo_neigh_setup        = bond_neigh_setup,
4467         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4468         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4469         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4470 #ifdef CONFIG_NET_POLL_CONTROLLER
4471         .ndo_netpoll_setup      = bond_netpoll_setup,
4472         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4473         .ndo_poll_controller    = bond_poll_controller,
4474 #endif
4475         .ndo_add_slave          = bond_enslave,
4476         .ndo_del_slave          = bond_release,
4477 };
4478
4479 static void bond_destructor(struct net_device *bond_dev)
4480 {
4481         struct bonding *bond = netdev_priv(bond_dev);
4482         if (bond->wq)
4483                 destroy_workqueue(bond->wq);
4484         free_netdev(bond_dev);
4485 }
4486
4487 static void bond_setup(struct net_device *bond_dev)
4488 {
4489         struct bonding *bond = netdev_priv(bond_dev);
4490
4491         /* initialize rwlocks */
4492         rwlock_init(&bond->lock);
4493         rwlock_init(&bond->curr_slave_lock);
4494
4495         bond->params = bonding_defaults;
4496
4497         /* Initialize pointers */
4498         bond->dev = bond_dev;
4499         INIT_LIST_HEAD(&bond->vlan_list);
4500
4501         /* Initialize the device entry points */
4502         ether_setup(bond_dev);
4503         bond_dev->netdev_ops = &bond_netdev_ops;
4504         bond_dev->ethtool_ops = &bond_ethtool_ops;
4505         bond_set_mode_ops(bond, bond->params.mode);
4506
4507         bond_dev->destructor = bond_destructor;
4508
4509         /* Initialize the device options */
4510         bond_dev->tx_queue_len = 0;
4511         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4512         bond_dev->priv_flags |= IFF_BONDING;
4513         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4514
4515         /* At first, we block adding VLANs. That's the only way to
4516          * prevent problems that occur when adding VLANs over an
4517          * empty bond. The block will be removed once non-challenged
4518          * slaves are enslaved.
4519          */
4520         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4521
4522         /* don't acquire bond device's netif_tx_lock when
4523          * transmitting */
4524         bond_dev->features |= NETIF_F_LLTX;
4525
4526         /* By default, we declare the bond to be fully
4527          * VLAN hardware accelerated capable. Special
4528          * care is taken in the various xmit functions
4529          * when there are slaves that are not hw accel
4530          * capable
4531          */
4532         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4533                                NETIF_F_HW_VLAN_RX |
4534                                NETIF_F_HW_VLAN_FILTER);
4535
4536         /* By default, we enable GRO on bonding devices.
4537          * Actual support requires lowlevel drivers are GRO ready.
4538          */
4539         bond_dev->features |= NETIF_F_GRO;
4540 }
4541
4542 static void bond_work_cancel_all(struct bonding *bond)
4543 {
4544         write_lock_bh(&bond->lock);
4545         bond->kill_timers = 1;
4546         write_unlock_bh(&bond->lock);
4547
4548         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4549                 cancel_delayed_work(&bond->mii_work);
4550
4551         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4552                 cancel_delayed_work(&bond->arp_work);
4553
4554         if (bond->params.mode == BOND_MODE_ALB &&
4555             delayed_work_pending(&bond->alb_work))
4556                 cancel_delayed_work(&bond->alb_work);
4557
4558         if (bond->params.mode == BOND_MODE_8023AD &&
4559             delayed_work_pending(&bond->ad_work))
4560                 cancel_delayed_work(&bond->ad_work);
4561
4562         if (delayed_work_pending(&bond->mcast_work))
4563                 cancel_delayed_work(&bond->mcast_work);
4564 }
4565
4566 /*
4567 * Destroy a bonding device.
4568 * Must be under rtnl_lock when this function is called.
4569 */
4570 static void bond_uninit(struct net_device *bond_dev)
4571 {
4572         struct bonding *bond = netdev_priv(bond_dev);
4573         struct vlan_entry *vlan, *tmp;
4574
4575         bond_netpoll_cleanup(bond_dev);
4576
4577         /* Release the bonded slaves */
4578         bond_release_all(bond_dev);
4579
4580         list_del(&bond->bond_list);
4581
4582         bond_work_cancel_all(bond);
4583
4584         bond_remove_proc_entry(bond);
4585
4586         bond_debug_unregister(bond);
4587
4588         __hw_addr_flush(&bond->mc_list);
4589
4590         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4591                 list_del(&vlan->vlan_list);
4592                 kfree(vlan);
4593         }
4594 }
4595
4596 /*------------------------- Module initialization ---------------------------*/
4597
4598 /*
4599  * Convert string input module parms.  Accept either the
4600  * number of the mode or its string name.  A bit complicated because
4601  * some mode names are substrings of other names, and calls from sysfs
4602  * may have whitespace in the name (trailing newlines, for example).
