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 int min_links;
 102 static char *ad_select;
 103 static char *xmit_hash_policy;
 104 static int arp_interval = BOND_LINK_ARP_INTERV;
 105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 106 static char *arp_validate;
 107 static char *fail_over_mac;
 108 static int all_slaves_active = 0;
 109 static struct bond_params bonding_defaults;
 110 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 111
 112 module_param(max_bonds, int, 0);
 113 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 114 module_param(tx_queues, int, 0);
 115 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 116 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 117 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 118                                "failover event (alias of num_unsol_na)");
 119 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 120 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 121                                "failover event (alias of num_grat_arp)");
 122 module_param(miimon, int, 0);
 123 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 124 module_param(updelay, int, 0);
 125 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 126 module_param(downdelay, int, 0);
 127 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 128                             "in milliseconds");
 129 module_param(use_carrier, int, 0);
 130 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 131                               "0 for off, 1 for on (default)");
 132 module_param(mode, charp, 0);
 133 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 134                        "1 for active-backup, 2 for balance-xor, "
 135                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 136                        "6 for balance-alb");
 137 module_param(primary, charp, 0);
 138 MODULE_PARM_DESC(primary, "Primary network device to use");
 139 module_param(primary_reselect, charp, 0);
 140 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 141                                    "once it comes up; "
 142                                    "0 for always (default), "
 143                                    "1 for only if speed of primary is "
 144                                    "better, "
 145                                    "2 for only on active slave "
 146                                    "failure");
 147 module_param(lacp_rate, charp, 0);
 148 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 149                             "0 for slow, 1 for fast");
 150 module_param(ad_select, charp, 0);
 151 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 152                             "0 for stable (default), 1 for bandwidth, "
 153                             "2 for count");
 154 module_param(min_links, int, 0);
 155 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 156
 157 module_param(xmit_hash_policy, charp, 0);
 158 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 159                                    "0 for layer 2 (default), 1 for layer 3+4, "
 160                                    "2 for layer 2+3");
 161 module_param(arp_interval, int, 0);
 162 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 163 module_param_array(arp_ip_target, charp, NULL, 0);
 164 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 165 module_param(arp_validate, charp, 0);
 166 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 167                                "0 for none (default), 1 for active, "
 168                                "2 for backup, 3 for all");
 169 module_param(fail_over_mac, charp, 0);
 170 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 171                                 "the same MAC; 0 for none (default), "
 172                                 "1 for active, 2 for follow");
 173 module_param(all_slaves_active, int, 0);
 174 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 175                                      "by setting active flag for all slaves; "
 176                                      "0 for never (default), 1 for always.");
 177 module_param(resend_igmp, int, 0);
 178 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 179                               "link failure");
 180
 181 /*----------------------------- Global variables ----------------------------*/
 182
 183 #ifdef CONFIG_NET_POLL_CONTROLLER
 184 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 185 #endif
 186
 187 int bond_net_id __read_mostly;
 188
 189 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 190 static int arp_ip_count;
 191 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 192 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 193 static int lacp_fast;
 194
 195 const struct bond_parm_tbl bond_lacp_tbl[] = {
 196 {       "slow",         AD_LACP_SLOW},
 197 {       "fast",         AD_LACP_FAST},
 198 {       NULL,           -1},
 199 };
 200
 201 const struct bond_parm_tbl bond_mode_tbl[] = {
 202 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 203 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 204 {       "balance-xor",          BOND_MODE_XOR},
 205 {       "broadcast",            BOND_MODE_BROADCAST},
 206 {       "802.3ad",              BOND_MODE_8023AD},
 207 {       "balance-tlb",          BOND_MODE_TLB},
 208 {       "balance-alb",          BOND_MODE_ALB},
 209 {       NULL,                   -1},
 210 };
 211
 212 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 213 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 214 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 215 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 216 {       NULL,                   -1},
 217 };
 218
 219 const struct bond_parm_tbl arp_validate_tbl[] = {
 220 {       "none",                 BOND_ARP_VALIDATE_NONE},
 221 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 222 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 223 {       "all",                  BOND_ARP_VALIDATE_ALL},
 224 {       NULL,                   -1},
 225 };
 226
 227 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 228 {       "none",                 BOND_FOM_NONE},
 229 {       "active",               BOND_FOM_ACTIVE},
 230 {       "follow",               BOND_FOM_FOLLOW},
 231 {       NULL,                   -1},
 232 };
 233
 234 const struct bond_parm_tbl pri_reselect_tbl[] = {
 235 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 236 {       "better",               BOND_PRI_RESELECT_BETTER},
 237 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 238 {       NULL,                   -1},
 239 };
 240
 241 struct bond_parm_tbl ad_select_tbl[] = {
 242 {       "stable",       BOND_AD_STABLE},
 243 {       "bandwidth",    BOND_AD_BANDWIDTH},
 244 {       "count",        BOND_AD_COUNT},
 245 {       NULL,           -1},
 246 };
 247
 248 /*-------------------------- Forward declarations ---------------------------*/
 249
 250 static int bond_init(struct net_device *bond_dev);
 251 static void bond_uninit(struct net_device *bond_dev);
 252
 253 /*---------------------------- General routines -----------------------------*/
 254
 255 const char *bond_mode_name(int mode)
 256 {
 257         static const char *names[] = {
 258                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 259                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 260                 [BOND_MODE_XOR] = "load balancing (xor)",
 261                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 262                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 263                 [BOND_MODE_TLB] = "transmit load balancing",
 264                 [BOND_MODE_ALB] = "adaptive load balancing",
 265         };
 266
 267         if (mode < 0 || mode > BOND_MODE_ALB)
 268                 return "unknown";
 269
 270         return names[mode];
 271 }
 272
 273 /*---------------------------------- VLAN -----------------------------------*/
 274
 275 /**
 276  * bond_add_vlan - add a new vlan id on bond
 277  * @bond: bond that got the notification
 278  * @vlan_id: the vlan id to add
 279  *
 280  * Returns -ENOMEM if allocation failed.
 281  */
 282 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 283 {
 284         struct vlan_entry *vlan;
 285
 286         pr_debug("bond: %s, vlan id %d\n",
 287                  (bond ? bond->dev->name : "None"), vlan_id);
 288
 289         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 290         if (!vlan)
 291                 return -ENOMEM;
 292
 293         INIT_LIST_HEAD(&vlan->vlan_list);
 294         vlan->vlan_id = vlan_id;
 295
 296         write_lock_bh(&bond->lock);
 297
 298         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 299
 300         write_unlock_bh(&bond->lock);
 301
 302         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 303
 304         return 0;
 305 }
 306
 307 /**
 308  * bond_del_vlan - delete a vlan id from bond
 309  * @bond: bond that got the notification
 310  * @vlan_id: the vlan id to delete
 311  *
 312  * returns -ENODEV if @vlan_id was not found in @bond.
 313  */
 314 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 315 {
 316         struct vlan_entry *vlan;
 317         int res = -ENODEV;
 318
 319         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 320
 321         block_netpoll_tx();
 322         write_lock_bh(&bond->lock);
 323
 324         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 325                 if (vlan->vlan_id == vlan_id) {
 326                         list_del(&vlan->vlan_list);
 327
 328                         if (bond_is_lb(bond))
 329                                 bond_alb_clear_vlan(bond, vlan_id);
 330
 331                         pr_debug("removed VLAN ID %d from bond %s\n",
 332                                  vlan_id, bond->dev->name);
 333
 334                         kfree(vlan);
 335
 336                         res = 0;
 337                         goto out;
 338                 }
 339         }
 340
 341         pr_debug("couldn't find VLAN ID %d in bond %s\n",
 342                  vlan_id, bond->dev->name);
 343
 344 out:
 345         write_unlock_bh(&bond->lock);
 346         unblock_netpoll_tx();
 347         return res;
 348 }
 349
 350 /**
 351  * bond_next_vlan - safely skip to the next item in the vlans list.
 352  * @bond: the bond we're working on
 353  * @curr: item we're advancing from
 354  *
 355  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 356  * or @curr->next otherwise (even if it is @curr itself again).
 357  *
 358  * Caller must hold bond->lock
 359  */
 360 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 361 {
 362         struct vlan_entry *next, *last;
 363
 364         if (list_empty(&bond->vlan_list))
 365                 return NULL;
 366
 367         if (!curr) {
 368                 next = list_entry(bond->vlan_list.next,
 369                                   struct vlan_entry, vlan_list);
 370         } else {
 371                 last = list_entry(bond->vlan_list.prev,
 372                                   struct vlan_entry, vlan_list);
 373                 if (last == curr) {
 374                         next = list_entry(bond->vlan_list.next,
 375                                           struct vlan_entry, vlan_list);
 376                 } else {
 377                         next = list_entry(curr->vlan_list.next,
 378                                           struct vlan_entry, vlan_list);
 379                 }
 380         }
 381
 382         return next;
 383 }
 384
 385 #define bond_queue_mapping(skb) (*(u16 *)((skb)->cb))
 386
 387 /**
 388  * bond_dev_queue_xmit - Prepare skb for xmit.
 389  *
 390  * @bond: bond device that got this skb for tx.
 391  * @skb: hw accel VLAN tagged skb to transmit
 392  * @slave_dev: slave that is supposed to xmit this skbuff
 393  */
 394 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 395                         struct net_device *slave_dev)
 396 {
 397         skb->dev = slave_dev;
 398         skb->priority = 1;
 399
 400         skb->queue_mapping = bond_queue_mapping(skb);
 401
 402         if (unlikely(netpoll_tx_running(slave_dev)))
 403                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 404         else
 405                 dev_queue_xmit(skb);
 406
 407         return 0;
 408 }
 409
 410 /*
 411  * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 412  * We don't protect the slave list iteration with a lock because:
 413  * a. This operation is performed in IOCTL context,
 414  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 415  * c. Holding a lock with BH disabled while directly calling a base driver
 416  *    entry point is generally a BAD idea.
 417  *
 418  * The design of synchronization/protection for this operation in the 8021q
 419  * module is good for one or more VLAN devices over a single physical device
 420  * and cannot be extended for a teaming solution like bonding, so there is a
 421  * potential race condition here where a net device from the vlan group might
 422  * be referenced (either by a base driver or the 8021q code) while it is being
 423  * removed from the system. However, it turns out we're not making matters
 424  * worse, and if it works for regular VLAN usage it will work here too.
 425 */
 426
 427 /**
 428  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 429  * @bond_dev: bonding net device that got called
 430  * @vid: vlan id being added
 431  */
 432 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 433 {
 434         struct bonding *bond = netdev_priv(bond_dev);
 435         struct slave *slave;
 436         int i, res;
 437
 438         bond_for_each_slave(bond, slave, i) {
 439                 struct net_device *slave_dev = slave->dev;
 440                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 441
 442                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 443                     slave_ops->ndo_vlan_rx_add_vid) {
 444                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
 445                 }
 446         }
 447
 448         res = bond_add_vlan(bond, vid);
 449         if (res) {
 450                 pr_err("%s: Error: Failed to add vlan id %d\n",
 451                        bond_dev->name, vid);
 452         }
 453 }
 454
 455 /**
 456  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 457  * @bond_dev: bonding net device that got called
 458  * @vid: vlan id being removed
 459  */
 460 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 461 {
 462         struct bonding *bond = netdev_priv(bond_dev);
 463         struct slave *slave;
 464         int i, res;
 465
 466         bond_for_each_slave(bond, slave, i) {
 467                 struct net_device *slave_dev = slave->dev;
 468                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 469
 470                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 471                     slave_ops->ndo_vlan_rx_kill_vid) {
 472                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
 473                 }
 474         }
 475
 476         res = bond_del_vlan(bond, vid);
 477         if (res) {
 478                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 479                        bond_dev->name, vid);
 480         }
 481 }
 482
 483 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 484 {
 485         struct vlan_entry *vlan;
 486         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 487
 488         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 489             !(slave_ops->ndo_vlan_rx_add_vid))
 490                 return;
 491
 492         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
 493                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 494 }
 495
 496 static void bond_del_vlans_from_slave(struct bonding *bond,
 497                                       struct net_device *slave_dev)
 498 {
 499         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 500         struct vlan_entry *vlan;
 501
 502         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 503             !(slave_ops->ndo_vlan_rx_kill_vid))
 504                 return;
 505
 506         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 507                 if (!vlan->vlan_id)
 508                         continue;
 509                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 510         }
 511 }
 512
 513 /*------------------------------- Link status -------------------------------*/
 514
 515 /*
 516  * Set the carrier state for the master according to the state of its
 517  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 518  * do special 802.3ad magic.
 519  *
 520  * Returns zero if carrier state does not change, nonzero if it does.
 521  */
 522 static int bond_set_carrier(struct bonding *bond)
 523 {
 524         struct slave *slave;
 525         int i;
 526
 527         if (bond->slave_cnt == 0)
 528                 goto down;
 529
 530         if (bond->params.mode == BOND_MODE_8023AD)
 531                 return bond_3ad_set_carrier(bond);
 532
 533         bond_for_each_slave(bond, slave, i) {
 534                 if (slave->link == BOND_LINK_UP) {
 535                         if (!netif_carrier_ok(bond->dev)) {
 536                                 netif_carrier_on(bond->dev);
 537                                 return 1;
 538                         }
 539                         return 0;
 540                 }
 541         }
 542
 543 down:
 544         if (netif_carrier_ok(bond->dev)) {
 545                 netif_carrier_off(bond->dev);
 546                 return 1;
 547         }
 548         return 0;
 549 }
 550
 551 /*
 552  * Get link speed and duplex from the slave's base driver
 553  * using ethtool. If for some reason the call fails or the
 554  * values are invalid, fake speed and duplex to 100/Full
 555  * and return error.
 556  */
 557 static int bond_update_speed_duplex(struct slave *slave)
 558 {
 559         struct net_device *slave_dev = slave->dev;
 560         struct ethtool_cmd ecmd;
 561         u32 slave_speed;
 562         int res;
 563
 564         /* Fake speed and duplex */
 565         slave->speed = SPEED_100;
 566         slave->duplex = DUPLEX_FULL;
 567
 568         res = __ethtool_get_settings(slave_dev, &ecmd);
 569         if (res < 0)
 570                 return -1;
 571
 572         slave_speed = ethtool_cmd_speed(&ecmd);
 573         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 574                 return -1;
 575
 576         switch (ecmd.duplex) {
 577         case DUPLEX_FULL:
 578         case DUPLEX_HALF:
 579                 break;
 580         default:
 581                 return -1;
 582         }
 583
 584         slave->speed = slave_speed;
 585         slave->duplex = ecmd.duplex;
 586
 587         return 0;
 588 }
 589
 590 /*
 591  * if <dev> supports MII link status reporting, check its link status.
 592  *
 593  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 594  * depending upon the setting of the use_carrier parameter.
 595  *
 596  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 597  * can't tell and just pretend it is), or 0, meaning that the link is
 598  * down.
 599  *
 600  * If reporting is non-zero, instead of faking link up, return -1 if
 601  * both ETHTOOL and MII ioctls fail (meaning the device does not
 602  * support them).  If use_carrier is set, return whatever it says.
 603  * It'd be nice if there was a good way to tell if a driver supports
 604  * netif_carrier, but there really isn't.
 605  */
 606 static int bond_check_dev_link(struct bonding *bond,
 607                                struct net_device *slave_dev, int reporting)
 608 {
 609         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 610         int (*ioctl)(struct net_device *, struct ifreq *, int);
 611         struct ifreq ifr;
 612         struct mii_ioctl_data *mii;
 613
 614         if (!reporting && !netif_running(slave_dev))
 615                 return 0;
 616
 617         if (bond->params.use_carrier)
 618                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 619
 620         /* Try to get link status using Ethtool first. */
 621         if (slave_dev->ethtool_ops) {
 622                 if (slave_dev->ethtool_ops->get_link) {
 623                         u32 link;
 624
 625                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 626
 627                         return link ? BMSR_LSTATUS : 0;
 628                 }
 629         }
 630
 631         /* Ethtool can't be used, fallback to MII ioctls. */
 632         ioctl = slave_ops->ndo_do_ioctl;
 633         if (ioctl) {
 634                 /* TODO: set pointer to correct ioctl on a per team member */
 635                 /*       bases to make this more efficient. that is, once  */
 636                 /*       we determine the correct ioctl, we will always    */
 637                 /*       call it and not the others for that team          */
 638                 /*       member.                                           */
 639
 640                 /*
 641                  * We cannot assume that SIOCGMIIPHY will also read a
 642                  * register; not all network drivers (e.g., e100)
 643                  * support that.
 644                  */
 645
 646                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 647                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 648                 mii = if_mii(&ifr);
 649                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 650                         mii->reg_num = MII_BMSR;
 651                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 652                                 return mii->val_out & BMSR_LSTATUS;
 653                 }
 654         }
 655
 656         /*
 657          * If reporting, report that either there's no dev->do_ioctl,
 658          * or both SIOCGMIIREG and get_link failed (meaning that we
 659          * cannot report link status).  If not reporting, pretend
 660          * we're ok.
 661          */
 662         return reporting ? -1 : BMSR_LSTATUS;
 663 }
 664
 665 /*----------------------------- Multicast list ------------------------------*/
 666
 667 /*
 668  * Push the promiscuity flag down to appropriate slaves
 669  */
 670 static int bond_set_promiscuity(struct bonding *bond, int inc)
 671 {
 672         int err = 0;
 673         if (USES_PRIMARY(bond->params.mode)) {
 674                 /* write lock already acquired */
 675                 if (bond->curr_active_slave) {
 676                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 677                                                   inc);
 678                 }
 679         } else {
 680                 struct slave *slave;
 681                 int i;
 682                 bond_for_each_slave(bond, slave, i) {
 683                         err = dev_set_promiscuity(slave->dev, inc);
 684                         if (err)
 685                                 return err;
 686                 }
 687         }
 688         return err;
 689 }
 690
 691 /*
 692  * Push the allmulti flag down to all slaves
 693  */
 694 static int bond_set_allmulti(struct bonding *bond, int inc)
 695 {
 696         int err = 0;
 697         if (USES_PRIMARY(bond->params.mode)) {
 698                 /* write lock already acquired */
 699                 if (bond->curr_active_slave) {
 700                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 701                                                inc);
 702                 }
 703         } else {
 704                 struct slave *slave;
 705                 int i;
 706                 bond_for_each_slave(bond, slave, i) {
 707                         err = dev_set_allmulti(slave->dev, inc);
 708                         if (err)
 709                                 return err;
 710                 }
 711         }
 712         return err;
 713 }
 714
 715 /*
 716  * Add a Multicast address to slaves
 717  * according to mode
 718  */
 719 static void bond_mc_add(struct bonding *bond, void *addr)
 720 {
 721         if (USES_PRIMARY(bond->params.mode)) {
 722                 /* write lock already acquired */
 723                 if (bond->curr_active_slave)
 724                         dev_mc_add(bond->curr_active_slave->dev, addr);
 725         } else {
 726                 struct slave *slave;
 727                 int i;
 728
 729                 bond_for_each_slave(bond, slave, i)
 730                         dev_mc_add(slave->dev, addr);
 731         }
 732 }
 733
 734 /*
 735  * Remove a multicast address from slave
 736  * according to mode
 737  */
 738 static void bond_mc_del(struct bonding *bond, void *addr)
 739 {
 740         if (USES_PRIMARY(bond->params.mode)) {
 741                 /* write lock already acquired */
 742                 if (bond->curr_active_slave)
 743                         dev_mc_del(bond->curr_active_slave->dev, addr);
 744         } else {
 745                 struct slave *slave;
 746                 int i;
 747                 bond_for_each_slave(bond, slave, i) {
 748                         dev_mc_del(slave->dev, addr);
 749                 }
 750         }
 751 }
 752
 753
 754 static void __bond_resend_igmp_join_requests(struct net_device *dev)
 755 {
 756         struct in_device *in_dev;
 757
 758         rcu_read_lock();
 759         in_dev = __in_dev_get_rcu(dev);
 760         if (in_dev)
 761                 ip_mc_rejoin_groups(in_dev);
 762         rcu_read_unlock();
 763 }
 764
 765 /*
 766  * Retrieve the list of registered multicast addresses for the bonding
 767  * device and retransmit an IGMP JOIN request to the current active
 768  * slave.
