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 etool = { .cmd = ETHTOOL_GSET };
 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         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
 569                 return -1;
 570
 571         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 572         if (res < 0)
 573                 return -1;
 574
 575         slave_speed = ethtool_cmd_speed(&etool);
 576         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 577                 return -1;
 578
 579         switch (etool.duplex) {
 580         case DUPLEX_FULL:
 581         case DUPLEX_HALF:
 582                 break;
 583         default:
 584                 return -1;
 585         }
 586
 587         slave->speed = slave_speed;
 588         slave->duplex = etool.duplex;
 589
 590         return 0;
 591 }
 592
 593 /*
 594  * if <dev> supports MII link status reporting, check its link status.
 595  *
 596  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 597  * depending upon the setting of the use_carrier parameter.
 598  *
 599  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 600  * can't tell and just pretend it is), or 0, meaning that the link is
 601  * down.
 602  *
 603  * If reporting is non-zero, instead of faking link up, return -1 if
 604  * both ETHTOOL and MII ioctls fail (meaning the device does not
 605  * support them).  If use_carrier is set, return whatever it says.
 606  * It'd be nice if there was a good way to tell if a driver supports
 607  * netif_carrier, but there really isn't.
 608  */
 609 static int bond_check_dev_link(struct bonding *bond,
 610                                struct net_device *slave_dev, int reporting)
 611 {
 612         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 613         int (*ioctl)(struct net_device *, struct ifreq *, int);
 614         struct ifreq ifr;
 615         struct mii_ioctl_data *mii;
 616
 617         if (!reporting && !netif_running(slave_dev))
 618                 return 0;
 619
 620         if (bond->params.use_carrier)
 621                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 622
 623         /* Try to get link status using Ethtool first. */
 624         if (slave_dev->ethtool_ops) {
 625                 if (slave_dev->ethtool_ops->get_link) {
 626                         u32 link;
 627
 628                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 629
 630                         return link ? BMSR_LSTATUS : 0;
 631                 }
 632         }
 633
 634         /* Ethtool can't be used, fallback to MII ioctls. */
 635         ioctl = slave_ops->ndo_do_ioctl;
 636         if (ioctl) {
 637                 /* TODO: set pointer to correct ioctl on a per team member */
 638                 /*       bases to make this more efficient. that is, once  */
 639                 /*       we determine the correct ioctl, we will always    */
 640                 /*       call it and not the others for that team          */
 641                 /*       member.                                           */
 642
 643                 /*
 644                  * We cannot assume that SIOCGMIIPHY will also read a
 645                  * register; not all network drivers (e.g., e100)
 646                  * support that.
 647                  */
 648
 649                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 650                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 651                 mii = if_mii(&ifr);
 652                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 653                         mii->reg_num = MII_BMSR;
 654                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 655                                 return mii->val_out & BMSR_LSTATUS;
 656                 }
 657         }
 658
 659         /*
 660          * If reporting, report that either there's no dev->do_ioctl,
 661          * or both SIOCGMIIREG and get_link failed (meaning that we
 662          * cannot report link status).  If not reporting, pretend
 663          * we're ok.
