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