 769  */
 770 static void bond_resend_igmp_join_requests(struct bonding *bond)
 771 {
 772         struct net_device *vlan_dev;
 773         struct vlan_entry *vlan;
 774
 775         read_lock(&bond->lock);
 776
 777         if (bond->kill_timers)
 778                 goto out;
 779
 780         /* rejoin all groups on bond device */
 781         __bond_resend_igmp_join_requests(bond->dev);
 782
 783         /* rejoin all groups on vlan devices */
 784         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 785                 rcu_read_lock();
 786                 vlan_dev = __vlan_find_dev_deep(bond->dev,
 787                                                 vlan->vlan_id);
 788                 rcu_read_unlock();
 789                 if (vlan_dev)
 790                         __bond_resend_igmp_join_requests(vlan_dev);
 791         }
 792
 793         if ((--bond->igmp_retrans > 0) && !bond->kill_timers)
 794                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 795 out:
 796         read_unlock(&bond->lock);
 797 }
 798
 799 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 800 {
 801         struct bonding *bond = container_of(work, struct bonding,
 802                                             mcast_work.work);
 803         bond_resend_igmp_join_requests(bond);
 804 }
 805
 806 /*
 807  * flush all members of flush->mc_list from device dev->mc_list
 808  */
 809 static void bond_mc_list_flush(struct net_device *bond_dev,
 810                                struct net_device *slave_dev)
 811 {
 812         struct bonding *bond = netdev_priv(bond_dev);
 813         struct netdev_hw_addr *ha;
 814
 815         netdev_for_each_mc_addr(ha, bond_dev)
 816                 dev_mc_del(slave_dev, ha->addr);
 817
 818         if (bond->params.mode == BOND_MODE_8023AD) {
 819                 /* del lacpdu mc addr from mc list */
 820                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 821
 822                 dev_mc_del(slave_dev, lacpdu_multicast);
 823         }
 824 }
 825
 826 /*--------------------------- Active slave change ---------------------------*/
 827
 828 /*
 829  * Update the mc list and multicast-related flags for the new and
 830  * old active slaves (if any) according to the multicast mode, and
 831  * promiscuous flags unconditionally.
 832  */
 833 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 834                          struct slave *old_active)
 835 {
 836         struct netdev_hw_addr *ha;
 837
 838         if (!USES_PRIMARY(bond->params.mode))
 839                 /* nothing to do -  mc list is already up-to-date on
 840                  * all slaves
 841                  */
 842                 return;
 843
 844         if (old_active) {
 845                 if (bond->dev->flags & IFF_PROMISC)
 846                         dev_set_promiscuity(old_active->dev, -1);
 847
 848                 if (bond->dev->flags & IFF_ALLMULTI)
 849                         dev_set_allmulti(old_active->dev, -1);
 850
 851                 netdev_for_each_mc_addr(ha, bond->dev)
 852                         dev_mc_del(old_active->dev, ha->addr);
 853         }
 854
 855         if (new_active) {
 856                 /* FIXME: Signal errors upstream. */
 857                 if (bond->dev->flags & IFF_PROMISC)
 858                         dev_set_promiscuity(new_active->dev, 1);
 859
 860                 if (bond->dev->flags & IFF_ALLMULTI)
 861                         dev_set_allmulti(new_active->dev, 1);
 862
 863                 netdev_for_each_mc_addr(ha, bond->dev)
 864                         dev_mc_add(new_active->dev, ha->addr);
 865         }
 866 }
 867
 868 /*
 869  * bond_do_fail_over_mac
 870  *
 871  * Perform special MAC address swapping for fail_over_mac settings
 872  *
 873  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 874  */
 875 static void bond_do_fail_over_mac(struct bonding *bond,
 876                                   struct slave *new_active,
 877                                   struct slave *old_active)
 878         __releases(&bond->curr_slave_lock)
 879         __releases(&bond->lock)
 880         __acquires(&bond->lock)
 881         __acquires(&bond->curr_slave_lock)
 882 {
 883         u8 tmp_mac[ETH_ALEN];
 884         struct sockaddr saddr;
 885         int rv;
 886
 887         switch (bond->params.fail_over_mac) {
 888         case BOND_FOM_ACTIVE:
 889                 if (new_active)
 890                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 891                                new_active->dev->addr_len);
 892                 break;
 893         case BOND_FOM_FOLLOW:
 894                 /*
 895                  * if new_active && old_active, swap them
 896                  * if just old_active, do nothing (going to no active slave)
 897                  * if just new_active, set new_active to bond's MAC
 898                  */
 899                 if (!new_active)
 900                         return;
 901
 902                 write_unlock_bh(&bond->curr_slave_lock);
 903                 read_unlock(&bond->lock);
 904
 905                 if (old_active) {
 906                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 907                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 908                                ETH_ALEN);
 909                         saddr.sa_family = new_active->dev->type;
 910                 } else {
 911                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 912                         saddr.sa_family = bond->dev->type;
 913                 }
 914
 915                 rv = dev_set_mac_address(new_active->dev, &saddr);
 916                 if (rv) {
 917                         pr_err("%s: Error %d setting MAC of slave %s\n",
 918                                bond->dev->name, -rv, new_active->dev->name);
 919                         goto out;
 920                 }
 921
 922                 if (!old_active)
 923                         goto out;
 924
 925                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 926                 saddr.sa_family = old_active->dev->type;
 927
 928                 rv = dev_set_mac_address(old_active->dev, &saddr);
 929                 if (rv)
 930                         pr_err("%s: Error %d setting MAC of slave %s\n",
 931                                bond->dev->name, -rv, new_active->dev->name);
 932 out:
 933                 read_lock(&bond->lock);
 934                 write_lock_bh(&bond->curr_slave_lock);
 935                 break;
 936         default:
 937                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 938                        bond->dev->name, bond->params.fail_over_mac);
 939                 break;
 940         }
 941
 942 }
 943
 944 static bool bond_should_change_active(struct bonding *bond)
 945 {
 946         struct slave *prim = bond->primary_slave;
 947         struct slave *curr = bond->curr_active_slave;
 948
 949         if (!prim || !curr || curr->link != BOND_LINK_UP)
 950                 return true;
 951         if (bond->force_primary) {
 952                 bond->force_primary = false;
 953                 return true;
 954         }
 955         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 956             (prim->speed < curr->speed ||
 957              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 958                 return false;
 959         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 960                 return false;
 961         return true;
 962 }
 963
 964 /**
 965  * find_best_interface - select the best available slave to be the active one
 966  * @bond: our bonding struct
 967  *
 968  * Warning: Caller must hold curr_slave_lock for writing.
 969  */
 970 static struct slave *bond_find_best_slave(struct bonding *bond)
 971 {
 972         struct slave *new_active, *old_active;
 973         struct slave *bestslave = NULL;
 974         int mintime = bond->params.updelay;
 975         int i;
 976
 977         new_active = bond->curr_active_slave;
 978
 979         if (!new_active) { /* there were no active slaves left */
 980                 if (bond->slave_cnt > 0)   /* found one slave */
 981                         new_active = bond->first_slave;
 982                 else
 983                         return NULL; /* still no slave, return NULL */
 984         }
 985
 986         if ((bond->primary_slave) &&
 987             bond->primary_slave->link == BOND_LINK_UP &&
 988             bond_should_change_active(bond)) {
 989                 new_active = bond->primary_slave;
 990         }
 991
 992         /* remember where to stop iterating over the slaves */
 993         old_active = new_active;
 994
 995         bond_for_each_slave_from(bond, new_active, i, old_active) {
 996                 if (new_active->link == BOND_LINK_UP) {
 997                         return new_active;
 998                 } else if (new_active->link == BOND_LINK_BACK &&
 999                            IS_UP(new_active->dev)) {
1000                         /* link up, but waiting for stabilization */
1001                         if (new_active->delay < mintime) {
1002                                 mintime = new_active->delay;
1003                                 bestslave = new_active;
1004                         }
1005                 }
1006         }
1007
1008         return bestslave;
1009 }
1010
1011 static bool bond_should_notify_peers(struct bonding *bond)
1012 {
1013         struct slave *slave = bond->curr_active_slave;
1014
1015         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1016                  bond->dev->name, slave ? slave->dev->name : "NULL");
1017
1018         if (!slave || !bond->send_peer_notif ||
1019             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1020                 return false;
1021
1022         bond->send_peer_notif--;
1023         return true;
1024 }
1025
1026 /**
1027  * change_active_interface - change the active slave into the specified one
1028  * @bond: our bonding struct
1029  * @new: the new slave to make the active one
1030  *
1031  * Set the new slave to the bond's settings and unset them on the old
1032  * curr_active_slave.
1033  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1034  *
1035  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1036  * because it is apparently the best available slave we have, even though its
1037  * updelay hasn't timed out yet.
1038  *
1039  * If new_active is not NULL, caller must hold bond->lock for read and
1040  * curr_slave_lock for write_bh.
1041  */
1042 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1043 {
1044         struct slave *old_active = bond->curr_active_slave;
1045
1046         if (old_active == new_active)
1047                 return;
1048
1049         if (new_active) {
1050                 new_active->jiffies = jiffies;
1051
1052                 if (new_active->link == BOND_LINK_BACK) {
1053                         if (USES_PRIMARY(bond->params.mode)) {
1054                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1055                                         bond->dev->name, new_active->dev->name,
1056                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1057                         }
1058
1059                         new_active->delay = 0;
1060                         new_active->link = BOND_LINK_UP;
1061
1062                         if (bond->params.mode == BOND_MODE_8023AD)
1063                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1064
1065                         if (bond_is_lb(bond))
1066                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1067                 } else {
1068                         if (USES_PRIMARY(bond->params.mode)) {
1069                                 pr_info("%s: making interface %s the new active one.\n",
1070                                         bond->dev->name, new_active->dev->name);
1071                         }
1072                 }
1073         }
1074
1075         if (USES_PRIMARY(bond->params.mode))
1076                 bond_mc_swap(bond, new_active, old_active);
1077
1078         if (bond_is_lb(bond)) {
1079                 bond_alb_handle_active_change(bond, new_active);
1080                 if (old_active)
1081                         bond_set_slave_inactive_flags(old_active);
1082                 if (new_active)
1083                         bond_set_slave_active_flags(new_active);
1084         } else {
1085                 bond->curr_active_slave = new_active;
1086         }
1087
1088         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1089                 if (old_active)
1090                         bond_set_slave_inactive_flags(old_active);
1091
1092                 if (new_active) {
1093                         bool should_notify_peers = false;
1094
1095                         bond_set_slave_active_flags(new_active);
1096
1097                         if (bond->params.fail_over_mac)
1098                                 bond_do_fail_over_mac(bond, new_active,
1099                                                       old_active);
1100
1101                         if (netif_running(bond->dev)) {
1102                                 bond->send_peer_notif =
1103                                         bond->params.num_peer_notif;
1104                                 should_notify_peers =
1105                                         bond_should_notify_peers(bond);
1106                         }
1107
1108                         write_unlock_bh(&bond->curr_slave_lock);
1109                         read_unlock(&bond->lock);
1110
1111                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1112                         if (should_notify_peers)
1113                                 netdev_bonding_change(bond->dev,
1114                                                       NETDEV_NOTIFY_PEERS);
1115
1116                         read_lock(&bond->lock);
1117                         write_lock_bh(&bond->curr_slave_lock);
1118                 }
1119         }
1120
1121         /* resend IGMP joins since active slave has changed or
1122          * all were sent on curr_active_slave.
1123          * resend only if bond is brought up with the affected
1124          * bonding modes and the retransmission is enabled */
1125         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1126             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1127              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1128                 bond->igmp_retrans = bond->params.resend_igmp;
1129                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1130         }
1131 }
1132
1133 /**
1134  * bond_select_active_slave - select a new active slave, if needed
1135  * @bond: our bonding struct
1136  *
1137  * This functions should be called when one of the following occurs:
1138  * - The old curr_active_slave has been released or lost its link.
1139  * - The primary_slave has got its link back.
1140  * - A slave has got its link back and there's no old curr_active_slave.
1141  *
1142  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1143  */
1144 void bond_select_active_slave(struct bonding *bond)
1145 {
1146         struct slave *best_slave;
1147         int rv;
1148
1149         best_slave = bond_find_best_slave(bond);
1150         if (best_slave != bond->curr_active_slave) {
1151                 bond_change_active_slave(bond, best_slave);
1152                 rv = bond_set_carrier(bond);
1153                 if (!rv)
1154                         return;
1155
1156                 if (netif_carrier_ok(bond->dev)) {
1157                         pr_info("%s: first active interface up!\n",
1158                                 bond->dev->name);
1159                 } else {
1160                         pr_info("%s: now running without any active interface !\n",
1161                                 bond->dev->name);
1162                 }
1163         }
1164 }
1165
1166 /*--------------------------- slave list handling ---------------------------*/
1167
1168 /*
1169  * This function attaches the slave to the end of list.
1170  *
1171  * bond->lock held for writing by caller.
1172  */
1173 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1174 {
1175         if (bond->first_slave == NULL) { /* attaching the first slave */
1176                 new_slave->next = new_slave;
1177                 new_slave->prev = new_slave;
1178                 bond->first_slave = new_slave;
1179         } else {
1180                 new_slave->next = bond->first_slave;
1181                 new_slave->prev = bond->first_slave->prev;
1182                 new_slave->next->prev = new_slave;
1183                 new_slave->prev->next = new_slave;
1184         }
1185
1186         bond->slave_cnt++;
1187 }
1188
1189 /*
1190  * This function detaches the slave from the list.
1191  * WARNING: no check is made to verify if the slave effectively
1192  * belongs to <bond>.
1193  * Nothing is freed on return, structures are just unchained.
1194  * If any slave pointer in bond was pointing to <slave>,
1195  * it should be changed by the calling function.
1196  *
1197  * bond->lock held for writing by caller.
1198  */
1199 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1200 {
1201         if (slave->next)
1202                 slave->next->prev = slave->prev;
1203
1204         if (slave->prev)
1205                 slave->prev->next = slave->next;
1206
1207         if (bond->first_slave == slave) { /* slave is the first slave */
1208                 if (bond->slave_cnt > 1) { /* there are more slave */
1209                         bond->first_slave = slave->next;
1210                 } else {
1211                         bond->first_slave = NULL; /* slave was the last one */
1212                 }
1213         }
1214
1215         slave->next = NULL;
1216         slave->prev = NULL;
1217         bond->slave_cnt--;
1218 }
1219
1220 #ifdef CONFIG_NET_POLL_CONTROLLER
1221 static inline int slave_enable_netpoll(struct slave *slave)
1222 {
1223         struct netpoll *np;
1224         int err = 0;
1225
1226         np = kzalloc(sizeof(*np), GFP_KERNEL);
1227         err = -ENOMEM;
1228         if (!np)
1229                 goto out;
1230
1231         np->dev = slave->dev;
1232         strlcpy(np->dev_name, slave->dev->name, IFNAMSIZ);
1233         err = __netpoll_setup(np);
1234         if (err) {
1235                 kfree(np);
1236                 goto out;
1237         }
1238         slave->np = np;
1239 out:
1240         return err;
1241 }
1242 static inline void slave_disable_netpoll(struct slave *slave)
1243 {
1244         struct netpoll *np = slave->np;
1245
1246         if (!np)
1247                 return;
1248
1249         slave->np = NULL;
1250         synchronize_rcu_bh();
1251         __netpoll_cleanup(np);
1252         kfree(np);
1253 }
1254 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1255 {
1256         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1257                 return false;
1258         if (!slave_dev->netdev_ops->ndo_poll_controller)
1259                 return false;
1260         return true;
1261 }
1262
1263 static void bond_poll_controller(struct net_device *bond_dev)
1264 {
1265 }
1266
1267 static void __bond_netpoll_cleanup(struct bonding *bond)
1268 {
1269         struct slave *slave;
1270         int i;
1271
1272         bond_for_each_slave(bond, slave, i)
1273                 if (IS_UP(slave->dev))
1274                         slave_disable_netpoll(slave);
1275 }
1276 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1277 {
1278         struct bonding *bond = netdev_priv(bond_dev);
1279
1280         read_lock(&bond->lock);
1281         __bond_netpoll_cleanup(bond);
1282         read_unlock(&bond->lock);
1283 }
1284
1285 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1286 {
1287         struct bonding *bond = netdev_priv(dev);
1288         struct slave *slave;
1289         int i, err = 0;
1290
1291         read_lock(&bond->lock);
1292         bond_for_each_slave(bond, slave, i) {
1293                 err = slave_enable_netpoll(slave);
1294                 if (err) {
1295                         __bond_netpoll_cleanup(bond);
1296                         break;
1297                 }
1298         }
1299         read_unlock(&bond->lock);
1300         return err;
1301 }
1302
1303 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1304 {
1305         return bond->dev->npinfo;
1306 }
1307
1308 #else
1309 static inline int slave_enable_netpoll(struct slave *slave)
1310 {
1311         return 0;
1312 }
1313 static inline void slave_disable_netpoll(struct slave *slave)
1314 {
1315 }
1316 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1317 {
1318 }
1319 #endif
1320
1321 /*---------------------------------- IOCTL ----------------------------------*/
1322
1323 static int bond_sethwaddr(struct net_device *bond_dev,
1324                           struct net_device *slave_dev)
1325 {
1326         pr_debug("bond_dev=%p\n", bond_dev);
1327         pr_debug("slave_dev=%p\n", slave_dev);
1328         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1329         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1330         return 0;
1331 }
1332
1333 static u32 bond_fix_features(struct net_device *dev, u32 features)
1334 {
1335         struct slave *slave;
1336         struct bonding *bond = netdev_priv(dev);
1337         u32 mask;
1338         int i;
1339
1340         read_lock(&bond->lock);
1341
1342         if (!bond->first_slave) {
1343                 /* Disable adding VLANs to empty bond. But why? --mq */
1344                 features |= NETIF_F_VLAN_CHALLENGED;
1345                 goto out;
1346         }
1347
1348         mask = features;
1349         features &= ~NETIF_F_ONE_FOR_ALL;
1350         features |= NETIF_F_ALL_FOR_ALL;
1351
1352         bond_for_each_slave(bond, slave, i) {
1353                 features = netdev_increment_features(features,
1354                                                      slave->dev->features,
1355                                                      mask);
1356         }
1357
1358 out:
1359         read_unlock(&bond->lock);
1360         return features;
1361 }
1362
1363 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1364                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1365                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1366
1367 static void bond_compute_features(struct bonding *bond)
1368 {
1369         struct slave *slave;
1370         struct net_device *bond_dev = bond->dev;
1371         u32 vlan_features = BOND_VLAN_FEATURES;
1372         unsigned short max_hard_header_len = ETH_HLEN;
1373         int i;
1374
1375         read_lock(&bond->lock);
1376
1377         if (!bond->first_slave)
1378                 goto done;
1379
1380         bond_for_each_slave(bond, slave, i) {
1381                 vlan_features = netdev_increment_features(vlan_features,
1382                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1383
1384                 if (slave->dev->hard_header_len > max_hard_header_len)
1385                         max_hard_header_len = slave->dev->hard_header_len;
1386         }
1387
1388 done:
1389         bond_dev->vlan_features = vlan_features;
1390         bond_dev->hard_header_len = max_hard_header_len;
1391
1392         read_unlock(&bond->lock);
1393
1394         netdev_change_features(bond_dev);
1395 }
1396
1397 static void bond_setup_by_slave(struct net_device *bond_dev,
1398                                 struct net_device *slave_dev)
1399 {
1400         struct bonding *bond = netdev_priv(bond_dev);
1401
1402         bond_dev->header_ops        = slave_dev->header_ops;
1403
1404         bond_dev->type              = slave_dev->type;
1405         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1406         bond_dev->addr_len          = slave_dev->addr_len;
1407
1408         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1409                 slave_dev->addr_len);
1410         bond->setup_by_slave = 1;
1411 }
1412
1413 /* On bonding slaves other than the currently active slave, suppress
1414  * duplicates except for alb non-mcast/bcast.