 664          */
 665         return reporting ? -1 : BMSR_LSTATUS;
 666 }
 667
 668 /*----------------------------- Multicast list ------------------------------*/
 669
 670 /*
 671  * Push the promiscuity flag down to appropriate slaves
 672  */
 673 static int bond_set_promiscuity(struct bonding *bond, int inc)
 674 {
 675         int err = 0;
 676         if (USES_PRIMARY(bond->params.mode)) {
 677                 /* write lock already acquired */
 678                 if (bond->curr_active_slave) {
 679                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 680                                                   inc);
 681                 }
 682         } else {
 683                 struct slave *slave;
 684                 int i;
 685                 bond_for_each_slave(bond, slave, i) {
 686                         err = dev_set_promiscuity(slave->dev, inc);
 687                         if (err)
 688                                 return err;
 689                 }
 690         }
 691         return err;
 692 }
 693
 694 /*
 695  * Push the allmulti flag down to all slaves
 696  */
 697 static int bond_set_allmulti(struct bonding *bond, int inc)
 698 {
 699         int err = 0;
 700         if (USES_PRIMARY(bond->params.mode)) {
 701                 /* write lock already acquired */
 702                 if (bond->curr_active_slave) {
 703                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 704                                                inc);
 705                 }
 706         } else {
 707                 struct slave *slave;
 708                 int i;
 709                 bond_for_each_slave(bond, slave, i) {
 710                         err = dev_set_allmulti(slave->dev, inc);
 711                         if (err)
 712                                 return err;
 713                 }
 714         }
 715         return err;
 716 }
 717
 718 /*
 719  * Add a Multicast address to slaves
 720  * according to mode
 721  */
 722 static void bond_mc_add(struct bonding *bond, void *addr)
 723 {
 724         if (USES_PRIMARY(bond->params.mode)) {
 725                 /* write lock already acquired */
 726                 if (bond->curr_active_slave)
 727                         dev_mc_add(bond->curr_active_slave->dev, addr);
 728         } else {
 729                 struct slave *slave;
 730                 int i;
 731
 732                 bond_for_each_slave(bond, slave, i)
 733                         dev_mc_add(slave->dev, addr);
 734         }
 735 }
 736
 737 /*
 738  * Remove a multicast address from slave
 739  * according to mode
 740  */
 741 static void bond_mc_del(struct bonding *bond, void *addr)
 742 {
 743         if (USES_PRIMARY(bond->params.mode)) {
 744                 /* write lock already acquired */
 745                 if (bond->curr_active_slave)
 746                         dev_mc_del(bond->curr_active_slave->dev, addr);
 747         } else {
 748                 struct slave *slave;
 749                 int i;
 750                 bond_for_each_slave(bond, slave, i) {
 751                         dev_mc_del(slave->dev, addr);
 752                 }
 753         }
 754 }
 755
 756
 757 static void __bond_resend_igmp_join_requests(struct net_device *dev)
 758 {
 759         struct in_device *in_dev;
 760
 761         rcu_read_lock();
 762         in_dev = __in_dev_get_rcu(dev);
 763         if (in_dev)
 764                 ip_mc_rejoin_groups(in_dev);
 765         rcu_read_unlock();
 766 }
 767
 768 /*
 769  * Retrieve the list of registered multicast addresses for the bonding
 770  * device and retransmit an IGMP JOIN request to the current active
 771  * slave.
 772  */
 773 static void bond_resend_igmp_join_requests(struct bonding *bond)
 774 {
 775         struct net_device *vlan_dev;
 776         struct vlan_entry *vlan;
 777
 778         read_lock(&bond->lock);
 779
 780         if (bond->kill_timers)
 781                 goto out;
 782
 783         /* rejoin all groups on bond device */
 784         __bond_resend_igmp_join_requests(bond->dev);
 785
 786         /* rejoin all groups on vlan devices */
 787         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 788                 rcu_read_lock();
 789                 vlan_dev = __vlan_find_dev_deep(bond->dev,
 790                                                 vlan->vlan_id);
 791                 rcu_read_unlock();
 792                 if (vlan_dev)
 793                         __bond_resend_igmp_join_requests(vlan_dev);
 794         }
 795
 796         if ((--bond->igmp_retrans > 0) && !bond->kill_timers)
 797                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 798 out:
 799         read_unlock(&bond->lock);
 800 }
 801
 802 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 803 {
 804         struct bonding *bond = container_of(work, struct bonding,
 805                                             mcast_work.work);
 806         bond_resend_igmp_join_requests(bond);
 807 }
 808
 809 /*
 810  * flush all members of flush->mc_list from device dev->mc_list
 811  */
 812 static void bond_mc_list_flush(struct net_device *bond_dev,
 813                                struct net_device *slave_dev)
 814 {
 815         struct bonding *bond = netdev_priv(bond_dev);
 816         struct netdev_hw_addr *ha;
 817
 818         netdev_for_each_mc_addr(ha, bond_dev)
 819                 dev_mc_del(slave_dev, ha->addr);
 820
 821         if (bond->params.mode == BOND_MODE_8023AD) {
 822                 /* del lacpdu mc addr from mc list */
 823                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 824
 825                 dev_mc_del(slave_dev, lacpdu_multicast);
 826         }
 827 }
 828
 829 /*--------------------------- Active slave change ---------------------------*/
 830
 831 /*
 832  * Update the mc list and multicast-related flags for the new and
 833  * old active slaves (if any) according to the multicast mode, and
 834  * promiscuous flags unconditionally.