4603  */
4604 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4605 {
4606         int modeint = -1, i, rv;
4607         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4608
4609         for (p = (char *)buf; *p; p++)
4610                 if (!(isdigit(*p) || isspace(*p)))
4611                         break;
4612
4613         if (*p)
4614                 rv = sscanf(buf, "%20s", modestr);
4615         else
4616                 rv = sscanf(buf, "%d", &modeint);
4617
4618         if (!rv)
4619                 return -1;
4620
4621         for (i = 0; tbl[i].modename; i++) {
4622                 if (modeint == tbl[i].mode)
4623                         return tbl[i].mode;
4624                 if (strcmp(modestr, tbl[i].modename) == 0)
4625                         return tbl[i].mode;
4626         }
4627
4628         return -1;
4629 }
4630
4631 static int bond_check_params(struct bond_params *params)
4632 {
4633         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4634
4635         /*
4636          * Convert string parameters.
4637          */
4638         if (mode) {
4639                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4640                 if (bond_mode == -1) {
4641                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4642                                mode == NULL ? "NULL" : mode);
4643                         return -EINVAL;
4644                 }
4645         }
4646
4647         if (xmit_hash_policy) {
4648                 if ((bond_mode != BOND_MODE_XOR) &&
4649                     (bond_mode != BOND_MODE_8023AD)) {
4650                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4651                                bond_mode_name(bond_mode));
4652                 } else {
4653                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4654                                                         xmit_hashtype_tbl);
4655                         if (xmit_hashtype == -1) {
4656                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4657                                        xmit_hash_policy == NULL ? "NULL" :
4658                                        xmit_hash_policy);
4659                                 return -EINVAL;
4660                         }
4661                 }
4662         }
4663
4664         if (lacp_rate) {
4665                 if (bond_mode != BOND_MODE_8023AD) {
4666                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4667                                 bond_mode_name(bond_mode));
4668                 } else {
4669                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4670                         if (lacp_fast == -1) {
4671                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4672                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4673                                 return -EINVAL;
4674                         }
4675                 }
4676         }
4677
4678         if (ad_select) {
4679                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4680                 if (params->ad_select == -1) {
4681                         pr_err("Error: Invalid ad_select \"%s\"\n",
4682                                ad_select == NULL ? "NULL" : ad_select);
4683                         return -EINVAL;
4684                 }
4685
4686                 if (bond_mode != BOND_MODE_8023AD) {
4687                         pr_warning("ad_select param only affects 802.3ad mode\n");
4688                 }
4689         } else {
4690                 params->ad_select = BOND_AD_STABLE;
4691         }
4692
4693         if (max_bonds < 0) {
4694                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4695                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4696                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4697         }
4698
4699         if (miimon < 0) {
4700                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4701                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4702                 miimon = BOND_LINK_MON_INTERV;
4703         }
4704
4705         if (updelay < 0) {
4706                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4707                            updelay, INT_MAX);
4708                 updelay = 0;
4709         }
4710
4711         if (downdelay < 0) {
4712                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4713                            downdelay, INT_MAX);
4714                 downdelay = 0;
4715         }
4716
4717         if ((use_carrier != 0) && (use_carrier != 1)) {
4718                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4719                            use_carrier);
4720                 use_carrier = 1;
4721         }
4722
4723         if (num_grat_arp < 0 || num_grat_arp > 255) {
4724                 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4725                            num_grat_arp);
4726                 num_grat_arp = 1;
4727         }
4728
4729         if (num_unsol_na < 0 || num_unsol_na > 255) {
4730                 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4731                            num_unsol_na);
4732                 num_unsol_na = 1;
4733         }
4734
4735         /* reset values for 802.3ad */
4736         if (bond_mode == BOND_MODE_8023AD) {
4737                 if (!miimon) {
4738                         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");
4739                         pr_warning("Forcing miimon to 100msec\n");
4740                         miimon = 100;
4741                 }
4742         }
4743
4744         if (tx_queues < 1 || tx_queues > 255) {
4745                 pr_warning("Warning: tx_queues (%d) should be between "