1415  */
1416 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1417                                             struct slave *slave,
1418                                             struct bonding *bond)
1419 {
1420         if (bond_is_slave_inactive(slave)) {
1421                 if (bond->params.mode == BOND_MODE_ALB &&
1422                     skb->pkt_type != PACKET_BROADCAST &&
1423                     skb->pkt_type != PACKET_MULTICAST)
1424                         return false;
1425                 return true;
1426         }
1427         return false;
1428 }
1429
1430 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1431 {
1432         struct sk_buff *skb = *pskb;
1433         struct slave *slave;
1434         struct bonding *bond;
1435
1436         skb = skb_share_check(skb, GFP_ATOMIC);
1437         if (unlikely(!skb))
1438                 return RX_HANDLER_CONSUMED;
1439
1440         *pskb = skb;
1441
1442         slave = bond_slave_get_rcu(skb->dev);
1443         bond = slave->bond;
1444
1445         if (bond->params.arp_interval)
1446                 slave->dev->last_rx = jiffies;
1447
1448         if (bond->recv_probe) {
1449                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1450
1451                 if (likely(nskb)) {
1452                         bond->recv_probe(nskb, bond, slave);
1453                         dev_kfree_skb(nskb);
1454                 }
1455         }
1456
1457         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1458                 return RX_HANDLER_EXACT;
1459         }
1460
1461         skb->dev = bond->dev;
1462
1463         if (bond->params.mode == BOND_MODE_ALB &&
1464             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1465             skb->pkt_type == PACKET_HOST) {
1466
1467                 if (unlikely(skb_cow_head(skb,
1468                                           skb->data - skb_mac_header(skb)))) {
1469                         kfree_skb(skb);
1470                         return RX_HANDLER_CONSUMED;
1471                 }
1472                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1473         }
1474
1475         return RX_HANDLER_ANOTHER;
1476 }
1477
1478 /* enslave device <slave> to bond device <master> */
1479 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1480 {
1481         struct bonding *bond = netdev_priv(bond_dev);
1482         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1483         struct slave *new_slave = NULL;
1484         struct netdev_hw_addr *ha;
1485         struct sockaddr addr;
1486         int link_reporting;
1487         int res = 0;
1488
1489         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1490                 slave_ops->ndo_do_ioctl == NULL) {
1491                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1492                            bond_dev->name, slave_dev->name);
1493         }
1494
1495         /* already enslaved */
1496         if (slave_dev->flags & IFF_SLAVE) {
1497                 pr_debug("Error, Device was already enslaved\n");
1498                 return -EBUSY;
1499         }
1500
1501         /* vlan challenged mutual exclusion */
1502         /* no need to lock since we're protected by rtnl_lock */
1503         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1504                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1505                 if (bond_vlan_used(bond)) {
1506                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1507                                bond_dev->name, slave_dev->name, bond_dev->name);
1508                         return -EPERM;
1509                 } else {
1510                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1511                                    bond_dev->name, slave_dev->name,
1512                                    slave_dev->name, bond_dev->name);
1513                 }
1514         } else {
1515                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1516         }
1517
1518         /*
1519          * Old ifenslave binaries are no longer supported.  These can
1520          * be identified with moderate accuracy by the state of the slave:
1521          * the current ifenslave will set the interface down prior to
1522          * enslaving it; the old ifenslave will not.
1523          */
1524         if ((slave_dev->flags & IFF_UP)) {
1525                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1526                        slave_dev->name);
1527                 res = -EPERM;
1528                 goto err_undo_flags;
1529         }
1530
1531         /* set bonding device ether type by slave - bonding netdevices are
1532          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1533          * there is a need to override some of the type dependent attribs/funcs.
1534          *
1535          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1536          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1537          */
1538         if (bond->slave_cnt == 0) {
1539                 if (bond_dev->type != slave_dev->type) {
1540                         pr_debug("%s: change device type from %d to %d\n",
1541                                  bond_dev->name,
1542                                  bond_dev->type, slave_dev->type);
1543
1544                         res = netdev_bonding_change(bond_dev,
1545                                                     NETDEV_PRE_TYPE_CHANGE);
1546                         res = notifier_to_errno(res);
1547                         if (res) {
1548                                 pr_err("%s: refused to change device type\n",
1549                                        bond_dev->name);
1550                                 res = -EBUSY;
1551                                 goto err_undo_flags;
1552                         }
1553
1554                         /* Flush unicast and multicast addresses */
1555                         dev_uc_flush(bond_dev);
1556                         dev_mc_flush(bond_dev);
1557
1558                         if (slave_dev->type != ARPHRD_ETHER)
1559                                 bond_setup_by_slave(bond_dev, slave_dev);
1560                         else {
1561                                 ether_setup(bond_dev);
1562                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1563                         }
1564
1565                         netdev_bonding_change(bond_dev,
1566                                               NETDEV_POST_TYPE_CHANGE);
1567                 }
1568         } else if (bond_dev->type != slave_dev->type) {
1569                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1570                        slave_dev->name,
1571                        slave_dev->type, bond_dev->type);
1572                 res = -EINVAL;
1573                 goto err_undo_flags;
1574         }
1575
1576         if (slave_ops->ndo_set_mac_address == NULL) {
1577                 if (bond->slave_cnt == 0) {
1578                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1579                                    bond_dev->name);
1580                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1581                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1582                         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",
1583                                bond_dev->name);
1584                         res = -EOPNOTSUPP;
1585                         goto err_undo_flags;
1586                 }
1587         }
1588
1589         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1590
1591         /* If this is the first slave, then we need to set the master's hardware
1592          * address to be the same as the slave's. */
1593         if (is_zero_ether_addr(bond->dev->dev_addr))
1594                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1595                        slave_dev->addr_len);
1596
1597
1598         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1599         if (!new_slave) {
1600                 res = -ENOMEM;
1601                 goto err_undo_flags;
1602         }
1603
1604         /*
1605          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1606          * is set via sysfs or module option if desired.
1607          */
1608         new_slave->queue_id = 0;
1609
1610         /* Save slave's original mtu and then set it to match the bond */
1611         new_slave->original_mtu = slave_dev->mtu;
1612         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1613         if (res) {
1614                 pr_debug("Error %d calling dev_set_mtu\n", res);
1615                 goto err_free;
1616         }
1617
1618         /*
1619          * Save slave's original ("permanent") mac address for modes
1620          * that need it, and for restoring it upon release, and then
1621          * set it to the master's address
1622          */
1623         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1624
1625         if (!bond->params.fail_over_mac) {
1626                 /*
1627                  * Set slave to master's mac address.  The application already
1628                  * set the master's mac address to that of the first slave
1629                  */
1630                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1631                 addr.sa_family = slave_dev->type;
1632                 res = dev_set_mac_address(slave_dev, &addr);
1633                 if (res) {
1634                         pr_debug("Error %d calling set_mac_address\n", res);
1635                         goto err_restore_mtu;
1636                 }
1637         }
1638
1639         res = netdev_set_bond_master(slave_dev, bond_dev);
1640         if (res) {
1641                 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1642                 goto err_restore_mac;
1643         }
1644
1645         /* open the slave since the application closed it */
1646         res = dev_open(slave_dev);
1647         if (res) {
1648                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1649                 goto err_unset_master;
1650         }
1651
1652         new_slave->bond = bond;
1653         new_slave->dev = slave_dev;
1654         slave_dev->priv_flags |= IFF_BONDING;
1655
1656         if (bond_is_lb(bond)) {
1657                 /* bond_alb_init_slave() must be called before all other stages since
1658                  * it might fail and we do not want to have to undo everything
1659                  */
1660                 res = bond_alb_init_slave(bond, new_slave);
1661                 if (res)
1662                         goto err_close;
1663         }
1664
1665         /* If the mode USES_PRIMARY, then the new slave gets the
1666          * master's promisc (and mc) settings only if it becomes the
1667          * curr_active_slave, and that is taken care of later when calling
1668          * bond_change_active()
1669          */
1670         if (!USES_PRIMARY(bond->params.mode)) {
1671                 /* set promiscuity level to new slave */
1672                 if (bond_dev->flags & IFF_PROMISC) {
1673                         res = dev_set_promiscuity(slave_dev, 1);
1674                         if (res)
1675                                 goto err_close;
1676                 }
1677
1678                 /* set allmulti level to new slave */
1679                 if (bond_dev->flags & IFF_ALLMULTI) {
1680                         res = dev_set_allmulti(slave_dev, 1);
1681                         if (res)
1682                                 goto err_close;
1683                 }
1684
1685                 netif_addr_lock_bh(bond_dev);
1686                 /* upload master's mc_list to new slave */
1687                 netdev_for_each_mc_addr(ha, bond_dev)
1688                         dev_mc_add(slave_dev, ha->addr);
1689                 netif_addr_unlock_bh(bond_dev);
1690         }
1691
1692         if (bond->params.mode == BOND_MODE_8023AD) {
1693                 /* add lacpdu mc addr to mc list */
1694                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1695
1696                 dev_mc_add(slave_dev, lacpdu_multicast);
1697         }
1698
1699         bond_add_vlans_on_slave(bond, slave_dev);
1700
1701         write_lock_bh(&bond->lock);
1702
1703         bond_attach_slave(bond, new_slave);
1704
1705         new_slave->delay = 0;
1706         new_slave->link_failure_count = 0;
1707
1708         write_unlock_bh(&bond->lock);
1709
1710         bond_compute_features(bond);
1711
1712         read_lock(&bond->lock);
1713
1714         new_slave->last_arp_rx = jiffies;
1715
1716         if (bond->params.miimon && !bond->params.use_carrier) {
1717                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1718
1719                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1720                         /*
1721                          * miimon is set but a bonded network driver
1722                          * does not support ETHTOOL/MII and
1723                          * arp_interval is not set.  Note: if
1724                          * use_carrier is enabled, we will never go
1725                          * here (because netif_carrier is always
1726                          * supported); thus, we don't need to change
1727                          * the messages for netif_carrier.
1728                          */
1729                         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",
1730                                bond_dev->name, slave_dev->name);
1731                 } else if (link_reporting == -1) {
1732                         /* unable get link status using mii/ethtool */
1733                         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",
1734                                    bond_dev->name, slave_dev->name);
1735                 }
1736         }
1737
1738         /* check for initial state */
1739         if (!bond->params.miimon ||
1740             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1741                 if (bond->params.updelay) {
1742                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1743                         new_slave->link  = BOND_LINK_BACK;
1744                         new_slave->delay = bond->params.updelay;
1745                 } else {
1746                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1747                         new_slave->link  = BOND_LINK_UP;
1748                 }
1749                 new_slave->jiffies = jiffies;
1750         } else {
1751                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1752                 new_slave->link  = BOND_LINK_DOWN;
1753         }
1754
1755         if (bond_update_speed_duplex(new_slave) &&
1756             (new_slave->link != BOND_LINK_DOWN)) {
1757                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1758                            bond_dev->name, new_slave->dev->name);
1759
1760                 if (bond->params.mode == BOND_MODE_8023AD) {
1761                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1762                                    bond_dev->name);
1763                 }
1764         }
1765
1766         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1767                 /* if there is a primary slave, remember it */
1768                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1769                         bond->primary_slave = new_slave;
1770                         bond->force_primary = true;
1771                 }
1772         }
1773
1774         write_lock_bh(&bond->curr_slave_lock);
1775
1776         switch (bond->params.mode) {
1777         case BOND_MODE_ACTIVEBACKUP:
1778                 bond_set_slave_inactive_flags(new_slave);
1779                 bond_select_active_slave(bond);
1780                 break;
1781         case BOND_MODE_8023AD:
1782                 /* in 802.3ad mode, the internal mechanism
1783                  * will activate the slaves in the selected
1784                  * aggregator
1785                  */
1786                 bond_set_slave_inactive_flags(new_slave);
1787                 /* if this is the first slave */
1788                 if (bond->slave_cnt == 1) {
1789                         SLAVE_AD_INFO(new_slave).id = 1;
1790                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1791                          * can be called only after the mac address of the bond is set
1792                          */
1793                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1794                 } else {
1795                         SLAVE_AD_INFO(new_slave).id =
1796                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1797                 }
1798
1799                 bond_3ad_bind_slave(new_slave);
1800                 break;
1801         case BOND_MODE_TLB:
1802         case BOND_MODE_ALB:
1803                 bond_set_active_slave(new_slave);
1804                 bond_set_slave_inactive_flags(new_slave);
1805                 bond_select_active_slave(bond);
1806                 break;
1807         default:
1808                 pr_debug("This slave is always active in trunk mode\n");
1809
1810                 /* always active in trunk mode */
1811                 bond_set_active_slave(new_slave);
1812
1813                 /* In trunking mode there is little meaning to curr_active_slave
1814                  * anyway (it holds no special properties of the bond device),
1815                  * so we can change it without calling change_active_interface()
1816                  */
1817                 if (!bond->curr_active_slave)
1818                         bond->curr_active_slave = new_slave;
1819
1820                 break;
1821         } /* switch(bond_mode) */
1822
1823         write_unlock_bh(&bond->curr_slave_lock);
1824
1825         bond_set_carrier(bond);
1826
1827 #ifdef CONFIG_NET_POLL_CONTROLLER
1828         slave_dev->npinfo = bond_netpoll_info(bond);
1829         if (slave_dev->npinfo) {
1830                 if (slave_enable_netpoll(new_slave)) {
1831                         read_unlock(&bond->lock);
1832                         pr_info("Error, %s: master_dev is using netpoll, "
1833                                  "but new slave device does not support netpoll.\n",
1834                                  bond_dev->name);
1835                         res = -EBUSY;
1836                         goto err_close;
1837                 }
1838         }
1839 #endif
1840
1841         read_unlock(&bond->lock);
1842
1843         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1844         if (res)
1845                 goto err_close;
1846
1847         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1848                                          new_slave);
1849         if (res) {
1850                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1851                 goto err_dest_symlinks;
1852         }
1853
1854         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1855                 bond_dev->name, slave_dev->name,
1856                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1857                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1858
1859         /* enslave is successful */
1860         return 0;
1861
1862 /* Undo stages on error */
1863 err_dest_symlinks:
1864         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1865
1866 err_close:
1867         dev_close(slave_dev);
1868
1869 err_unset_master:
1870         netdev_set_bond_master(slave_dev, NULL);
1871
1872 err_restore_mac:
1873         if (!bond->params.fail_over_mac) {
1874                 /* XXX TODO - fom follow mode needs to change master's
1875                  * MAC if this slave's MAC is in use by the bond, or at
1876                  * least print a warning.
1877                  */
1878                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1879                 addr.sa_family = slave_dev->type;
1880                 dev_set_mac_address(slave_dev, &addr);
1881         }
1882
1883 err_restore_mtu:
1884         dev_set_mtu(slave_dev, new_slave->original_mtu);
1885
1886 err_free:
1887         kfree(new_slave);
1888
1889 err_undo_flags:
1890         bond_compute_features(bond);
1891
1892         return res;
1893 }
1894
1895 /*
1896  * Try to release the slave device <slave> from the bond device <master>
1897  * It is legal to access curr_active_slave without a lock because all the function
1898  * is write-locked.
1899  *
1900  * The rules for slave state should be:
1901  *   for Active/Backup:
1902  *     Active stays on all backups go down
1903  *   for Bonded connections:
1904  *     The first up interface should be left on and all others downed.
1905  */
1906 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1907 {
1908         struct bonding *bond = netdev_priv(bond_dev);
1909         struct slave *slave, *oldcurrent;
1910         struct sockaddr addr;
1911         u32 old_features = bond_dev->features;
1912
1913         /* slave is not a slave or master is not master of this slave */
1914         if (!(slave_dev->flags & IFF_SLAVE) ||
1915             (slave_dev->master != bond_dev)) {
1916                 pr_err("%s: Error: cannot release %s.\n",
1917                        bond_dev->name, slave_dev->name);
1918                 return -EINVAL;
1919         }
1920
1921         block_netpoll_tx();
1922         netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1923         write_lock_bh(&bond->lock);
1924
1925         slave = bond_get_slave_by_dev(bond, slave_dev);
1926         if (!slave) {
1927                 /* not a slave of this bond */
1928                 pr_info("%s: %s not enslaved\n",
1929                         bond_dev->name, slave_dev->name);
1930                 write_unlock_bh(&bond->lock);
1931                 unblock_netpoll_tx();
1932                 return -EINVAL;
1933         }
1934
1935         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1936          * for this slave anymore.