 835  */
 836 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 837                          struct slave *old_active)
 838 {
 839         struct netdev_hw_addr *ha;
 840
 841         if (!USES_PRIMARY(bond->params.mode))
 842                 /* nothing to do -  mc list is already up-to-date on
 843                  * all slaves
 844                  */
 845                 return;
 846
 847         if (old_active) {
 848                 if (bond->dev->flags & IFF_PROMISC)
 849                         dev_set_promiscuity(old_active->dev, -1);
 850
 851                 if (bond->dev->flags & IFF_ALLMULTI)
 852                         dev_set_allmulti(old_active->dev, -1);
 853
 854                 netdev_for_each_mc_addr(ha, bond->dev)
 855                         dev_mc_del(old_active->dev, ha->addr);
 856         }
 857
 858         if (new_active) {
 859                 /* FIXME: Signal errors upstream. */
 860                 if (bond->dev->flags & IFF_PROMISC)
 861                         dev_set_promiscuity(new_active->dev, 1);
 862
 863                 if (bond->dev->flags & IFF_ALLMULTI)
 864                         dev_set_allmulti(new_active->dev, 1);
 865
 866                 netdev_for_each_mc_addr(ha, bond->dev)
 867                         dev_mc_add(new_active->dev, ha->addr);
 868         }
 869 }
 870
 871 /*
 872  * bond_do_fail_over_mac
 873  *
 874  * Perform special MAC address swapping for fail_over_mac settings
 875  *
 876  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 877  */
 878 static void bond_do_fail_over_mac(struct bonding *bond,
 879                                   struct slave *new_active,
 880                                   struct slave *old_active)
 881         __releases(&bond->curr_slave_lock)
 882         __releases(&bond->lock)
 883         __acquires(&bond->lock)
 884         __acquires(&bond->curr_slave_lock)
 885 {
 886         u8 tmp_mac[ETH_ALEN];
 887         struct sockaddr saddr;
 888         int rv;
 889
 890         switch (bond->params.fail_over_mac) {
 891         case BOND_FOM_ACTIVE:
 892                 if (new_active)
 893                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 894                                new_active->dev->addr_len);
 895                 break;
 896         case BOND_FOM_FOLLOW:
 897                 /*
 898                  * if new_active && old_active, swap them
 899                  * if just old_active, do nothing (going to no active slave)
 900                  * if just new_active, set new_active to bond's MAC
 901                  */
 902                 if (!new_active)
 903                         return;
 904
 905                 write_unlock_bh(&bond->curr_slave_lock);
 906                 read_unlock(&bond->lock);
 907
 908                 if (old_active) {
 909                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 910                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 911                                ETH_ALEN);
 912                         saddr.sa_family = new_active->dev->type;
 913                 } else {
 914                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 915                         saddr.sa_family = bond->dev->type;
 916                 }
 917
 918                 rv = dev_set_mac_address(new_active->dev, &saddr);
 919                 if (rv) {
 920                         pr_err("%s: Error %d setting MAC of slave %s\n",
 921                                bond->dev->name, -rv, new_active->dev->name);
 922                         goto out;
 923                 }
 924
 925                 if (!old_active)
 926                         goto out;
 927
 928                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 929                 saddr.sa_family = old_active->dev->type;
 930
 931                 rv = dev_set_mac_address(old_active->dev, &saddr);
 932                 if (rv)
 933                         pr_err("%s: Error %d setting MAC of slave %s\n",
 934                                bond->dev->name, -rv, new_active->dev->name);
 935 out:
 936                 read_lock(&bond->lock);