4746                            "1 and 255, resetting to %d\n",
4747                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4748                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4749         }
4750
4751         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4752                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4753                            "not of valid value (0/1), so it was set to "
4754                            "0\n", all_slaves_active);
4755                 all_slaves_active = 0;
4756         }
4757
4758         if (resend_igmp < 0 || resend_igmp > 255) {
4759                 pr_warning("Warning: resend_igmp (%d) should be between "
4760                            "0 and 255, resetting to %d\n",
4761                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4762                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4763         }
4764
4765         /* reset values for TLB/ALB */
4766         if ((bond_mode == BOND_MODE_TLB) ||
4767             (bond_mode == BOND_MODE_ALB)) {
4768                 if (!miimon) {
4769                         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");
4770                         pr_warning("Forcing miimon to 100msec\n");
4771                         miimon = 100;
4772                 }
4773         }
4774
4775         if (bond_mode == BOND_MODE_ALB) {
4776                 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",
4777                           updelay);
4778         }
4779
4780         if (!miimon) {
4781                 if (updelay || downdelay) {
4782                         /* just warn the user the up/down delay will have
4783                          * no effect since miimon is zero...
4784                          */
4785                         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",
4786                                    updelay, downdelay);
4787                 }
4788         } else {
4789                 /* don't allow arp monitoring */
4790                 if (arp_interval) {
4791                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4792                                    miimon, arp_interval);
4793                         arp_interval = 0;
4794                 }
4795
4796                 if ((updelay % miimon) != 0) {
4797                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4798                                    updelay, miimon,
4799                                    (updelay / miimon) * miimon);
4800                 }
4801
4802                 updelay /= miimon;
4803
4804                 if ((downdelay % miimon) != 0) {
4805                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4806                                    downdelay, miimon,
4807                                    (downdelay / miimon) * miimon);
4808                 }
4809
4810                 downdelay /= miimon;
4811         }
4812
4813         if (arp_interval < 0) {
4814                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4815                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4816                 arp_interval = BOND_LINK_ARP_INTERV;
4817         }
4818
4819         for (arp_ip_count = 0;
4820              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4821              arp_ip_count++) {
4822                 /* not complete check, but should be good enough to
4823                    catch mistakes */
4824                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4825                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4826                                    arp_ip_target[arp_ip_count]);
4827                         arp_interval = 0;
4828                 } else {
4829                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4830                         arp_target[arp_ip_count] = ip;
4831                 }
4832         }
4833
4834         if (arp_interval && !arp_ip_count) {
4835                 /* don't allow arping if no arp_ip_target given... */
4836                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4837                            arp_interval);
4838                 arp_interval = 0;
4839         }
4840
4841         if (arp_validate) {
4842                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4843                         pr_err("arp_validate only supported in active-backup mode\n");
4844                         return -EINVAL;
4845                 }
4846                 if (!arp_interval) {
4847                         pr_err("arp_validate requires arp_interval\n");
4848                         return -EINVAL;
4849                 }
4850
4851                 arp_validate_value = bond_parse_parm(arp_validate,
4852                                                      arp_validate_tbl);
4853                 if (arp_validate_value == -1) {
4854                         pr_err("Error: invalid arp_validate \"%s\"\n",
4855                                arp_validate == NULL ? "NULL" : arp_validate);
4856                         return -EINVAL;
4857                 }
4858         } else
4859                 arp_validate_value = 0;
4860
4861         if (miimon) {
4862                 pr_info("MII link monitoring set to %d ms\n", miimon);
4863         } else if (arp_interval) {
4864                 int i;
4865
4866                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4867                         arp_interval,
4868                         arp_validate_tbl[arp_validate_value].modename,
4869                         arp_ip_count);
4870
4871                 for (i = 0; i < arp_ip_count; i++)
4872                         pr_info(" %s", arp_ip_target[i]);