1937          */
1938         netdev_rx_handler_unregister(slave_dev);
1939         write_unlock_bh(&bond->lock);
1940         synchronize_net();
1941         write_lock_bh(&bond->lock);
1942
1943         if (!bond->params.fail_over_mac) {
1944                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1945                     bond->slave_cnt > 1)
1946                         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",
1947                                    bond_dev->name, slave_dev->name,
1948                                    slave->perm_hwaddr,
1949                                    bond_dev->name, slave_dev->name);
1950         }
1951
1952         /* Inform AD package of unbinding of slave. */
1953         if (bond->params.mode == BOND_MODE_8023AD) {
1954                 /* must be called before the slave is
1955                  * detached from the list
1956                  */
1957                 bond_3ad_unbind_slave(slave);
1958         }
1959
1960         pr_info("%s: releasing %s interface %s\n",
1961                 bond_dev->name,
1962                 bond_is_active_slave(slave) ? "active" : "backup",
1963                 slave_dev->name);
1964
1965         oldcurrent = bond->curr_active_slave;
1966
1967         bond->current_arp_slave = NULL;
1968
1969         /* release the slave from its bond */
1970         bond_detach_slave(bond, slave);
1971
1972         if (bond->primary_slave == slave)
1973                 bond->primary_slave = NULL;
1974
1975         if (oldcurrent == slave)
1976                 bond_change_active_slave(bond, NULL);
1977
1978         if (bond_is_lb(bond)) {
1979                 /* Must be called only after the slave has been
1980                  * detached from the list and the curr_active_slave
1981                  * has been cleared (if our_slave == old_current),
1982                  * but before a new active slave is selected.
1983                  */
1984                 write_unlock_bh(&bond->lock);
1985                 bond_alb_deinit_slave(bond, slave);
1986                 write_lock_bh(&bond->lock);
1987         }
1988
1989         if (oldcurrent == slave) {
1990                 /*
1991                  * Note that we hold RTNL over this sequence, so there
1992                  * is no concern that another slave add/remove event
1993                  * will interfere.
1994                  */
1995                 write_unlock_bh(&bond->lock);
1996                 read_lock(&bond->lock);
1997                 write_lock_bh(&bond->curr_slave_lock);
1998
1999                 bond_select_active_slave(bond);
2000
2001                 write_unlock_bh(&bond->curr_slave_lock);
2002                 read_unlock(&bond->lock);
2003                 write_lock_bh(&bond->lock);
2004         }
2005
2006         if (bond->slave_cnt == 0) {
2007                 bond_set_carrier(bond);
2008
2009                 /* if the last slave was removed, zero the mac address
2010                  * of the master so it will be set by the application
2011                  * to the mac address of the first slave
2012                  */
2013                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2014
2015                 if (bond_vlan_used(bond)) {
2016                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2017                                    bond_dev->name, bond_dev->name);
2018                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2019                                    bond_dev->name);
2020                 }
2021         }
2022
2023         write_unlock_bh(&bond->lock);
2024         unblock_netpoll_tx();
2025
2026         bond_compute_features(bond);
2027         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2028             (old_features & NETIF_F_VLAN_CHALLENGED))
2029                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2030                         bond_dev->name, slave_dev->name, bond_dev->name);
2031
2032         /* must do this from outside any spinlocks */
2033         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2034
2035         bond_del_vlans_from_slave(bond, slave_dev);
2036
2037         /* If the mode USES_PRIMARY, then we should only remove its
2038          * promisc and mc settings if it was the curr_active_slave, but that was
2039          * already taken care of above when we detached the slave
2040          */
2041         if (!USES_PRIMARY(bond->params.mode)) {
2042                 /* unset promiscuity level from slave */
2043                 if (bond_dev->flags & IFF_PROMISC)
2044                         dev_set_promiscuity(slave_dev, -1);
2045
2046                 /* unset allmulti level from slave */
2047                 if (bond_dev->flags & IFF_ALLMULTI)
2048                         dev_set_allmulti(slave_dev, -1);
2049
2050                 /* flush master's mc_list from slave */
2051                 netif_addr_lock_bh(bond_dev);
2052                 bond_mc_list_flush(bond_dev, slave_dev);
2053                 netif_addr_unlock_bh(bond_dev);
2054         }
2055
2056         netdev_set_bond_master(slave_dev, NULL);
2057
2058         slave_disable_netpoll(slave);
2059
2060         /* close slave before restoring its mac address */
2061         dev_close(slave_dev);
2062
2063         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2064                 /* restore original ("permanent") mac address */
2065                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2066                 addr.sa_family = slave_dev->type;
2067                 dev_set_mac_address(slave_dev, &addr);
2068         }
2069
2070         dev_set_mtu(slave_dev, slave->original_mtu);
2071
2072         slave_dev->priv_flags &= ~IFF_BONDING;
2073
2074         kfree(slave);
2075
2076         return 0;  /* deletion OK */
2077 }
2078
2079 /*
2080 * First release a slave and then destroy the bond if no more slaves are left.
2081 * Must be under rtnl_lock when this function is called.
2082 */
2083 static int  bond_release_and_destroy(struct net_device *bond_dev,
2084                                      struct net_device *slave_dev)
2085 {
2086         struct bonding *bond = netdev_priv(bond_dev);
2087         int ret;
2088
2089         ret = bond_release(bond_dev, slave_dev);
2090         if ((ret == 0) && (bond->slave_cnt == 0)) {
2091                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2092                 pr_info("%s: destroying bond %s.\n",
2093                         bond_dev->name, bond_dev->name);
2094                 unregister_netdevice(bond_dev);
2095         }
2096         return ret;
2097 }
2098
2099 /*
2100  * This function releases all slaves.
2101  */
2102 static int bond_release_all(struct net_device *bond_dev)
2103 {
2104         struct bonding *bond = netdev_priv(bond_dev);
2105         struct slave *slave;
2106         struct net_device *slave_dev;
2107         struct sockaddr addr;
2108
2109         write_lock_bh(&bond->lock);
2110
2111         netif_carrier_off(bond_dev);
2112
2113         if (bond->slave_cnt == 0)
2114                 goto out;
2115
2116         bond->current_arp_slave = NULL;
2117         bond->primary_slave = NULL;
2118         bond_change_active_slave(bond, NULL);
2119
2120         while ((slave = bond->first_slave) != NULL) {
2121                 /* Inform AD package of unbinding of slave
2122                  * before slave is detached from the list.
2123                  */
2124                 if (bond->params.mode == BOND_MODE_8023AD)
2125                         bond_3ad_unbind_slave(slave);
2126
2127                 slave_dev = slave->dev;
2128                 bond_detach_slave(bond, slave);
2129
2130                 /* now that the slave is detached, unlock and perform
2131                  * all the undo steps that should not be called from
2132                  * within a lock.
2133                  */
2134                 write_unlock_bh(&bond->lock);
2135
2136                 /* unregister rx_handler early so bond_handle_frame wouldn't
2137                  * be called for this slave anymore.
2138                  */
2139                 netdev_rx_handler_unregister(slave_dev);
2140                 synchronize_net();
2141
2142                 if (bond_is_lb(bond)) {
2143                         /* must be called only after the slave
2144                          * has been detached from the list
2145                          */
2146                         bond_alb_deinit_slave(bond, slave);
2147                 }
2148
2149                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2150                 bond_del_vlans_from_slave(bond, slave_dev);
2151
2152                 /* If the mode USES_PRIMARY, then we should only remove its
2153                  * promisc and mc settings if it was the curr_active_slave, but that was
2154                  * already taken care of above when we detached the slave
2155                  */
2156                 if (!USES_PRIMARY(bond->params.mode)) {
2157                         /* unset promiscuity level from slave */
2158                         if (bond_dev->flags & IFF_PROMISC)
2159                                 dev_set_promiscuity(slave_dev, -1);
2160
2161                         /* unset allmulti level from slave */
2162                         if (bond_dev->flags & IFF_ALLMULTI)
2163                                 dev_set_allmulti(slave_dev, -1);
2164
2165                         /* flush master's mc_list from slave */
2166                         netif_addr_lock_bh(bond_dev);
2167                         bond_mc_list_flush(bond_dev, slave_dev);
2168                         netif_addr_unlock_bh(bond_dev);
2169                 }
2170
2171                 netdev_set_bond_master(slave_dev, NULL);
2172
2173                 slave_disable_netpoll(slave);
2174
2175                 /* close slave before restoring its mac address */
2176                 dev_close(slave_dev);
2177
2178                 if (!bond->params.fail_over_mac) {
2179                         /* restore original ("permanent") mac address*/
2180                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2181                         addr.sa_family = slave_dev->type;
2182                         dev_set_mac_address(slave_dev, &addr);
2183                 }
2184
2185                 kfree(slave);
2186
2187                 /* re-acquire the lock before getting the next slave */
2188                 write_lock_bh(&bond->lock);
2189         }
2190
2191         /* zero the mac address of the master so it will be
2192          * set by the application to the mac address of the
2193          * first slave
2194          */
2195         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2196
2197         if (bond_vlan_used(bond)) {
2198                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2199                            bond_dev->name, bond_dev->name);
2200                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2201                            bond_dev->name);
2202         }
2203
2204         pr_info("%s: released all slaves\n", bond_dev->name);
2205
2206 out:
2207         write_unlock_bh(&bond->lock);
2208
2209         bond_compute_features(bond);
2210
2211         return 0;
2212 }
2213
2214 /*
2215  * This function changes the active slave to slave <slave_dev>.
2216  * It returns -EINVAL in the following cases.
2217  *  - <slave_dev> is not found in the list.
2218  *  - There is not active slave now.
2219  *  - <slave_dev> is already active.
2220  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2221  *  - <slave_dev> is not running.
2222  * In these cases, this function does nothing.
2223  * In the other cases, current_slave pointer is changed and 0 is returned.
2224  */
2225 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2226 {
2227         struct bonding *bond = netdev_priv(bond_dev);
2228         struct slave *old_active = NULL;
2229         struct slave *new_active = NULL;
2230         int res = 0;
2231
2232         if (!USES_PRIMARY(bond->params.mode))
2233                 return -EINVAL;
2234
2235         /* Verify that master_dev is indeed the master of slave_dev */
2236         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2237                 return -EINVAL;
2238
2239         read_lock(&bond->lock);
2240
2241         read_lock(&bond->curr_slave_lock);
2242         old_active = bond->curr_active_slave;
2243         read_unlock(&bond->curr_slave_lock);
2244
2245         new_active = bond_get_slave_by_dev(bond, slave_dev);
2246
2247         /*
2248          * Changing to the current active: do nothing; return success.
2249          */
2250         if (new_active && (new_active == old_active)) {
2251                 read_unlock(&bond->lock);
2252                 return 0;
2253         }
2254
2255         if ((new_active) &&
2256             (old_active) &&
2257             (new_active->link == BOND_LINK_UP) &&
2258             IS_UP(new_active->dev)) {
2259                 block_netpoll_tx();
2260                 write_lock_bh(&bond->curr_slave_lock);
2261                 bond_change_active_slave(bond, new_active);
2262                 write_unlock_bh(&bond->curr_slave_lock);
2263                 unblock_netpoll_tx();
2264         } else
2265                 res = -EINVAL;
2266
2267         read_unlock(&bond->lock);
2268
2269         return res;
2270 }
2271
2272 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2273 {
2274         struct bonding *bond = netdev_priv(bond_dev);
2275
2276         info->bond_mode = bond->params.mode;
2277         info->miimon = bond->params.miimon;
2278
2279         read_lock(&bond->lock);
2280         info->num_slaves = bond->slave_cnt;
2281         read_unlock(&bond->lock);
2282
2283         return 0;
2284 }
2285
2286 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2287 {
2288         struct bonding *bond = netdev_priv(bond_dev);
2289         struct slave *slave;
2290         int i, res = -ENODEV;
2291
2292         read_lock(&bond->lock);
2293
2294         bond_for_each_slave(bond, slave, i) {
2295                 if (i == (int)info->slave_id) {
2296                         res = 0;
2297                         strcpy(info->slave_name, slave->dev->name);
2298                         info->link = slave->link;
2299                         info->state = bond_slave_state(slave);
2300                         info->link_failure_count = slave->link_failure_count;
2301                         break;
2302                 }
2303         }
2304
2305         read_unlock(&bond->lock);
2306
2307         return res;
2308 }
2309
2310 /*-------------------------------- Monitoring -------------------------------*/
2311
2312
2313 static int bond_miimon_inspect(struct bonding *bond)
2314 {
2315         struct slave *slave;
2316         int i, link_state, commit = 0;
2317         bool ignore_updelay;
2318
2319         ignore_updelay = !bond->curr_active_slave ? true : false;
2320
2321         bond_for_each_slave(bond, slave, i) {
2322                 slave->new_link = BOND_LINK_NOCHANGE;
2323
2324                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2325
2326                 switch (slave->link) {
2327                 case BOND_LINK_UP:
2328                         if (link_state)
2329                                 continue;
2330
2331                         slave->link = BOND_LINK_FAIL;
2332                         slave->delay = bond->params.downdelay;
2333                         if (slave->delay) {
2334                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2335                                         bond->dev->name,
2336                                         (bond->params.mode ==
2337                                          BOND_MODE_ACTIVEBACKUP) ?
2338                                         (bond_is_active_slave(slave) ?
2339                                          "active " : "backup ") : "",
2340                                         slave->dev->name,
2341                                         bond->params.downdelay * bond->params.miimon);
2342                         }
2343                         /*FALLTHRU*/
2344                 case BOND_LINK_FAIL:
2345                         if (link_state) {
2346                                 /*
2347                                  * recovered before downdelay expired
2348                                  */
2349                                 slave->link = BOND_LINK_UP;
2350                                 slave->jiffies = jiffies;
2351                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2352                                         bond->dev->name,
2353                                         (bond->params.downdelay - slave->delay) *
2354                                         bond->params.miimon,
2355                                         slave->dev->name);
2356                                 continue;
2357                         }
2358
2359                         if (slave->delay <= 0) {
2360                                 slave->new_link = BOND_LINK_DOWN;
2361                                 commit++;
2362                                 continue;
2363                         }
2364
2365                         slave->delay--;
2366                         break;
2367
2368                 case BOND_LINK_DOWN:
2369                         if (!link_state)
2370                                 continue;
2371
2372                         slave->link = BOND_LINK_BACK;
2373                         slave->delay = bond->params.updelay;
2374
2375                         if (slave->delay) {
2376                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2377                                         bond->dev->name, slave->dev->name,
2378                                         ignore_updelay ? 0 :
2379                                         bond->params.updelay *
2380                                         bond->params.miimon);
2381                         }
2382                         /*FALLTHRU*/
2383                 case BOND_LINK_BACK:
2384                         if (!link_state) {
2385                                 slave->link = BOND_LINK_DOWN;
2386                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2387                                         bond->dev->name,
2388                                         (bond->params.updelay - slave->delay) *
2389                                         bond->params.miimon,
2390                                         slave->dev->name);
2391
2392                                 continue;
2393                         }
2394
2395                         if (ignore_updelay)
2396                                 slave->delay = 0;
2397
2398                         if (slave->delay <= 0) {
2399                                 slave->new_link = BOND_LINK_UP;
2400                                 commit++;
2401                                 ignore_updelay = false;
2402                                 continue;
2403                         }
2404
2405                         slave->delay--;
2406                         break;
2407                 }
2408         }
2409
2410         return commit;
2411 }
2412
2413 static void bond_miimon_commit(struct bonding *bond)
2414 {
2415         struct slave *slave;
2416         int i;
2417
2418         bond_for_each_slave(bond, slave, i) {
2419                 switch (slave->new_link) {
2420                 case BOND_LINK_NOCHANGE:
2421                         continue;
2422
2423                 case BOND_LINK_UP:
2424                         slave->link = BOND_LINK_UP;
2425                         slave->jiffies = jiffies;
2426
2427                         if (bond->params.mode == BOND_MODE_8023AD) {
2428                                 /* prevent it from being the active one */
2429                                 bond_set_backup_slave(slave);
2430                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2431                                 /* make it immediately active */
2432                                 bond_set_active_slave(slave);
2433                         } else if (slave != bond->primary_slave) {
2434                                 /* prevent it from being the active one */
2435                                 bond_set_backup_slave(slave);
2436                         }
2437
2438                         bond_update_speed_duplex(slave);
2439
2440                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2441                                 bond->dev->name, slave->dev->name,
2442                                 slave->speed, slave->duplex ? "full" : "half");
2443
2444                         /* notify ad that the link status has changed */
2445                         if (bond->params.mode == BOND_MODE_8023AD)
2446                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2447
2448                         if (bond_is_lb(bond))
2449                                 bond_alb_handle_link_change(bond, slave,
2450                                                             BOND_LINK_UP);
2451
2452                         if (!bond->curr_active_slave ||
2453                             (slave == bond->primary_slave))
2454                                 goto do_failover;
2455
2456                         continue;
2457
2458                 case BOND_LINK_DOWN:
2459                         if (slave->link_failure_count < UINT_MAX)
2460                                 slave->link_failure_count++;
2461
2462                         slave->link = BOND_LINK_DOWN;
2463
2464                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2465                             bond->params.mode == BOND_MODE_8023AD)
2466                                 bond_set_slave_inactive_flags(slave);
2467
2468                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2469                                 bond->dev->name, slave->dev->name);
2470
2471                         if (bond->params.mode == BOND_MODE_8023AD)
2472                                 bond_3ad_handle_link_change(slave,
2473                                                             BOND_LINK_DOWN);
2474
2475                         if (bond_is_lb(bond))
2476                                 bond_alb_handle_link_change(bond, slave,
2477                                                             BOND_LINK_DOWN);
2478
2479                         if (slave == bond->curr_active_slave)
2480                                 goto do_failover;
2481
2482                         continue;
2483
2484                 default:
2485                         pr_err("%s: invalid new link %d on slave %s\n",
2486                                bond->dev->name, slave->new_link,
2487                                slave->dev->name);
2488                         slave->new_link = BOND_LINK_NOCHANGE;
2489
2490                         continue;
2491                 }
2492
2493 do_failover:
2494                 ASSERT_RTNL();
2495                 block_netpoll_tx();
2496                 write_lock_bh(&bond->curr_slave_lock);
2497                 bond_select_active_slave(bond);
2498                 write_unlock_bh(&bond->curr_slave_lock);
2499                 unblock_netpoll_tx();
2500         }
2501
2502         bond_set_carrier(bond);
2503 }
2504
2505 /*
2506  * bond_mii_monitor
2507  *
2508  * Really a wrapper that splits the mii monitor into two phases: an
2509  * inspection, then (if inspection indicates something needs to be done)
2510  * an acquisition of appropriate locks followed by a commit phase to
2511  * implement whatever link state changes are indicated.