 937                 write_lock_bh(&bond->curr_slave_lock);
 938                 break;
 939         default:
 940                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 941                        bond->dev->name, bond->params.fail_over_mac);
 942                 break;
 943         }
 944
 945 }
 946
 947 static bool bond_should_change_active(struct bonding *bond)
 948 {
 949         struct slave *prim = bond->primary_slave;
 950         struct slave *curr = bond->curr_active_slave;
 951
 952         if (!prim || !curr || curr->link != BOND_LINK_UP)
 953                 return true;
 954         if (bond->force_primary) {
 955                 bond->force_primary = false;
 956                 return true;
 957         }
 958         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 959             (prim->speed < curr->speed ||
 960              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 961                 return false;
 962         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 963                 return false;
 964         return true;
 965 }
 966
 967 /**
 968  * find_best_interface - select the best available slave to be the active one
 969  * @bond: our bonding struct
 970  *
 971  * Warning: Caller must hold curr_slave_lock for writing.
 972  */
 973 static struct slave *bond_find_best_slave(struct bonding *bond)
 974 {
 975         struct slave *new_active, *old_active;
 976         struct slave *bestslave = NULL;
 977         int mintime = bond->params.updelay;
 978         int i;
 979
 980         new_active = bond->curr_active_slave;
 981
 982         if (!new_active) { /* there were no active slaves left */
 983                 if (bond->slave_cnt > 0)   /* found one slave */
 984                         new_active = bond->first_slave;
 985                 else
 986                         return NULL; /* still no slave, return NULL */
 987         }
 988
 989         if ((bond->primary_slave) &&
 990             bond->primary_slave->link == BOND_LINK_UP &&
 991             bond_should_change_active(bond)) {
 992                 new_active = bond->primary_slave;
 993         }
 994
 995         /* remember where to stop iterating over the slaves */
 996         old_active = new_active;
 997
 998         bond_for_each_slave_from(bond, new_active, i, old_active) {
 999                 if (new_active->link == BOND_LINK_UP) {
1000                         return new_active;
1001                 } else if (new_active->link == BOND_LINK_BACK &&
1002                            IS_UP(new_active->dev)) {
1003                         /* link up, but waiting for stabilization */
1004                         if (new_active->delay < mintime) {
1005                                 mintime = new_active->delay;
1006                                 bestslave = new_active;
1007                         }
1008                 }
1009         }
1010
1011         return bestslave;
1012 }
1013
1014 static bool bond_should_notify_peers(struct bonding *bond)
1015 {
1016         struct slave *slave = bond->curr_active_slave;
1017
1018         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1019                  bond->dev->name, slave ? slave->dev->name : "NULL");
1020
1021         if (!slave || !bond->send_peer_notif ||
1022             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1023                 return false;
1024
1025         bond->send_peer_notif--;
1026         return true;
1027 }
1028
1029 /**
1030  * change_active_interface - change the active slave into the specified one
1031  * @bond: our bonding struct
1032  * @new: the new slave to make the active one
1033  *
1034  * Set the new slave to the bond's settings and unset them on the old
1035  * curr_active_slave.
1036  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1037  *
1038  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1039  * because it is apparently the best available slave we have, even though its
1040  * updelay hasn't timed out yet.
1041  *
1042  * If new_active is not NULL, caller must hold bond->lock for read and
1043  * curr_slave_lock for write_bh.