4873
4874                 pr_info("\n");
4875
4876         } else if (max_bonds) {
4877                 /* miimon and arp_interval not set, we need one so things
4878                  * work as expected, see bonding.txt for details
4879                  */
4880                 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");
4881         }
4882
4883         if (primary && !USES_PRIMARY(bond_mode)) {
4884                 /* currently, using a primary only makes sense
4885                  * in active backup, TLB or ALB modes
4886                  */
4887                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4888                            primary, bond_mode_name(bond_mode));
4889                 primary = NULL;
4890         }
4891
4892         if (primary && primary_reselect) {
4893                 primary_reselect_value = bond_parse_parm(primary_reselect,
4894                                                          pri_reselect_tbl);
4895                 if (primary_reselect_value == -1) {
4896                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4897                                primary_reselect ==
4898                                         NULL ? "NULL" : primary_reselect);
4899                         return -EINVAL;
4900                 }
4901         } else {
4902                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4903         }
4904
4905         if (fail_over_mac) {
4906                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4907                                                       fail_over_mac_tbl);
4908                 if (fail_over_mac_value == -1) {
4909                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4910                                arp_validate == NULL ? "NULL" : arp_validate);
4911                         return -EINVAL;
4912                 }
4913
4914                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4915                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4916         } else {
4917                 fail_over_mac_value = BOND_FOM_NONE;
4918         }
4919
4920         /* fill params struct with the proper values */
4921         params->mode = bond_mode;
4922         params->xmit_policy = xmit_hashtype;
4923         params->miimon = miimon;
4924         params->num_grat_arp = num_grat_arp;
4925         params->num_unsol_na = num_unsol_na;
4926         params->arp_interval = arp_interval;
4927         params->arp_validate = arp_validate_value;
4928         params->updelay = updelay;
4929         params->downdelay = downdelay;
4930         params->use_carrier = use_carrier;
4931         params->lacp_fast = lacp_fast;
4932         params->primary[0] = 0;
4933         params->primary_reselect = primary_reselect_value;
4934         params->fail_over_mac = fail_over_mac_value;
4935         params->tx_queues = tx_queues;
4936         params->all_slaves_active = all_slaves_active;
4937         params->resend_igmp = resend_igmp;
4938
4939         if (primary) {
4940                 strncpy(params->primary, primary, IFNAMSIZ);
4941                 params->primary[IFNAMSIZ - 1] = 0;
4942         }
4943
4944         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4945
4946         return 0;
4947 }
4948
4949 static struct lock_class_key bonding_netdev_xmit_lock_key;
4950 static struct lock_class_key bonding_netdev_addr_lock_key;
4951
4952 static void bond_set_lockdep_class_one(struct net_device *dev,
4953                                        struct netdev_queue *txq,
4954                                        void *_unused)
4955 {
4956         lockdep_set_class(&txq->_xmit_lock,
4957                           &bonding_netdev_xmit_lock_key);
4958 }
4959
4960 static void bond_set_lockdep_class(struct net_device *dev)
4961 {
4962         lockdep_set_class(&dev->addr_list_lock,
4963                           &bonding_netdev_addr_lock_key);
4964         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4965 }
4966
4967 /*
4968  * Called from registration process
4969  */
4970 static int bond_init(struct net_device *bond_dev)
4971 {
4972         struct bonding *bond = netdev_priv(bond_dev);
4973         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4974
4975         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4976
4977         bond->wq = create_singlethread_workqueue(bond_dev->name);
4978         if (!bond->wq)
4979                 return -ENOMEM;
4980
4981         bond_set_lockdep_class(bond_dev);
4982
4983         bond_create_proc_entry(bond);
4984         list_add_tail(&bond->bond_list, &bn->dev_list);
4985
4986         bond_prepare_sysfs_group(bond);
4987
4988         bond_debug_register(bond);
4989
4990         __hw_addr_init(&bond->mc_list);
4991         return 0;
4992 }
4993
4994 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4995 {
4996         if (tb[IFLA_ADDRESS]) {
4997                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4998                         return -EINVAL;
4999                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
5000                         return -EADDRNOTAVAIL;
5001         }
5002         return 0;
5003 }
5004
5005 static struct rtnl_link_ops bond_link_ops __read_mostly = {
5006         .kind           = "bond",
5007         .priv_size      = sizeof(struct bonding),
5008         .setup          = bond_setup,
5009         .validate       = bond_validate,
5010 };
5011
5012 /* Create a new bond based on the specified name and bonding parameters.
5013  * If name is NULL, obtain a suitable "bond%d" name for us.
5014  * Caller must NOT hold rtnl_lock; we need to release it here before we
5015  * set up our sysfs entries.