2512  */
2513 void bond_mii_monitor(struct work_struct *work)
2514 {
2515         struct bonding *bond = container_of(work, struct bonding,
2516                                             mii_work.work);
2517         bool should_notify_peers = false;
2518
2519         read_lock(&bond->lock);
2520         if (bond->kill_timers)
2521                 goto out;
2522
2523         if (bond->slave_cnt == 0)
2524                 goto re_arm;
2525
2526         should_notify_peers = bond_should_notify_peers(bond);
2527
2528         if (bond_miimon_inspect(bond)) {
2529                 read_unlock(&bond->lock);
2530                 rtnl_lock();
2531                 read_lock(&bond->lock);
2532
2533                 bond_miimon_commit(bond);
2534
2535                 read_unlock(&bond->lock);
2536                 rtnl_unlock();  /* might sleep, hold no other locks */
2537                 read_lock(&bond->lock);
2538         }
2539
2540 re_arm:
2541         if (bond->params.miimon && !bond->kill_timers)
2542                 queue_delayed_work(bond->wq, &bond->mii_work,
2543                                    msecs_to_jiffies(bond->params.miimon));
2544 out:
2545         read_unlock(&bond->lock);
2546
2547         if (should_notify_peers) {
2548                 rtnl_lock();
2549                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2550                 rtnl_unlock();
2551         }
2552 }
2553
2554 static __be32 bond_glean_dev_ip(struct net_device *dev)
2555 {
2556         struct in_device *idev;
2557         struct in_ifaddr *ifa;
2558         __be32 addr = 0;
2559
2560         if (!dev)
2561                 return 0;
2562
2563         rcu_read_lock();
2564         idev = __in_dev_get_rcu(dev);
2565         if (!idev)
2566                 goto out;
2567
2568         ifa = idev->ifa_list;
2569         if (!ifa)
2570                 goto out;
2571
2572         addr = ifa->ifa_local;
2573 out:
2574         rcu_read_unlock();
2575         return addr;
2576 }
2577
2578 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2579 {
2580         struct vlan_entry *vlan;
2581
2582         if (ip == bond->master_ip)
2583                 return 1;
2584
2585         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2586                 if (ip == vlan->vlan_ip)
2587                         return 1;
2588         }
2589
2590         return 0;
2591 }
2592
2593 /*
2594  * We go to the (large) trouble of VLAN tagging ARP frames because
2595  * switches in VLAN mode (especially if ports are configured as
2596  * "native" to a VLAN) might not pass non-tagged frames.
2597  */
2598 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2599 {
2600         struct sk_buff *skb;
2601
2602         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2603                  slave_dev->name, dest_ip, src_ip, vlan_id);
2604
2605         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2606                          NULL, slave_dev->dev_addr, NULL);
2607
2608         if (!skb) {
2609                 pr_err("ARP packet allocation failed\n");
2610                 return;
2611         }
2612         if (vlan_id) {
2613                 skb = vlan_put_tag(skb, vlan_id);
2614                 if (!skb) {
2615                         pr_err("failed to insert VLAN tag\n");
2616                         return;
2617                 }
2618         }
2619         arp_xmit(skb);
2620 }
2621
2622
2623 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2624 {
2625         int i, vlan_id;
2626         __be32 *targets = bond->params.arp_targets;
2627         struct vlan_entry *vlan;
2628         struct net_device *vlan_dev;
2629         struct rtable *rt;
2630
2631         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2632                 if (!targets[i])
2633                         break;
2634                 pr_debug("basa: target %x\n", targets[i]);
2635                 if (!bond_vlan_used(bond)) {
2636                         pr_debug("basa: empty vlan: arp_send\n");
2637                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2638                                       bond->master_ip, 0);
2639                         continue;
2640                 }
2641
2642                 /*
2643                  * If VLANs are configured, we do a route lookup to
2644                  * determine which VLAN interface would be used, so we
2645                  * can tag the ARP with the proper VLAN tag.
2646                  */
2647                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2648                                      RTO_ONLINK, 0);
2649                 if (IS_ERR(rt)) {
2650                         if (net_ratelimit()) {
2651                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2652                                            bond->dev->name, &targets[i]);
2653                         }
2654                         continue;
2655                 }
2656
2657                 /*
2658                  * This target is not on a VLAN
2659                  */
2660                 if (rt->dst.dev == bond->dev) {
2661                         ip_rt_put(rt);
2662                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2663                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2664                                       bond->master_ip, 0);
2665                         continue;
2666                 }
2667
2668                 vlan_id = 0;
2669                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2670                         rcu_read_lock();
2671                         vlan_dev = __vlan_find_dev_deep(bond->dev,
2672                                                         vlan->vlan_id);
2673                         rcu_read_unlock();
2674                         if (vlan_dev == rt->dst.dev) {
2675                                 vlan_id = vlan->vlan_id;
2676                                 pr_debug("basa: vlan match on %s %d\n",
2677                                        vlan_dev->name, vlan_id);
2678                                 break;
2679                         }
2680                 }
2681
2682                 if (vlan_id) {
2683                         ip_rt_put(rt);
2684                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2685                                       vlan->vlan_ip, vlan_id);
2686                         continue;
2687                 }
2688
2689                 if (net_ratelimit()) {
2690                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2691                                    bond->dev->name, &targets[i],
2692                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2693                 }
2694                 ip_rt_put(rt);
2695         }
2696 }
2697
2698 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2699 {
2700         int i;
2701         __be32 *targets = bond->params.arp_targets;
2702
2703         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2704                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2705                          &sip, &tip, i, &targets[i],
2706                          bond_has_this_ip(bond, tip));
2707                 if (sip == targets[i]) {
2708                         if (bond_has_this_ip(bond, tip))
2709                                 slave->last_arp_rx = jiffies;
2710                         return;
2711                 }
2712         }
2713 }
2714
2715 static void bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2716                          struct slave *slave)
2717 {
2718         struct arphdr *arp;
2719         unsigned char *arp_ptr;
2720         __be32 sip, tip;
2721
2722         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2723                 return;
2724
2725         read_lock(&bond->lock);
2726
2727         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2728                  bond->dev->name, skb->dev->name);
2729
2730         if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
2731                 goto out_unlock;
2732
2733         arp = arp_hdr(skb);
2734         if (arp->ar_hln != bond->dev->addr_len ||
2735             skb->pkt_type == PACKET_OTHERHOST ||
2736             skb->pkt_type == PACKET_LOOPBACK ||
2737             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2738             arp->ar_pro != htons(ETH_P_IP) ||
2739             arp->ar_pln != 4)
2740                 goto out_unlock;
2741
2742         arp_ptr = (unsigned char *)(arp + 1);
2743         arp_ptr += bond->dev->addr_len;
2744         memcpy(&sip, arp_ptr, 4);
2745         arp_ptr += 4 + bond->dev->addr_len;
2746         memcpy(&tip, arp_ptr, 4);
2747
2748         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2749                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2750                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2751                  &sip, &tip);
2752
2753         /*
2754          * Backup slaves won't see the ARP reply, but do come through
2755          * here for each ARP probe (so we swap the sip/tip to validate
2756          * the probe).  In a "redundant switch, common router" type of
2757          * configuration, the ARP probe will (hopefully) travel from
2758          * the active, through one switch, the router, then the other
2759          * switch before reaching the backup.
2760          */
2761         if (bond_is_active_slave(slave))
2762                 bond_validate_arp(bond, slave, sip, tip);
2763         else
2764                 bond_validate_arp(bond, slave, tip, sip);
2765
2766 out_unlock:
2767         read_unlock(&bond->lock);
2768 }
2769
2770 /*
2771  * this function is called regularly to monitor each slave's link
2772  * ensuring that traffic is being sent and received when arp monitoring
2773  * is used in load-balancing mode. if the adapter has been dormant, then an
2774  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2775  * arp monitoring in active backup mode.
2776  */
2777 void bond_loadbalance_arp_mon(struct work_struct *work)
2778 {
2779         struct bonding *bond = container_of(work, struct bonding,
2780                                             arp_work.work);
2781         struct slave *slave, *oldcurrent;
2782         int do_failover = 0;
2783         int delta_in_ticks;
2784         int i;
2785
2786         read_lock(&bond->lock);
2787
2788         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2789
2790         if (bond->kill_timers)
2791                 goto out;
2792
2793         if (bond->slave_cnt == 0)
2794                 goto re_arm;
2795
2796         read_lock(&bond->curr_slave_lock);
2797         oldcurrent = bond->curr_active_slave;
2798         read_unlock(&bond->curr_slave_lock);
2799
2800         /* see if any of the previous devices are up now (i.e. they have
2801          * xmt and rcv traffic). the curr_active_slave does not come into
2802          * the picture unless it is null. also, slave->jiffies is not needed
2803          * here because we send an arp on each slave and give a slave as
2804          * long as it needs to get the tx/rx within the delta.
2805          * TODO: what about up/down delay in arp mode? it wasn't here before
2806          *       so it can wait
2807          */
2808         bond_for_each_slave(bond, slave, i) {
2809                 unsigned long trans_start = dev_trans_start(slave->dev);
2810
2811                 if (slave->link != BOND_LINK_UP) {
2812                         if (time_in_range(jiffies,
2813                                 trans_start - delta_in_ticks,
2814                                 trans_start + delta_in_ticks) &&
2815                             time_in_range(jiffies,
2816                                 slave->dev->last_rx - delta_in_ticks,
2817                                 slave->dev->last_rx + delta_in_ticks)) {
2818
2819                                 slave->link  = BOND_LINK_UP;
2820                                 bond_set_active_slave(slave);
2821
2822                                 /* primary_slave has no meaning in round-robin
2823                                  * mode. the window of a slave being up and
2824                                  * curr_active_slave being null after enslaving
2825                                  * is closed.
2826                                  */
2827                                 if (!oldcurrent) {
2828                                         pr_info("%s: link status definitely up for interface %s, ",
2829                                                 bond->dev->name,
2830                                                 slave->dev->name);
2831                                         do_failover = 1;
2832                                 } else {
2833                                         pr_info("%s: interface %s is now up\n",
2834                                                 bond->dev->name,
2835                                                 slave->dev->name);
2836                                 }
2837                         }
2838                 } else {
2839                         /* slave->link == BOND_LINK_UP */
2840
2841                         /* not all switches will respond to an arp request
2842                          * when the source ip is 0, so don't take the link down
2843                          * if we don't know our ip yet
2844                          */
2845                         if (!time_in_range(jiffies,
2846                                 trans_start - delta_in_ticks,
2847                                 trans_start + 2 * delta_in_ticks) ||
2848                             !time_in_range(jiffies,
2849                                 slave->dev->last_rx - delta_in_ticks,
2850                                 slave->dev->last_rx + 2 * delta_in_ticks)) {
2851
2852                                 slave->link  = BOND_LINK_DOWN;
2853                                 bond_set_backup_slave(slave);
2854
2855                                 if (slave->link_failure_count < UINT_MAX)
2856                                         slave->link_failure_count++;
2857
2858                                 pr_info("%s: interface %s is now down.\n",
2859                                         bond->dev->name,
2860                                         slave->dev->name);
2861
2862                                 if (slave == oldcurrent)
2863                                         do_failover = 1;
2864                         }
2865                 }
2866
2867                 /* note: if switch is in round-robin mode, all links
2868                  * must tx arp to ensure all links rx an arp - otherwise
2869                  * links may oscillate or not come up at all; if switch is
2870                  * in something like xor mode, there is nothing we can
2871                  * do - all replies will be rx'ed on same link causing slaves
2872                  * to be unstable during low/no traffic periods
2873                  */
2874                 if (IS_UP(slave->dev))
2875                         bond_arp_send_all(bond, slave);
2876         }
2877
2878         if (do_failover) {
2879                 block_netpoll_tx();
2880                 write_lock_bh(&bond->curr_slave_lock);
2881
2882                 bond_select_active_slave(bond);
2883
2884                 write_unlock_bh(&bond->curr_slave_lock);
2885                 unblock_netpoll_tx();
2886         }
2887
2888 re_arm:
2889         if (bond->params.arp_interval && !bond->kill_timers)
2890                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2891 out:
2892         read_unlock(&bond->lock);
2893 }
2894
2895 /*
2896  * Called to inspect slaves for active-backup mode ARP monitor link state
2897  * changes.  Sets new_link in slaves to specify what action should take
2898  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2899  * to link states must be committed.
2900  *
2901  * Called with bond->lock held for read.
2902  */
2903 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2904 {
2905         struct slave *slave;
2906         int i, commit = 0;
2907         unsigned long trans_start;
2908
2909         bond_for_each_slave(bond, slave, i) {
2910                 slave->new_link = BOND_LINK_NOCHANGE;
2911
2912                 if (slave->link != BOND_LINK_UP) {
2913                         if (time_in_range(jiffies,
2914                                 slave_last_rx(bond, slave) - delta_in_ticks,
2915                                 slave_last_rx(bond, slave) + delta_in_ticks)) {
2916
2917                                 slave->new_link = BOND_LINK_UP;
2918                                 commit++;
2919                         }
2920
2921                         continue;
2922                 }
2923
2924                 /*
2925                  * Give slaves 2*delta after being enslaved or made
2926                  * active.  This avoids bouncing, as the last receive
2927                  * times need a full ARP monitor cycle to be updated.
2928                  */
2929                 if (time_in_range(jiffies,
2930                                   slave->jiffies - delta_in_ticks,
2931                                   slave->jiffies + 2 * delta_in_ticks))
2932                         continue;
2933
2934                 /*
2935                  * Backup slave is down if:
2936                  * - No current_arp_slave AND
2937                  * - more than 3*delta since last receive AND
2938                  * - the bond has an IP address
2939                  *
2940                  * Note: a non-null current_arp_slave indicates
2941                  * the curr_active_slave went down and we are
2942                  * searching for a new one; under this condition
2943                  * we only take the curr_active_slave down - this
2944                  * gives each slave a chance to tx/rx traffic
2945                  * before being taken out
2946                  */
2947                 if (!bond_is_active_slave(slave) &&
2948                     !bond->current_arp_slave &&
2949                     !time_in_range(jiffies,
2950                         slave_last_rx(bond, slave) - delta_in_ticks,
2951                         slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2952
2953                         slave->new_link = BOND_LINK_DOWN;
2954                         commit++;
2955                 }
2956
2957                 /*
2958                  * Active slave is down if:
2959                  * - more than 2*delta since transmitting OR
2960                  * - (more than 2*delta since receive AND
2961                  *    the bond has an IP address)
2962                  */
2963                 trans_start = dev_trans_start(slave->dev);
2964                 if (bond_is_active_slave(slave) &&
2965                     (!time_in_range(jiffies,
2966                         trans_start - delta_in_ticks,
2967                         trans_start + 2 * delta_in_ticks) ||
2968                      !time_in_range(jiffies,
2969                         slave_last_rx(bond, slave) - delta_in_ticks,
2970                         slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2971
2972                         slave->new_link = BOND_LINK_DOWN;
2973                         commit++;
2974                 }
2975         }
2976
2977         return commit;
2978 }
2979
2980 /*
2981  * Called to commit link state changes noted by inspection step of
2982  * active-backup mode ARP monitor.
2983  *
2984  * Called with RTNL and bond->lock for read.
2985  */
2986 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2987 {
2988         struct slave *slave;
2989         int i;
2990         unsigned long trans_start;
2991
2992         bond_for_each_slave(bond, slave, i) {
2993                 switch (slave->new_link) {
2994                 case BOND_LINK_NOCHANGE:
2995                         continue;
2996
2997                 case BOND_LINK_UP:
2998                         trans_start = dev_trans_start(slave->dev);
2999                         if ((!bond->curr_active_slave &&
3000                              time_in_range(jiffies,
3001                                            trans_start - delta_in_ticks,
3002                                            trans_start + delta_in_ticks)) ||
3003                             bond->curr_active_slave != slave) {
3004                                 slave->link = BOND_LINK_UP;
3005                                 bond->current_arp_slave = NULL;
3006
3007                                 pr_info("%s: link status definitely up for interface %s.\n",
3008                                         bond->dev->name, slave->dev->name);
3009
3010                                 if (!bond->curr_active_slave ||
3011                                     (slave == bond->primary_slave))
3012                                         goto do_failover;
3013
3014                         }
3015
3016                         continue;
3017
3018                 case BOND_LINK_DOWN:
3019                         if (slave->link_failure_count < UINT_MAX)
3020                                 slave->link_failure_count++;
3021
3022                         slave->link = BOND_LINK_DOWN;
3023                         bond_set_slave_inactive_flags(slave);
3024
3025                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3026                                 bond->dev->name, slave->dev->name);
3027
3028                         if (slave == bond->curr_active_slave) {
3029                                 bond->current_arp_slave = NULL;
3030                                 goto do_failover;
3031                         }
3032
3033                         continue;
3034
3035                 default:
3036                         pr_err("%s: impossible: new_link %d on slave %s\n",
3037                                bond->dev->name, slave->new_link,
3038                                slave->dev->name);
3039                         continue;
3040                 }
3041
3042 do_failover:
3043                 ASSERT_RTNL();
3044                 block_netpoll_tx();
3045                 write_lock_bh(&bond->curr_slave_lock);
3046                 bond_select_active_slave(bond);
3047                 write_unlock_bh(&bond->curr_slave_lock);
3048                 unblock_netpoll_tx();
3049         }
3050
3051         bond_set_carrier(bond);
3052 }
3053
3054 /*
3055  * Send ARP probes for active-backup mode ARP monitor.
3056  *
3057  * Called with bond->lock held for read.
