drivers/net/bonding/bond_main.c

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