1044  */
1045 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1046 {
1047         struct slave *old_active = bond->curr_active_slave;
1048
1049         if (old_active == new_active)
1050                 return;
1051
1052         if (new_active) {
1053                 new_active->jiffies = jiffies;
1054
1055                 if (new_active->link == BOND_LINK_BACK) {
1056                         if (USES_PRIMARY(bond->params.mode)) {
1057                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1058                                         bond->dev->name, new_active->dev->name,
1059                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1060                         }
1061
1062                         new_active->delay = 0;
1063                         new_active->link = BOND_LINK_UP;
1064
1065                         if (bond->params.mode == BOND_MODE_8023AD)
1066                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1067
1068                         if (bond_is_lb(bond))
1069                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1070                 } else {
1071                         if (USES_PRIMARY(bond->params.mode)) {
1072                                 pr_info("%s: making interface %s the new active one.\n",
1073                                         bond->dev->name, new_active->dev->name);
1074                         }
1075                 }
1076         }
1077
1078         if (USES_PRIMARY(bond->params.mode))
1079                 bond_mc_swap(bond, new_active, old_active);
1080
1081         if (bond_is_lb(bond)) {
1082                 bond_alb_handle_active_change(bond, new_active);
1083                 if (old_active)
1084                         bond_set_slave_inactive_flags(old_active);
1085                 if (new_active)
1086                         bond_set_slave_active_flags(new_active);
1087         } else {
1088                 bond->curr_active_slave = new_active;
1089         }
1090
1091         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1092                 if (old_active)
1093                         bond_set_slave_inactive_flags(old_active);
1094
1095                 if (new_active) {
1096                         bool should_notify_peers = false;
1097
1098                         bond_set_slave_active_flags(new_active);
1099
1100                         if (bond->params.fail_over_mac)
1101                                 bond_do_fail_over_mac(bond, new_active,
1102                                                       old_active);
1103
1104                         if (netif_running(bond->dev)) {
1105                                 bond->send_peer_notif =
1106                                         bond->params.num_peer_notif;
1107                                 should_notify_peers =
1108                                         bond_should_notify_peers(bond);
1109                         }
1110
1111                         write_unlock_bh(&bond->curr_slave_lock);
1112                         read_unlock(&bond->lock);
1113
1114                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1115                         if (should_notify_peers)
1116                                 netdev_bonding_change(bond->dev,
1117                                                       NETDEV_NOTIFY_PEERS);
1118
1119                         read_lock(&bond->lock);
1120                         write_lock_bh(&bond->curr_slave_lock);
1121                 }
1122         }
1123
1124         /* resend IGMP joins since active slave has changed or
1125          * all were sent on curr_active_slave.
1126          * resend only if bond is brought up with the affected
1127          * bonding modes and the retransmission is enabled */
1128         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1129             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1130              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1131                 bond->igmp_retrans = bond->params.resend_igmp;
1132                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1133         }
1134 }
1135
1136 /**
1137  * bond_select_active_slave - select a new active slave, if needed
1138  * @bond: our bonding struct
1139  *
1140  * This functions should be called when one of the following occurs:
1141  * - The old curr_active_slave has been released or lost its link.
1142  * - The primary_slave has got its link back.
1143  * - A slave has got its link back and there's no old curr_active_slave.
1144  *
1145  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1146  */
1147 void bond_select_active_slave(struct bonding *bond)
1148 {
1149         struct slave *best_slave;
1150         int rv;
1151
1152         best_slave = bond_find_best_slave(bond);
1153         if (best_slave != bond->curr_active_slave) {
1154                 bond_change_active_slave(bond, best_slave);
1155                 rv = bond_set_carrier(bond);
1156                 if (!rv)
1157                         return;
1158
1159                 if (netif_carrier_ok(bond->dev)) {
1160                         pr_info("%s: first active interface up!\n",
1161                                 bond->dev->name);
1162                 } else {
1163                         pr_info("%s: now running without any active interface !\n",
1164                                 bond->dev->name);
1165                 }
1166         }
1167 }
1168
1169 /*--------------------------- slave list handling ---------------------------*/
1170
1171 /*
1172  * This function attaches the slave to the end of list.