3058  */
3059 static void bond_ab_arp_probe(struct bonding *bond)
3060 {
3061         struct slave *slave;
3062         int i;
3063
3064         read_lock(&bond->curr_slave_lock);
3065
3066         if (bond->current_arp_slave && bond->curr_active_slave)
3067                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3068                         bond->current_arp_slave->dev->name,
3069                         bond->curr_active_slave->dev->name);
3070
3071         if (bond->curr_active_slave) {
3072                 bond_arp_send_all(bond, bond->curr_active_slave);
3073                 read_unlock(&bond->curr_slave_lock);
3074                 return;
3075         }
3076
3077         read_unlock(&bond->curr_slave_lock);
3078
3079         /* if we don't have a curr_active_slave, search for the next available
3080          * backup slave from the current_arp_slave and make it the candidate
3081          * for becoming the curr_active_slave
3082          */
3083
3084         if (!bond->current_arp_slave) {
3085                 bond->current_arp_slave = bond->first_slave;
3086                 if (!bond->current_arp_slave)
3087                         return;
3088         }
3089
3090         bond_set_slave_inactive_flags(bond->current_arp_slave);
3091
3092         /* search for next candidate */
3093         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3094                 if (IS_UP(slave->dev)) {
3095                         slave->link = BOND_LINK_BACK;
3096                         bond_set_slave_active_flags(slave);
3097                         bond_arp_send_all(bond, slave);
3098                         slave->jiffies = jiffies;
3099                         bond->current_arp_slave = slave;
3100                         break;
3101                 }
3102
3103                 /* if the link state is up at this point, we
3104                  * mark it down - this can happen if we have
3105                  * simultaneous link failures and
3106                  * reselect_active_interface doesn't make this
3107                  * one the current slave so it is still marked
3108                  * up when it is actually down
3109                  */
3110                 if (slave->link == BOND_LINK_UP) {
3111                         slave->link = BOND_LINK_DOWN;
3112                         if (slave->link_failure_count < UINT_MAX)
3113                                 slave->link_failure_count++;
3114
3115                         bond_set_slave_inactive_flags(slave);
3116
3117                         pr_info("%s: backup interface %s is now down.\n",
3118                                 bond->dev->name, slave->dev->name);
3119                 }
3120         }
3121 }
3122
3123 void bond_activebackup_arp_mon(struct work_struct *work)
3124 {
3125         struct bonding *bond = container_of(work, struct bonding,
3126                                             arp_work.work);
3127         bool should_notify_peers = false;
3128         int delta_in_ticks;
3129
3130         read_lock(&bond->lock);
3131
3132         if (bond->kill_timers)
3133                 goto out;
3134
3135         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3136
3137         if (bond->slave_cnt == 0)
3138                 goto re_arm;
3139
3140         should_notify_peers = bond_should_notify_peers(bond);
3141
3142         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3143                 read_unlock(&bond->lock);
3144                 rtnl_lock();
3145                 read_lock(&bond->lock);
3146
3147                 bond_ab_arp_commit(bond, delta_in_ticks);
3148
3149                 read_unlock(&bond->lock);
3150                 rtnl_unlock();
3151                 read_lock(&bond->lock);
3152         }
3153
3154         bond_ab_arp_probe(bond);
3155
3156 re_arm:
3157         if (bond->params.arp_interval && !bond->kill_timers)
3158                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3159 out:
3160         read_unlock(&bond->lock);
3161
3162         if (should_notify_peers) {
3163                 rtnl_lock();
3164                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3165                 rtnl_unlock();
3166         }
3167 }
3168
3169 /*-------------------------- netdev event handling --------------------------*/
3170
3171 /*
3172  * Change device name
3173  */
3174 static int bond_event_changename(struct bonding *bond)
3175 {
3176         bond_remove_proc_entry(bond);
3177         bond_create_proc_entry(bond);
3178
3179         bond_debug_reregister(bond);
3180
3181         return NOTIFY_DONE;
3182 }
3183
3184 static int bond_master_netdev_event(unsigned long event,
3185                                     struct net_device *bond_dev)
3186 {
3187         struct bonding *event_bond = netdev_priv(bond_dev);
3188
3189         switch (event) {
3190         case NETDEV_CHANGENAME:
3191                 return bond_event_changename(event_bond);
3192         default:
3193                 break;
3194         }
3195
3196         return NOTIFY_DONE;
3197 }
3198
3199 static int bond_slave_netdev_event(unsigned long event,
3200                                    struct net_device *slave_dev)
3201 {
3202         struct net_device *bond_dev = slave_dev->master;
3203         struct bonding *bond = netdev_priv(bond_dev);
3204
3205         switch (event) {
3206         case NETDEV_UNREGISTER:
3207                 if (bond_dev) {
3208                         if (bond->setup_by_slave)
3209                                 bond_release_and_destroy(bond_dev, slave_dev);
3210                         else
3211                                 bond_release(bond_dev, slave_dev);
3212                 }
3213                 break;
3214         case NETDEV_CHANGE:
3215                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3216                         struct slave *slave;
3217
3218                         slave = bond_get_slave_by_dev(bond, slave_dev);
3219                         if (slave) {
3220                                 u32 old_speed = slave->speed;
3221                                 u8  old_duplex = slave->duplex;
3222
3223                                 bond_update_speed_duplex(slave);
3224
3225                                 if (bond_is_lb(bond))
3226                                         break;
3227
3228                                 if (old_speed != slave->speed)
3229                                         bond_3ad_adapter_speed_changed(slave);
3230                                 if (old_duplex != slave->duplex)
3231                                         bond_3ad_adapter_duplex_changed(slave);
3232                         }
3233                 }
3234
3235                 break;
3236         case NETDEV_DOWN:
3237                 /*
3238                  * ... Or is it this?
3239                  */
3240                 break;
3241         case NETDEV_CHANGEMTU:
3242                 /*
3243                  * TODO: Should slaves be allowed to
3244                  * independently alter their MTU?  For
3245                  * an active-backup bond, slaves need
3246                  * not be the same type of device, so
3247                  * MTUs may vary.  For other modes,
3248                  * slaves arguably should have the
3249                  * same MTUs. To do this, we'd need to
3250                  * take over the slave's change_mtu
3251                  * function for the duration of their
3252                  * servitude.
3253                  */
3254                 break;
3255         case NETDEV_CHANGENAME:
3256                 /*
3257                  * TODO: handle changing the primary's name
3258                  */
3259                 break;
3260         case NETDEV_FEAT_CHANGE:
3261                 bond_compute_features(bond);
3262                 break;
3263         default:
3264                 break;
3265         }
3266
3267         return NOTIFY_DONE;
3268 }
3269
3270 /*
3271  * bond_netdev_event: handle netdev notifier chain events.
3272  *
3273  * This function receives events for the netdev chain.  The caller (an
3274  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3275  * locks for us to safely manipulate the slave devices (RTNL lock,
3276  * dev_probe_lock).
3277  */
3278 static int bond_netdev_event(struct notifier_block *this,
3279                              unsigned long event, void *ptr)
3280 {
3281         struct net_device *event_dev = (struct net_device *)ptr;
3282
3283         pr_debug("event_dev: %s, event: %lx\n",
3284                  event_dev ? event_dev->name : "None",
3285                  event);
3286
3287         if (!(event_dev->priv_flags & IFF_BONDING))
3288                 return NOTIFY_DONE;
3289
3290         if (event_dev->flags & IFF_MASTER) {
3291                 pr_debug("IFF_MASTER\n");
3292                 return bond_master_netdev_event(event, event_dev);
3293         }
3294
3295         if (event_dev->flags & IFF_SLAVE) {
3296                 pr_debug("IFF_SLAVE\n");
3297                 return bond_slave_netdev_event(event, event_dev);
3298         }
3299
3300         return NOTIFY_DONE;
3301 }
3302
3303 /*
3304  * bond_inetaddr_event: handle inetaddr notifier chain events.
3305  *
3306  * We keep track of device IPs primarily to use as source addresses in
3307  * ARP monitor probes (rather than spewing out broadcasts all the time).
3308  *
3309  * We track one IP for the main device (if it has one), plus one per VLAN.
3310  */
3311 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3312 {
3313         struct in_ifaddr *ifa = ptr;
3314         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3315         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3316         struct bonding *bond;
3317         struct vlan_entry *vlan;
3318
3319         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3320                 if (bond->dev == event_dev) {
3321                         switch (event) {
3322                         case NETDEV_UP:
3323                                 bond->master_ip = ifa->ifa_local;
3324                                 return NOTIFY_OK;
3325                         case NETDEV_DOWN:
3326                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3327                                 return NOTIFY_OK;
3328                         default:
3329                                 return NOTIFY_DONE;
3330                         }
3331                 }
3332
3333                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3334                         vlan_dev = __vlan_find_dev_deep(bond->dev,
3335                                                         vlan->vlan_id);
3336                         if (vlan_dev == event_dev) {
3337                                 switch (event) {
3338                                 case NETDEV_UP:
3339                                         vlan->vlan_ip = ifa->ifa_local;
3340                                         return NOTIFY_OK;
3341                                 case NETDEV_DOWN:
3342                                         vlan->vlan_ip =
3343                                                 bond_glean_dev_ip(vlan_dev);
3344                                         return NOTIFY_OK;
3345                                 default:
3346                                         return NOTIFY_DONE;
3347                                 }
3348                         }
3349                 }
3350         }
3351         return NOTIFY_DONE;
3352 }
3353
3354 static struct notifier_block bond_netdev_notifier = {
3355         .notifier_call = bond_netdev_event,
3356 };
3357
3358 static struct notifier_block bond_inetaddr_notifier = {
3359         .notifier_call = bond_inetaddr_event,
3360 };
3361
3362 /*---------------------------- Hashing Policies -----------------------------*/
3363
3364 /*
3365  * Hash for the output device based upon layer 2 and layer 3 data. If
3366  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3367  */
3368 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3369 {
3370         struct ethhdr *data = (struct ethhdr *)skb->data;
3371         struct iphdr *iph = ip_hdr(skb);
3372
3373         if (skb->protocol == htons(ETH_P_IP)) {
3374                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3375                         (data->h_dest[5] ^ data->h_source[5])) % count;
3376         }
3377
3378         return (data->h_dest[5] ^ data->h_source[5]) % count;
3379 }
3380
3381 /*
3382  * Hash for the output device based upon layer 3 and layer 4 data. If
3383  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3384  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3385  */
3386 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3387 {
3388         struct ethhdr *data = (struct ethhdr *)skb->data;
3389         struct iphdr *iph = ip_hdr(skb);
3390         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3391         int layer4_xor = 0;
3392
3393         if (skb->protocol == htons(ETH_P_IP)) {
3394                 if (!ip_is_fragment(iph) &&
3395                     (iph->protocol == IPPROTO_TCP ||
3396                      iph->protocol == IPPROTO_UDP)) {
3397                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3398                 }
3399                 return (layer4_xor ^
3400                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3401
3402         }
3403
3404         return (data->h_dest[5] ^ data->h_source[5]) % count;
3405 }
3406
3407 /*
3408  * Hash for the output device based upon layer 2 data
3409  */
3410 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3411 {
3412         struct ethhdr *data = (struct ethhdr *)skb->data;
3413
3414         return (data->h_dest[5] ^ data->h_source[5]) % count;
3415 }
3416
3417 /*-------------------------- Device entry points ----------------------------*/
3418
3419 static int bond_open(struct net_device *bond_dev)
3420 {
3421         struct bonding *bond = netdev_priv(bond_dev);
3422         struct slave *slave;
3423         int i;
3424
3425         bond->kill_timers = 0;
3426
3427         /* reset slave->backup and slave->inactive */
3428         read_lock(&bond->lock);
3429         if (bond->slave_cnt > 0) {
3430                 read_lock(&bond->curr_slave_lock);
3431                 bond_for_each_slave(bond, slave, i) {
3432                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3433                                 && (slave != bond->curr_active_slave)) {
3434                                 bond_set_slave_inactive_flags(slave);
3435                         } else {
3436                                 bond_set_slave_active_flags(slave);
3437                         }
3438                 }
3439                 read_unlock(&bond->curr_slave_lock);
3440         }
3441         read_unlock(&bond->lock);
3442
3443         INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3444
3445         if (bond_is_lb(bond)) {
3446                 /* bond_alb_initialize must be called before the timer
3447                  * is started.
3448                  */
3449                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3450                         /* something went wrong - fail the open operation */
3451                         return -ENOMEM;
3452                 }
3453
3454                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3455                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3456         }
3457
3458         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3459                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3460                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3461         }
3462
3463         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3464                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3465                         INIT_DELAYED_WORK(&bond->arp_work,
3466                                           bond_activebackup_arp_mon);
3467                 else
3468                         INIT_DELAYED_WORK(&bond->arp_work,
3469                                           bond_loadbalance_arp_mon);
3470
3471                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3472                 if (bond->params.arp_validate)
3473                         bond->recv_probe = bond_arp_rcv;
3474         }
3475
3476         if (bond->params.mode == BOND_MODE_8023AD) {
3477                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3478                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3479                 /* register to receive LACPDUs */
3480                 bond->recv_probe = bond_3ad_lacpdu_recv;
3481                 bond_3ad_initiate_agg_selection(bond, 1);
3482         }
3483
3484         return 0;
3485 }
3486
3487 static int bond_close(struct net_device *bond_dev)
3488 {
3489         struct bonding *bond = netdev_priv(bond_dev);
3490
3491         write_lock_bh(&bond->lock);
3492
3493         bond->send_peer_notif = 0;
3494
3495         /* signal timers not to re-arm */
3496         bond->kill_timers = 1;
3497
3498         write_unlock_bh(&bond->lock);
3499
3500         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3501                 cancel_delayed_work(&bond->mii_work);
3502         }
3503
3504         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3505                 cancel_delayed_work(&bond->arp_work);
3506         }
3507
3508         switch (bond->params.mode) {
3509         case BOND_MODE_8023AD:
3510                 cancel_delayed_work(&bond->ad_work);
3511                 break;
3512         case BOND_MODE_TLB:
3513         case BOND_MODE_ALB:
3514                 cancel_delayed_work(&bond->alb_work);
3515                 break;
3516         default:
3517                 break;
3518         }
3519
3520         if (delayed_work_pending(&bond->mcast_work))
3521                 cancel_delayed_work(&bond->mcast_work);
3522
3523         if (bond_is_lb(bond)) {
3524                 /* Must be called only after all
3525                  * slaves have been released
3526                  */
3527                 bond_alb_deinitialize(bond);
3528         }
3529         bond->recv_probe = NULL;
3530
3531         return 0;
3532 }
3533
3534 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3535                                                 struct rtnl_link_stats64 *stats)
3536 {
3537         struct bonding *bond = netdev_priv(bond_dev);
3538         struct rtnl_link_stats64 temp;
3539         struct slave *slave;
3540         int i;
3541
3542         memset(stats, 0, sizeof(*stats));
3543
3544         read_lock_bh(&bond->lock);
3545
3546         bond_for_each_slave(bond, slave, i) {
3547                 const struct rtnl_link_stats64 *sstats =
3548                         dev_get_stats(slave->dev, &temp);
3549
3550                 stats->rx_packets += sstats->rx_packets;
3551                 stats->rx_bytes += sstats->rx_bytes;
3552                 stats->rx_errors += sstats->rx_errors;
3553                 stats->rx_dropped += sstats->rx_dropped;
3554
3555                 stats->tx_packets += sstats->tx_packets;
3556                 stats->tx_bytes += sstats->tx_bytes;
3557                 stats->tx_errors += sstats->tx_errors;
3558                 stats->tx_dropped += sstats->tx_dropped;
3559
3560                 stats->multicast += sstats->multicast;
3561                 stats->collisions += sstats->collisions;
3562
3563                 stats->rx_length_errors += sstats->rx_length_errors;
3564                 stats->rx_over_errors += sstats->rx_over_errors;
3565                 stats->rx_crc_errors += sstats->rx_crc_errors;
3566                 stats->rx_frame_errors += sstats->rx_frame_errors;
3567                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3568                 stats->rx_missed_errors += sstats->rx_missed_errors;
3569
3570                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3571                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3572                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3573                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3574                 stats->tx_window_errors += sstats->tx_window_errors;
3575         }
3576
3577         read_unlock_bh(&bond->lock);
3578
3579         return stats;
3580 }
3581
3582 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3583 {
3584         struct net_device *slave_dev = NULL;
3585         struct ifbond k_binfo;
3586         struct ifbond __user *u_binfo = NULL;
3587         struct ifslave k_sinfo;
3588         struct ifslave __user *u_sinfo = NULL;
3589         struct mii_ioctl_data *mii = NULL;
3590         int res = 0;
3591
3592         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3593
3594         switch (cmd) {
3595         case SIOCGMIIPHY:
3596                 mii = if_mii(ifr);
3597                 if (!mii)
3598                         return -EINVAL;
3599
3600                 mii->phy_id = 0;
3601                 /* Fall Through */
3602         case SIOCGMIIREG:
3603                 /*
3604                  * We do this again just in case we were called by SIOCGMIIREG
3605                  * instead of SIOCGMIIPHY.
3606                  */
3607                 mii = if_mii(ifr);
3608                 if (!mii)
3609                         return -EINVAL;
3610
3611
3612                 if (mii->reg_num == 1) {
3613                         struct bonding *bond = netdev_priv(bond_dev);
3614                         mii->val_out = 0;
3615                         read_lock(&bond->lock);
3616                         read_lock(&bond->curr_slave_lock);
3617                         if (netif_carrier_ok(bond->dev))
3618                                 mii->val_out = BMSR_LSTATUS;
3619
3620                         read_unlock(&bond->curr_slave_lock);
3621                         read_unlock(&bond->lock);
3622                 }
3623
3624                 return 0;
3625         case BOND_INFO_QUERY_OLD:
3626         case SIOCBONDINFOQUERY:
3627                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3628
3629                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3630                         return -EFAULT;
3631
3632                 res = bond_info_query(bond_dev, &k_binfo);
3633                 if (res == 0 &&
3634                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3635                         return -EFAULT;
3636
3637                 return res;
3638         case BOND_SLAVE_INFO_QUERY_OLD:
3639         case SIOCBONDSLAVEINFOQUERY:
3640                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3641
3642                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3643                         return -EFAULT;
3644
3645                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3646                 if (res == 0 &&
3647                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3648                         return -EFAULT;
3649
3650                 return res;
3651         default:
3652                 /* Go on */
3653                 break;
3654         }
3655
3656         if (!capable(CAP_NET_ADMIN))
3657                 return -EPERM;
3658
3659         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3660
3661         pr_debug("slave_dev=%p:\n", slave_dev);
3662
3663         if (!slave_dev)
3664                 res = -ENODEV;
3665         else {
3666                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3667                 switch (cmd) {
3668                 case BOND_ENSLAVE_OLD:
3669                 case SIOCBONDENSLAVE:
3670                         res = bond_enslave(bond_dev, slave_dev);
3671                         break;
3672                 case BOND_RELEASE_OLD:
3673                 case SIOCBONDRELEASE:
3674                         res = bond_release(bond_dev, slave_dev);
3675                         break;
3676                 case BOND_SETHWADDR_OLD:
3677                 case SIOCBONDSETHWADDR:
3678                         res = bond_sethwaddr(bond_dev, slave_dev);
3679                         break;
3680                 case BOND_CHANGE_ACTIVE_OLD:
3681                 case SIOCBONDCHANGEACTIVE:
3682                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3683                         break;
3684                 default:
3685                         res = -EOPNOTSUPP;
3686                 }
3687
3688                 dev_put(slave_dev);
3689         }
3690
3691         return res;
3692 }
3693
3694 static bool bond_addr_in_mc_list(unsigned char *addr,
3695                                  struct netdev_hw_addr_list *list,
3696                                  int addrlen)
3697 {
3698         struct netdev_hw_addr *ha;
3699
3700         netdev_hw_addr_list_for_each(ha, list)
3701                 if (!memcmp(ha->addr, addr, addrlen))
3702                         return true;
3703
3704         return false;
3705 }
3706
3707 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3708 {
3709         struct bonding *bond = netdev_priv(bond_dev);
3710
3711         if (change & IFF_PROMISC)
3712                 bond_set_promiscuity(bond,
3713                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3714
3715         if (change & IFF_ALLMULTI)
3716                 bond_set_allmulti(bond,
3717                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3718 }
3719
3720 static void bond_set_multicast_list(struct net_device *bond_dev)
3721 {
3722         struct bonding *bond = netdev_priv(bond_dev);
3723         struct netdev_hw_addr *ha;
3724         bool found;
3725
3726         read_lock(&bond->lock);
3727
3728         /* looking for addresses to add to slaves' mc list */
3729         netdev_for_each_mc_addr(ha, bond_dev) {
3730                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3731                                              bond_dev->addr_len);
3732                 if (!found)
3733                         bond_mc_add(bond, ha->addr);
3734         }
3735
3736         /* looking for addresses to delete from slaves' list */
3737         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3738                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3739                                              bond_dev->addr_len);
3740                 if (!found)
3741                         bond_mc_del(bond, ha->addr);
3742         }
3743
3744         /* save master's multicast list */
3745         __hw_addr_flush(&bond->mc_list);
3746         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3747                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3748
3749         read_unlock(&bond->lock);
3750 }
3751
3752 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3753 {
3754         struct bonding *bond = netdev_priv(dev);
3755         struct slave *slave = bond->first_slave;
3756
3757         if (slave) {
3758                 const struct net_device_ops *slave_ops
3759                         = slave->dev->netdev_ops;
3760                 if (slave_ops->ndo_neigh_setup)
3761                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
3762         }
3763         return 0;
3764 }
3765
3766 /*
3767  * Change the MTU of all of a master's slaves to match the master
3768  */
3769 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3770 {
3771         struct bonding *bond = netdev_priv(bond_dev);
3772         struct slave *slave, *stop_at;
3773         int res = 0;
3774         int i;
3775
3776         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3777                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3778
3779         /* Can't hold bond->lock with bh disabled here since
3780          * some base drivers panic. On the other hand we can't
3781          * hold bond->lock without bh disabled because we'll
3782          * deadlock. The only solution is to rely on the fact
3783          * that we're under rtnl_lock here, and the slaves
3784          * list won't change. This doesn't solve the problem
3785          * of setting the slave's MTU while it is
3786          * transmitting, but the assumption is that the base
3787          * driver can handle that.