5016  */
5017 int bond_create(struct net *net, const char *name)
5018 {
5019         struct net_device *bond_dev;
5020         int res;
5021
5022         rtnl_lock();
5023
5024         bond_dev = alloc_netdev_mq(sizeof(struct bonding), name ? name : "",
5025                                 bond_setup, tx_queues);
5026         if (!bond_dev) {
5027                 pr_err("%s: eek! can't alloc netdev!\n", name);
5028                 rtnl_unlock();
5029                 return -ENOMEM;
5030         }
5031
5032         dev_net_set(bond_dev, net);
5033         bond_dev->rtnl_link_ops = &bond_link_ops;
5034
5035         if (!name) {
5036                 res = dev_alloc_name(bond_dev, "bond%d");
5037                 if (res < 0)
5038                         goto out;
5039         } else {
5040                 /*
5041                  * If we're given a name to register
5042                  * we need to ensure that its not already
5043                  * registered
5044                  */
5045                 res = -EEXIST;
5046                 if (__dev_get_by_name(net, name) != NULL)
5047                         goto out;
5048         }
5049
5050         res = register_netdevice(bond_dev);
5051
5052         netif_carrier_off(bond_dev);
5053
5054 out:
5055         rtnl_unlock();
5056         if (res < 0)
5057                 bond_destructor(bond_dev);
5058         return res;
5059 }
5060
5061 static int __net_init bond_net_init(struct net *net)
5062 {
5063         struct bond_net *bn = net_generic(net, bond_net_id);
5064
5065         bn->net = net;
5066         INIT_LIST_HEAD(&bn->dev_list);
5067
5068         bond_create_proc_dir(bn);
5069
5070         return 0;
5071 }
5072
5073 static void __net_exit bond_net_exit(struct net *net)
5074 {
5075         struct bond_net *bn = net_generic(net, bond_net_id);
5076
5077         bond_destroy_proc_dir(bn);
5078 }
5079
5080 static struct pernet_operations bond_net_ops = {
5081         .init = bond_net_init,
5082         .exit = bond_net_exit,
5083         .id   = &bond_net_id,
5084         .size = sizeof(struct bond_net),
5085 };
5086
5087 static int __init bonding_init(void)
5088 {
5089         int i;
5090         int res;
5091
5092         pr_info("%s", bond_version);
5093
5094         res = bond_check_params(&bonding_defaults);
5095         if (res)
5096                 goto out;
5097
5098         res = register_pernet_subsys(&bond_net_ops);
5099         if (res)
5100                 goto out;
5101
5102         res = rtnl_link_register(&bond_link_ops);
5103         if (res)
5104                 goto err_link;
5105
5106         bond_create_debugfs();
5107
5108         for (i = 0; i < max_bonds; i++) {
5109                 res = bond_create(&init_net, NULL);
5110                 if (res)
5111                         goto err;
5112         }
5113
5114         res = bond_create_sysfs();
5115         if (res)
5116                 goto err;
5117
5118         register_netdevice_notifier(&bond_netdev_notifier);
5119         register_inetaddr_notifier(&bond_inetaddr_notifier);
5120         bond_register_ipv6_notifier();
5121 out:
5122         return res;
5123 err:
5124         rtnl_link_unregister(&bond_link_ops);
5125 err_link:
5126         unregister_pernet_subsys(&bond_net_ops);
5127         goto out;
5128
5129 }
5130
5131 static void __exit bonding_exit(void)
5132 {
5133         unregister_netdevice_notifier(&bond_netdev_notifier);
5134         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5135         bond_unregister_ipv6_notifier();
5136
5137         bond_destroy_sysfs();
5138         bond_destroy_debugfs();
5139
5140         rtnl_link_unregister(&bond_link_ops);
5141         unregister_pernet_subsys(&bond_net_ops);
5142
5143 #ifdef CONFIG_NET_POLL_CONTROLLER
5144         /*
5145          * Make sure we don't have an imbalance on our netpoll blocking
5146          */
5147         WARN_ON(atomic_read(&netpoll_block_tx));
5148 #endif
5149 }
5150
5151 module_init(bonding_init);
5152 module_exit(bonding_exit);
5153 MODULE_LICENSE("GPL");
5154 MODULE_VERSION(DRV_VERSION);
5155 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5156 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5157 MODULE_ALIAS_RTNL_LINK("bond");