3788          *
3789          * TODO: figure out a way to safely iterate the slaves
3790          * list, but without holding a lock around the actual
3791          * call to the base driver.
3792          */
3793
3794         bond_for_each_slave(bond, slave, i) {
3795                 pr_debug("s %p s->p %p c_m %p\n",
3796                          slave,
3797                          slave->prev,
3798                          slave->dev->netdev_ops->ndo_change_mtu);
3799
3800                 res = dev_set_mtu(slave->dev, new_mtu);
3801
3802                 if (res) {
3803                         /* If we failed to set the slave's mtu to the new value
3804                          * we must abort the operation even in ACTIVE_BACKUP
3805                          * mode, because if we allow the backup slaves to have
3806                          * different mtu values than the active slave we'll
3807                          * need to change their mtu when doing a failover. That
3808                          * means changing their mtu from timer context, which
3809                          * is probably not a good idea.
3810                          */
3811                         pr_debug("err %d %s\n", res, slave->dev->name);
3812                         goto unwind;
3813                 }
3814         }
3815
3816         bond_dev->mtu = new_mtu;
3817
3818         return 0;
3819
3820 unwind:
3821         /* unwind from head to the slave that failed */
3822         stop_at = slave;
3823         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3824                 int tmp_res;
3825
3826                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3827                 if (tmp_res) {
3828                         pr_debug("unwind err %d dev %s\n",
3829                                  tmp_res, slave->dev->name);
3830                 }
3831         }
3832
3833         return res;
3834 }
3835
3836 /*
3837  * Change HW address
3838  *
3839  * Note that many devices must be down to change the HW address, and
3840  * downing the master releases all slaves.  We can make bonds full of
3841  * bonding devices to test this, however.
3842  */
3843 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3844 {
3845         struct bonding *bond = netdev_priv(bond_dev);
3846         struct sockaddr *sa = addr, tmp_sa;
3847         struct slave *slave, *stop_at;
3848         int res = 0;
3849         int i;
3850
3851         if (bond->params.mode == BOND_MODE_ALB)
3852                 return bond_alb_set_mac_address(bond_dev, addr);
3853
3854
3855         pr_debug("bond=%p, name=%s\n",
3856                  bond, bond_dev ? bond_dev->name : "None");
3857
3858         /*
3859          * If fail_over_mac is set to active, do nothing and return
3860          * success.  Returning an error causes ifenslave to fail.
3861          */
3862         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3863                 return 0;
3864
3865         if (!is_valid_ether_addr(sa->sa_data))
3866                 return -EADDRNOTAVAIL;
3867
3868         /* Can't hold bond->lock with bh disabled here since
3869          * some base drivers panic. On the other hand we can't
3870          * hold bond->lock without bh disabled because we'll
3871          * deadlock. The only solution is to rely on the fact
3872          * that we're under rtnl_lock here, and the slaves
3873          * list won't change. This doesn't solve the problem
3874          * of setting the slave's hw address while it is
3875          * transmitting, but the assumption is that the base
3876          * driver can handle that.
3877          *
3878          * TODO: figure out a way to safely iterate the slaves
3879          * list, but without holding a lock around the actual
3880          * call to the base driver.
3881          */
3882
3883         bond_for_each_slave(bond, slave, i) {
3884                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3885                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3886
3887                 if (slave_ops->ndo_set_mac_address == NULL) {
3888                         res = -EOPNOTSUPP;
3889                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3890                         goto unwind;
3891                 }
3892
3893                 res = dev_set_mac_address(slave->dev, addr);
3894                 if (res) {
3895                         /* TODO: consider downing the slave
3896                          * and retry ?
3897                          * User should expect communications
3898                          * breakage anyway until ARP finish
3899                          * updating, so...
3900                          */
3901                         pr_debug("err %d %s\n", res, slave->dev->name);
3902                         goto unwind;
3903                 }
3904         }
3905
3906         /* success */
3907         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3908         return 0;
3909
3910 unwind:
3911         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3912         tmp_sa.sa_family = bond_dev->type;
3913
3914         /* unwind from head to the slave that failed */
3915         stop_at = slave;
3916         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3917                 int tmp_res;
3918
3919                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3920                 if (tmp_res) {
3921                         pr_debug("unwind err %d dev %s\n",
3922                                  tmp_res, slave->dev->name);
3923                 }
3924         }
3925
3926         return res;
3927 }
3928
3929 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3930 {
3931         struct bonding *bond = netdev_priv(bond_dev);
3932         struct slave *slave, *start_at;
3933         int i, slave_no, res = 1;
3934         struct iphdr *iph = ip_hdr(skb);
3935
3936         /*
3937          * Start with the curr_active_slave that joined the bond as the
3938          * default for sending IGMP traffic.  For failover purposes one
3939          * needs to maintain some consistency for the interface that will
3940          * send the join/membership reports.  The curr_active_slave found
3941          * will send all of this type of traffic.
3942          */
3943         if ((iph->protocol == IPPROTO_IGMP) &&
3944             (skb->protocol == htons(ETH_P_IP))) {
3945
3946                 read_lock(&bond->curr_slave_lock);
3947                 slave = bond->curr_active_slave;
3948                 read_unlock(&bond->curr_slave_lock);
3949
3950                 if (!slave)
3951                         goto out;
3952         } else {
3953                 /*
3954                  * Concurrent TX may collide on rr_tx_counter; we accept
3955                  * that as being rare enough not to justify using an
3956                  * atomic op here.
3957                  */
3958                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3959
3960                 bond_for_each_slave(bond, slave, i) {
3961                         slave_no--;
3962                         if (slave_no < 0)
3963                                 break;
3964                 }
3965         }
3966
3967         start_at = slave;
3968         bond_for_each_slave_from(bond, slave, i, start_at) {
3969                 if (IS_UP(slave->dev) &&
3970                     (slave->link == BOND_LINK_UP) &&
3971                     bond_is_active_slave(slave)) {
3972                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
3973                         break;
3974                 }
3975         }
3976
3977 out:
3978         if (res) {
3979                 /* no suitable interface, frame not sent */
3980                 dev_kfree_skb(skb);
3981         }
3982
3983         return NETDEV_TX_OK;
3984 }
3985
3986
3987 /*
3988  * in active-backup mode, we know that bond->curr_active_slave is always valid if
3989  * the bond has a usable interface.
3990  */
3991 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3992 {
3993         struct bonding *bond = netdev_priv(bond_dev);
3994         int res = 1;
3995
3996         read_lock(&bond->curr_slave_lock);
3997
3998         if (bond->curr_active_slave)
3999                 res = bond_dev_queue_xmit(bond, skb,
4000                         bond->curr_active_slave->dev);
4001
4002         if (res)
4003                 /* no suitable interface, frame not sent */
4004                 dev_kfree_skb(skb);
4005
4006         read_unlock(&bond->curr_slave_lock);
4007
4008         return NETDEV_TX_OK;
4009 }
4010
4011 /*
4012  * In bond_xmit_xor() , we determine the output device by using a pre-
4013  * determined xmit_hash_policy(), If the selected device is not enabled,
4014  * find the next active slave.
4015  */
4016 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4017 {
4018         struct bonding *bond = netdev_priv(bond_dev);
4019         struct slave *slave, *start_at;
4020         int slave_no;
4021         int i;
4022         int res = 1;
4023
4024         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4025
4026         bond_for_each_slave(bond, slave, i) {
4027                 slave_no--;
4028                 if (slave_no < 0)
4029                         break;
4030         }
4031
4032         start_at = slave;
4033
4034         bond_for_each_slave_from(bond, slave, i, start_at) {
4035                 if (IS_UP(slave->dev) &&
4036                     (slave->link == BOND_LINK_UP) &&
4037                     bond_is_active_slave(slave)) {
4038                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4039                         break;
4040                 }
4041         }
4042
4043         if (res) {
4044                 /* no suitable interface, frame not sent */
4045                 dev_kfree_skb(skb);
4046         }
4047
4048         return NETDEV_TX_OK;
4049 }
4050
4051 /*
4052  * in broadcast mode, we send everything to all usable interfaces.
4053  */
4054 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4055 {
4056         struct bonding *bond = netdev_priv(bond_dev);
4057         struct slave *slave, *start_at;
4058         struct net_device *tx_dev = NULL;
4059         int i;
4060         int res = 1;
4061
4062         read_lock(&bond->curr_slave_lock);
4063         start_at = bond->curr_active_slave;
4064         read_unlock(&bond->curr_slave_lock);
4065
4066         if (!start_at)
4067                 goto out;
4068
4069         bond_for_each_slave_from(bond, slave, i, start_at) {
4070                 if (IS_UP(slave->dev) &&
4071                     (slave->link == BOND_LINK_UP) &&
4072                     bond_is_active_slave(slave)) {
4073                         if (tx_dev) {
4074                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4075                                 if (!skb2) {
4076                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4077                                                bond_dev->name);
4078                                         continue;
4079                                 }
4080
4081                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4082                                 if (res) {
4083                                         dev_kfree_skb(skb2);
4084                                         continue;
4085                                 }
4086                         }
4087                         tx_dev = slave->dev;
4088                 }
4089         }
4090
4091         if (tx_dev)
4092                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4093
4094 out:
4095         if (res)
4096                 /* no suitable interface, frame not sent */
4097                 dev_kfree_skb(skb);
4098
4099         /* frame sent to all suitable interfaces */
4100         return NETDEV_TX_OK;
4101 }
4102
4103 /*------------------------- Device initialization ---------------------------*/
4104
4105 static void bond_set_xmit_hash_policy(struct bonding *bond)
4106 {
4107         switch (bond->params.xmit_policy) {
4108         case BOND_XMIT_POLICY_LAYER23:
4109                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4110                 break;
4111         case BOND_XMIT_POLICY_LAYER34:
4112                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4113                 break;
4114         case BOND_XMIT_POLICY_LAYER2:
4115         default:
4116                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4117                 break;
4118         }
4119 }
4120
4121 /*
4122  * Lookup the slave that corresponds to a qid
4123  */
4124 static inline int bond_slave_override(struct bonding *bond,
4125                                       struct sk_buff *skb)
4126 {
4127         int i, res = 1;
4128         struct slave *slave = NULL;
4129         struct slave *check_slave;
4130
4131         if (!skb->queue_mapping)
4132                 return 1;
4133
4134         /* Find out if any slaves have the same mapping as this skb. */
4135         bond_for_each_slave(bond, check_slave, i) {
4136                 if (check_slave->queue_id == skb->queue_mapping) {
4137                         slave = check_slave;
4138                         break;
4139                 }
4140         }
4141
4142         /* If the slave isn't UP, use default transmit policy. */
4143         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4144             (slave->link == BOND_LINK_UP)) {
4145                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4146         }
4147
4148         return res;
4149 }
4150
4151
4152 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4153 {
4154         /*
4155          * This helper function exists to help dev_pick_tx get the correct
4156          * destination queue.  Using a helper function skips a call to
4157          * skb_tx_hash and will put the skbs in the queue we expect on their
4158          * way down to the bonding driver.
4159          */
4160         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4161
4162         /*
4163          * Save the original txq to restore before passing to the driver
4164          */
4165         bond_queue_mapping(skb) = skb->queue_mapping;
4166
4167         if (unlikely(txq >= dev->real_num_tx_queues)) {
4168                 do {
4169                         txq -= dev->real_num_tx_queues;
4170                 } while (txq >= dev->real_num_tx_queues);
4171         }
4172         return txq;
4173 }
4174
4175 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4176 {
4177         struct bonding *bond = netdev_priv(dev);
4178
4179         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4180                 if (!bond_slave_override(bond, skb))
4181                         return NETDEV_TX_OK;
4182         }
4183
4184         switch (bond->params.mode) {
4185         case BOND_MODE_ROUNDROBIN:
4186                 return bond_xmit_roundrobin(skb, dev);
4187         case BOND_MODE_ACTIVEBACKUP:
4188                 return bond_xmit_activebackup(skb, dev);
4189         case BOND_MODE_XOR:
4190                 return bond_xmit_xor(skb, dev);
4191         case BOND_MODE_BROADCAST:
4192                 return bond_xmit_broadcast(skb, dev);
4193         case BOND_MODE_8023AD:
4194                 return bond_3ad_xmit_xor(skb, dev);
4195         case BOND_MODE_ALB:
4196         case BOND_MODE_TLB:
4197                 return bond_alb_xmit(skb, dev);
4198         default:
4199                 /* Should never happen, mode already checked */
4200                 pr_err("%s: Error: Unknown bonding mode %d\n",
4201                        dev->name, bond->params.mode);
4202                 WARN_ON_ONCE(1);
4203                 dev_kfree_skb(skb);
4204                 return NETDEV_TX_OK;
4205         }
4206 }
4207
4208 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4209 {
4210         struct bonding *bond = netdev_priv(dev);
4211         netdev_tx_t ret = NETDEV_TX_OK;
4212
4213         /*
4214          * If we risk deadlock from transmitting this in the
4215          * netpoll path, tell netpoll to queue the frame for later tx
4216          */
4217         if (is_netpoll_tx_blocked(dev))
4218                 return NETDEV_TX_BUSY;
4219
4220         read_lock(&bond->lock);
4221
4222         if (bond->slave_cnt)
4223                 ret = __bond_start_xmit(skb, dev);
4224         else
4225                 dev_kfree_skb(skb);
4226
4227         read_unlock(&bond->lock);
4228
4229         return ret;
4230 }
4231
4232 /*
4233  * set bond mode specific net device operations
4234  */
4235 void bond_set_mode_ops(struct bonding *bond, int mode)
4236 {
4237         struct net_device *bond_dev = bond->dev;
4238
4239         switch (mode) {
4240         case BOND_MODE_ROUNDROBIN:
4241                 break;
4242         case BOND_MODE_ACTIVEBACKUP:
4243                 break;
4244         case BOND_MODE_XOR:
4245                 bond_set_xmit_hash_policy(bond);
4246                 break;
4247         case BOND_MODE_BROADCAST:
4248                 break;
4249         case BOND_MODE_8023AD:
4250                 bond_set_xmit_hash_policy(bond);
4251                 break;
4252         case BOND_MODE_ALB:
4253                 /* FALLTHRU */
4254         case BOND_MODE_TLB:
4255                 break;
4256         default:
4257                 /* Should never happen, mode already checked */
4258                 pr_err("%s: Error: Unknown bonding mode %d\n",
4259                        bond_dev->name, mode);
4260                 break;
4261         }
4262 }
4263
4264 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4265                                     struct ethtool_drvinfo *drvinfo)
4266 {
4267         strncpy(drvinfo->driver, DRV_NAME, 32);
4268         strncpy(drvinfo->version, DRV_VERSION, 32);
4269         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4270 }
4271
4272 static const struct ethtool_ops bond_ethtool_ops = {
4273         .get_drvinfo            = bond_ethtool_get_drvinfo,
4274         .get_link               = ethtool_op_get_link,
4275 };
4276
4277 static const struct net_device_ops bond_netdev_ops = {
4278         .ndo_init               = bond_init,
4279         .ndo_uninit             = bond_uninit,
4280         .ndo_open               = bond_open,
4281         .ndo_stop               = bond_close,
4282         .ndo_start_xmit         = bond_start_xmit,
4283         .ndo_select_queue       = bond_select_queue,
4284         .ndo_get_stats64        = bond_get_stats,
4285         .ndo_do_ioctl           = bond_do_ioctl,
4286         .ndo_change_rx_flags    = bond_change_rx_flags,
4287         .ndo_set_rx_mode        = bond_set_multicast_list,
4288         .ndo_change_mtu         = bond_change_mtu,
4289         .ndo_set_mac_address    = bond_set_mac_address,
4290         .ndo_neigh_setup        = bond_neigh_setup,
4291         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4292         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4293 #ifdef CONFIG_NET_POLL_CONTROLLER
4294         .ndo_netpoll_setup      = bond_netpoll_setup,
4295         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4296         .ndo_poll_controller    = bond_poll_controller,
4297 #endif
4298         .ndo_add_slave          = bond_enslave,
4299         .ndo_del_slave          = bond_release,
4300         .ndo_fix_features       = bond_fix_features,
4301 };
4302
4303 static void bond_destructor(struct net_device *bond_dev)
4304 {
4305         struct bonding *bond = netdev_priv(bond_dev);
4306         if (bond->wq)
4307                 destroy_workqueue(bond->wq);
4308         free_netdev(bond_dev);
4309 }
4310
4311 static void bond_setup(struct net_device *bond_dev)
4312 {
4313         struct bonding *bond = netdev_priv(bond_dev);
4314
4315         /* initialize rwlocks */
4316         rwlock_init(&bond->lock);
4317         rwlock_init(&bond->curr_slave_lock);
4318
4319         bond->params = bonding_defaults;
4320
4321         /* Initialize pointers */
4322         bond->dev = bond_dev;
4323         INIT_LIST_HEAD(&bond->vlan_list);
4324
4325         /* Initialize the device entry points */
4326         ether_setup(bond_dev);
4327         bond_dev->netdev_ops = &bond_netdev_ops;
4328         bond_dev->ethtool_ops = &bond_ethtool_ops;
4329         bond_set_mode_ops(bond, bond->params.mode);
4330
4331         bond_dev->destructor = bond_destructor;
4332
4333         /* Initialize the device options */
4334         bond_dev->tx_queue_len = 0;
4335         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4336         bond_dev->priv_flags |= IFF_BONDING;
4337         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4338
4339         /* At first, we block adding VLANs. That's the only way to
4340          * prevent problems that occur when adding VLANs over an
4341          * empty bond. The block will be removed once non-challenged
4342          * slaves are enslaved.
4343          */
4344         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4345
4346         /* don't acquire bond device's netif_tx_lock when
4347          * transmitting */
4348         bond_dev->features |= NETIF_F_LLTX;
4349
4350         /* By default, we declare the bond to be fully
4351          * VLAN hardware accelerated capable. Special
4352          * care is taken in the various xmit functions
4353          * when there are slaves that are not hw accel
4354          * capable
4355          */
4356
4357         bond_dev->hw_features = BOND_VLAN_FEATURES |
4358                                 NETIF_F_HW_VLAN_TX |
4359                                 NETIF_F_HW_VLAN_RX |
4360                                 NETIF_F_HW_VLAN_FILTER;
4361
4362         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM);
4363         bond_dev->features |= bond_dev->hw_features;
4364 }
4365
4366 static void bond_work_cancel_all(struct bonding *bond)
4367 {
4368         write_lock_bh(&bond->lock);
4369         bond->kill_timers = 1;
4370         write_unlock_bh(&bond->lock);
4371
4372         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4373                 cancel_delayed_work(&bond->mii_work);
4374
4375         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4376                 cancel_delayed_work(&bond->arp_work);
4377
4378         if (bond->params.mode == BOND_MODE_ALB &&
4379             delayed_work_pending(&bond->alb_work))
4380                 cancel_delayed_work(&bond->alb_work);
4381
4382         if (bond->params.mode == BOND_MODE_8023AD &&
4383             delayed_work_pending(&bond->ad_work))
4384                 cancel_delayed_work(&bond->ad_work);
4385
4386         if (delayed_work_pending(&bond->mcast_work))
4387                 cancel_delayed_work(&bond->mcast_work);
4388 }
4389
4390 /*
4391 * Destroy a bonding device.
4392 * Must be under rtnl_lock when this function is called.
4393 */
4394 static void bond_uninit(struct net_device *bond_dev)
4395 {
4396         struct bonding *bond = netdev_priv(bond_dev);
4397         struct vlan_entry *vlan, *tmp;
4398
4399         bond_netpoll_cleanup(bond_dev);
4400
4401         /* Release the bonded slaves */
4402         bond_release_all(bond_dev);
4403
4404         list_del(&bond->bond_list);
4405
4406         bond_work_cancel_all(bond);
4407
4408         bond_remove_proc_entry(bond);
4409
4410         bond_debug_unregister(bond);
4411
4412         __hw_addr_flush(&bond->mc_list);
4413
4414         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4415                 list_del(&vlan->vlan_list);
4416                 kfree(vlan);
4417         }
4418 }
4419
4420 /*------------------------- Module initialization ---------------------------*/
4421
4422 /*
4423  * Convert string input module parms.  Accept either the
4424  * number of the mode or its string name.  A bit complicated because
4425  * some mode names are substrings of other names, and calls from sysfs
4426  * may have whitespace in the name (trailing newlines, for example).
4427  */
4428 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4429 {
4430         int modeint = -1, i, rv;
4431         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4432
4433         for (p = (char *)buf; *p; p++)
4434                 if (!(isdigit(*p) || isspace(*p)))
4435                         break;
4436
4437         if (*p)
4438                 rv = sscanf(buf, "%20s", modestr);
4439         else
4440                 rv = sscanf(buf, "%d", &modeint);
4441
4442         if (!rv)
4443                 return -1;
4444
4445         for (i = 0; tbl[i].modename; i++) {
4446                 if (modeint == tbl[i].mode)
4447                         return tbl[i].mode;
4448                 if (strcmp(modestr, tbl[i].modename) == 0)
4449                         return tbl[i].mode;
4450         }
4451
4452         return -1;
4453 }
4454
4455 static int bond_check_params(struct bond_params *params)
4456 {
4457         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4458
4459         /*
4460          * Convert string parameters.
4461          */
4462         if (mode) {
4463                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4464                 if (bond_mode == -1) {
4465                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4466                                mode == NULL ? "NULL" : mode);
4467                         return -EINVAL;
4468                 }
4469         }
4470
4471         if (xmit_hash_policy) {
4472                 if ((bond_mode != BOND_MODE_XOR) &&
4473                     (bond_mode != BOND_MODE_8023AD)) {
4474                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4475                                bond_mode_name(bond_mode));
4476                 } else {
4477                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4478                                                         xmit_hashtype_tbl);
4479                         if (xmit_hashtype == -1) {
4480                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4481                                        xmit_hash_policy == NULL ? "NULL" :
4482                                        xmit_hash_policy);
4483                                 return -EINVAL;
4484                         }
4485                 }
4486         }
4487
4488         if (lacp_rate) {
4489                 if (bond_mode != BOND_MODE_8023AD) {
4490                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4491                                 bond_mode_name(bond_mode));
4492                 } else {
4493                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4494                         if (lacp_fast == -1) {
4495                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4496                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4497                                 return -EINVAL;
4498                         }
4499                 }
4500         }
4501
4502         if (ad_select) {
4503                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4504                 if (params->ad_select == -1) {
4505                         pr_err("Error: Invalid ad_select \"%s\"\n",
4506                                ad_select == NULL ? "NULL" : ad_select);
4507                         return -EINVAL;
4508                 }
4509
4510                 if (bond_mode != BOND_MODE_8023AD) {
4511                         pr_warning("ad_select param only affects 802.3ad mode\n");
4512                 }
4513         } else {
4514                 params->ad_select = BOND_AD_STABLE;
4515         }
4516
4517         if (max_bonds < 0) {
4518                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4519                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4520                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4521         }
4522
4523         if (miimon < 0) {
4524                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4525                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4526                 miimon = BOND_LINK_MON_INTERV;
4527         }
4528
4529         if (updelay < 0) {
4530                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4531                            updelay, INT_MAX);
4532                 updelay = 0;
4533         }
4534
4535         if (downdelay < 0) {
4536                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4537                            downdelay, INT_MAX);
4538                 downdelay = 0;
4539         }
4540
4541         if ((use_carrier != 0) && (use_carrier != 1)) {
4542                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4543                            use_carrier);
4544                 use_carrier = 1;
4545         }
4546
4547         if (num_peer_notif < 0 || num_peer_notif > 255) {
4548                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4549                            num_peer_notif);
4550                 num_peer_notif = 1;
4551         }
4552
4553         /* reset values for 802.3ad */
4554         if (bond_mode == BOND_MODE_8023AD) {
4555                 if (!miimon) {
4556                         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");
4557                         pr_warning("Forcing miimon to 100msec\n");
4558                         miimon = 100;
4559                 }
4560         }
4561
4562         if (tx_queues < 1 || tx_queues > 255) {
4563                 pr_warning("Warning: tx_queues (%d) should be between "
4564                            "1 and 255, resetting to %d\n",
4565                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4566                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4567         }
4568
4569         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4570                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4571                            "not of valid value (0/1), so it was set to "
4572                            "0\n", all_slaves_active);
4573                 all_slaves_active = 0;
4574         }
4575
4576         if (resend_igmp < 0 || resend_igmp > 255) {
4577                 pr_warning("Warning: resend_igmp (%d) should be between "
4578                            "0 and 255, resetting to %d\n",
4579                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4580                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4581         }
4582
4583         /* reset values for TLB/ALB */
4584         if ((bond_mode == BOND_MODE_TLB) ||
4585             (bond_mode == BOND_MODE_ALB)) {
4586                 if (!miimon) {
4587                         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");
4588                         pr_warning("Forcing miimon to 100msec\n");
4589                         miimon = 100;
4590                 }
4591         }
4592
4593         if (bond_mode == BOND_MODE_ALB) {
4594                 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",
4595                           updelay);
4596         }
4597
4598         if (!miimon) {
4599                 if (updelay || downdelay) {
4600                         /* just warn the user the up/down delay will have
4601                          * no effect since miimon is zero...
4602                          */
4603                         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",
4604                                    updelay, downdelay);
4605                 }
4606         } else {
4607                 /* don't allow arp monitoring */
4608                 if (arp_interval) {
4609                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4610                                    miimon, arp_interval);
4611                         arp_interval = 0;
4612                 }
4613
4614                 if ((updelay % miimon) != 0) {
4615                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4616                                    updelay, miimon,
4617                                    (updelay / miimon) * miimon);
4618                 }
4619
4620                 updelay /= miimon;
4621
4622                 if ((downdelay % miimon) != 0) {
4623                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4624                                    downdelay, miimon,
4625                                    (downdelay / miimon) * miimon);
4626                 }
4627
4628                 downdelay /= miimon;
4629         }
4630
4631         if (arp_interval < 0) {
4632                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4633                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4634                 arp_interval = BOND_LINK_ARP_INTERV;
4635         }
4636
4637         for (arp_ip_count = 0;
4638              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4639              arp_ip_count++) {
4640                 /* not complete check, but should be good enough to
4641                    catch mistakes */
4642                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4643                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4644                                    arp_ip_target[arp_ip_count]);
4645                         arp_interval = 0;
4646                 } else {
4647                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4648                         arp_target[arp_ip_count] = ip;
4649                 }
4650         }
4651
4652         if (arp_interval && !arp_ip_count) {
4653                 /* don't allow arping if no arp_ip_target given... */
4654                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4655                            arp_interval);
4656                 arp_interval = 0;
4657         }
4658
4659         if (arp_validate) {
4660                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4661                         pr_err("arp_validate only supported in active-backup mode\n");
4662                         return -EINVAL;
4663                 }
4664                 if (!arp_interval) {
4665                         pr_err("arp_validate requires arp_interval\n");
4666                         return -EINVAL;
4667                 }
4668
4669                 arp_validate_value = bond_parse_parm(arp_validate,
4670                                                      arp_validate_tbl);
4671                 if (arp_validate_value == -1) {
4672                         pr_err("Error: invalid arp_validate \"%s\"\n",
4673                                arp_validate == NULL ? "NULL" : arp_validate);
4674                         return -EINVAL;
4675                 }
4676         } else
4677                 arp_validate_value = 0;
4678
4679         if (miimon) {
4680                 pr_info("MII link monitoring set to %d ms\n", miimon);
4681         } else if (arp_interval) {
4682                 int i;
4683
4684                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4685                         arp_interval,
4686                         arp_validate_tbl[arp_validate_value].modename,
4687                         arp_ip_count);
4688
4689                 for (i = 0; i < arp_ip_count; i++)
4690                         pr_info(" %s", arp_ip_target[i]);
4691
4692                 pr_info("\n");
4693
4694         } else if (max_bonds) {
4695                 /* miimon and arp_interval not set, we need one so things
4696                  * work as expected, see bonding.txt for details
4697                  */
4698                 pr_debug("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");
4699         }
4700
4701         if (primary && !USES_PRIMARY(bond_mode)) {
4702                 /* currently, using a primary only makes sense
4703                  * in active backup, TLB or ALB modes
4704                  */
4705                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4706                            primary, bond_mode_name(bond_mode));
4707                 primary = NULL;
4708         }
4709
4710         if (primary && primary_reselect) {
4711                 primary_reselect_value = bond_parse_parm(primary_reselect,
4712                                                          pri_reselect_tbl);
4713                 if (primary_reselect_value == -1) {
4714                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4715                                primary_reselect ==
4716                                         NULL ? "NULL" : primary_reselect);
4717                         return -EINVAL;
4718                 }
4719         } else {
4720                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4721         }
4722
4723         if (fail_over_mac) {
4724                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4725                                                       fail_over_mac_tbl);
4726                 if (fail_over_mac_value == -1) {
4727                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4728                                arp_validate == NULL ? "NULL" : arp_validate);
4729                         return -EINVAL;
4730                 }
4731
4732                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4733                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4734         } else {
4735                 fail_over_mac_value = BOND_FOM_NONE;
4736         }
4737
4738         /* fill params struct with the proper values */
4739         params->mode = bond_mode;
4740         params->xmit_policy = xmit_hashtype;
4741         params->miimon = miimon;
4742         params->num_peer_notif = num_peer_notif;
4743         params->arp_interval = arp_interval;
4744         params->arp_validate = arp_validate_value;
4745         params->updelay = updelay;
4746         params->downdelay = downdelay;
4747         params->use_carrier = use_carrier;
4748         params->lacp_fast = lacp_fast;
4749         params->primary[0] = 0;
4750         params->primary_reselect = primary_reselect_value;
4751         params->fail_over_mac = fail_over_mac_value;
4752         params->tx_queues = tx_queues;
4753         params->all_slaves_active = all_slaves_active;
4754         params->resend_igmp = resend_igmp;
4755         params->min_links = min_links;
4756
4757         if (primary) {
4758                 strncpy(params->primary, primary, IFNAMSIZ);
4759                 params->primary[IFNAMSIZ - 1] = 0;
4760         }
4761
4762         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4763
4764         return 0;
4765 }
4766
4767 static struct lock_class_key bonding_netdev_xmit_lock_key;
4768 static struct lock_class_key bonding_netdev_addr_lock_key;
4769
4770 static void bond_set_lockdep_class_one(struct net_device *dev,
4771                                        struct netdev_queue *txq,
4772                                        void *_unused)
4773 {
4774         lockdep_set_class(&txq->_xmit_lock,
4775                           &bonding_netdev_xmit_lock_key);
4776 }
4777
4778 static void bond_set_lockdep_class(struct net_device *dev)
4779 {
4780         lockdep_set_class(&dev->addr_list_lock,
4781                           &bonding_netdev_addr_lock_key);
4782         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4783 }
4784
4785 /*
4786  * Called from registration process
4787  */
4788 static int bond_init(struct net_device *bond_dev)
4789 {
4790         struct bonding *bond = netdev_priv(bond_dev);
4791         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4792         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4793
4794         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4795
4796         /*
4797          * Initialize locks that may be required during
4798          * en/deslave operations.  All of the bond_open work
4799          * (of which this is part) should really be moved to
4800          * a phase prior to dev_open
4801          */
4802         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4803         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4804
4805         bond->wq = create_singlethread_workqueue(bond_dev->name);
4806         if (!bond->wq)
4807                 return -ENOMEM;
4808
4809         bond_set_lockdep_class(bond_dev);
4810
4811         bond_create_proc_entry(bond);
4812         list_add_tail(&bond->bond_list, &bn->dev_list);
4813
4814         bond_prepare_sysfs_group(bond);
4815
4816         bond_debug_register(bond);
4817
4818         __hw_addr_init(&bond->mc_list);
4819         return 0;
4820 }
4821
4822 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4823 {
4824         if (tb[IFLA_ADDRESS]) {
4825                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4826                         return -EINVAL;
4827                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4828                         return -EADDRNOTAVAIL;
4829         }
4830         return 0;
4831 }
4832
4833 static int bond_get_tx_queues(struct net *net, struct nlattr *tb[],
4834                               unsigned int *num_queues,
4835                               unsigned int *real_num_queues)
4836 {
4837         *num_queues = tx_queues;
4838         return 0;
4839 }
4840
4841 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4842         .kind           = "bond",
4843         .priv_size      = sizeof(struct bonding),
4844         .setup          = bond_setup,
4845         .validate       = bond_validate,
4846         .get_tx_queues  = bond_get_tx_queues,
4847 };
4848
4849 /* Create a new bond based on the specified name and bonding parameters.
4850  * If name is NULL, obtain a suitable "bond%d" name for us.
4851  * Caller must NOT hold rtnl_lock; we need to release it here before we
4852  * set up our sysfs entries.
4853  */
4854 int bond_create(struct net *net, const char *name)
4855 {
4856         struct net_device *bond_dev;
4857         int res;
4858
4859         rtnl_lock();
4860
4861         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4862                                    name ? name : "bond%d",
4863                                    bond_setup, tx_queues);
4864         if (!bond_dev) {
4865                 pr_err("%s: eek! can't alloc netdev!\n", name);
4866                 rtnl_unlock();
4867                 return -ENOMEM;
4868         }
4869
4870         dev_net_set(bond_dev, net);
4871         bond_dev->rtnl_link_ops = &bond_link_ops;
4872
4873         res = register_netdevice(bond_dev);
4874
4875         netif_carrier_off(bond_dev);
4876
4877         rtnl_unlock();
4878         if (res < 0)
4879                 bond_destructor(bond_dev);
4880         return res;
4881 }
4882
4883 static int __net_init bond_net_init(struct net *net)
4884 {
4885         struct bond_net *bn = net_generic(net, bond_net_id);
4886
4887         bn->net = net;
4888         INIT_LIST_HEAD(&bn->dev_list);
4889
4890         bond_create_proc_dir(bn);
4891
4892         return 0;
4893 }
4894
4895 static void __net_exit bond_net_exit(struct net *net)
4896 {
4897         struct bond_net *bn = net_generic(net, bond_net_id);
4898
4899         bond_destroy_proc_dir(bn);
4900 }
4901
4902 static struct pernet_operations bond_net_ops = {
4903         .init = bond_net_init,
4904         .exit = bond_net_exit,
4905         .id   = &bond_net_id,
4906         .size = sizeof(struct bond_net),
4907 };
4908
4909 static int __init bonding_init(void)
4910 {
4911         int i;
4912         int res;
4913
4914         pr_info("%s", bond_version);
4915
4916         res = bond_check_params(&bonding_defaults);
4917         if (res)
4918                 goto out;
4919
4920         res = register_pernet_subsys(&bond_net_ops);
4921         if (res)
4922                 goto out;
4923
4924         res = rtnl_link_register(&bond_link_ops);
4925         if (res)
4926                 goto err_link;
4927
4928         bond_create_debugfs();
4929
4930         for (i = 0; i < max_bonds; i++) {
4931                 res = bond_create(&init_net, NULL);
4932                 if (res)
4933                         goto err;
4934         }
4935
4936         res = bond_create_sysfs();
4937         if (res)
4938                 goto err;
4939
4940         register_netdevice_notifier(&bond_netdev_notifier);
4941         register_inetaddr_notifier(&bond_inetaddr_notifier);
4942 out:
4943         return res;
4944 err:
4945         rtnl_link_unregister(&bond_link_ops);
4946 err_link:
4947         unregister_pernet_subsys(&bond_net_ops);
4948         goto out;
4949
4950 }
4951
4952 static void __exit bonding_exit(void)
4953 {
4954         unregister_netdevice_notifier(&bond_netdev_notifier);
4955         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4956
4957         bond_destroy_sysfs();
4958         bond_destroy_debugfs();
4959
4960         rtnl_link_unregister(&bond_link_ops);
4961         unregister_pernet_subsys(&bond_net_ops);
4962
4963 #ifdef CONFIG_NET_POLL_CONTROLLER
4964         /*
4965          * Make sure we don't have an imbalance on our netpoll blocking
4966          */
4967         WARN_ON(atomic_read(&netpoll_block_tx));
4968 #endif
4969 }
4970
4971 module_init(bonding_init);
4972 module_exit(bonding_exit);
4973 MODULE_LICENSE("GPL");
4974 MODULE_VERSION(DRV_VERSION);
4975 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4976 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4977 MODULE_ALIAS_RTNL_LINK("bond");