1173  *
1174  * bond->lock held for writing by caller.
1175  */
1176 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1177 {
1178         if (bond->first_slave == NULL) { /* attaching the first slave */
1179                 new_slave->next = new_slave;
1180                 new_slave->prev = new_slave;
1181                 bond->first_slave = new_slave;
1182         } else {
1183                 new_slave->next = bond->first_slave;
1184                 new_slave->prev = bond->first_slave->prev;
1185                 new_slave->next->prev = new_slave;
1186                 new_slave->prev->next = new_slave;
1187         }
1188
1189         bond->slave_cnt++;
1190 }
1191
1192 /*
1193  * This function detaches the slave from the list.
1194  * WARNING: no check is made to verify if the slave effectively
1195  * belongs to <bond>.
1196  * Nothing is freed on return, structures are just unchained.
1197  * If any slave pointer in bond was pointing to <slave>,
1198  * it should be changed by the calling function.
1199  *
1200  * bond->lock held for writing by caller.
1201  */
1202 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1203 {
1204         if (slave->next)
1205                 slave->next->prev = slave->prev;
1206
1207         if (slave->prev)
1208                 slave->prev->next = slave->next;
1209
1210         if (bond->first_slave == slave) { /* slave is the first slave */
1211                 if (bond->slave_cnt > 1) { /* there are more slave */
1212                         bond->first_slave = slave->next;
1213                 } else {
1214                         bond->first_slave = NULL; /* slave was the last one */
1215                 }
1216         }
1217
1218         slave->next = NULL;
1219         slave->prev = NULL;
1220         bond->slave_cnt--;
1221 }
1222
1223 #ifdef CONFIG_NET_POLL_CONTROLLER
1224 static inline int slave_enable_netpoll(struct slave *slave)
1225 {
1226         struct netpoll *np;
1227         int err = 0;
1228
1229         np = kzalloc(sizeof(*np), GFP_KERNEL);
1230         err = -ENOMEM;
1231         if (!np)
1232                 goto out;
1233
1234         np->dev = slave->dev;
1235         strlcpy(np->dev_name, slave->dev->name, IFNAMSIZ);
1236         err = __netpoll_setup(np);
1237         if (err) {
1238                 kfree(np);
1239                 goto out;
1240         }
1241         slave->np = np;
1242 out:
1243         return err;
1244 }
1245 static inline void slave_disable_netpoll(struct slave *slave)
1246 {
1247         struct netpoll *np = slave->np;
1248
1249         if (!np)
1250                 return;
1251
1252         slave->np = NULL;
1253         synchronize_rcu_bh();
1254         __netpoll_cleanup(np);
1255         kfree(np);
1256 }
1257 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1258 {
1259         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1260                 return false;
1261         if (!slave_dev->netdev_ops->ndo_poll_controller)
1262                 return false;
1263         return true;
1264 }
1265
1266 static void bond_poll_controller(struct net_device *bond_dev)
1267 {
1268 }
1269
1270 static void __bond_netpoll_cleanup(struct bonding *bond)
1271 {
1272         struct slave *slave;
1273         int i;
1274
1275         bond_for_each_slave(bond, slave, i)
1276                 if (IS_UP(slave->dev))
1277                         slave_disable_netpoll(slave);
1278 }
1279 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1280 {
1281         struct bonding *bond = netdev_priv(bond_dev);
1282
1283         read_lock(&bond->lock);
1284         __bond_netpoll_cleanup(bond);
1285         read_unlock(&bond->lock);
1286 }
1287
1288 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1289 {
1290         struct bonding *bond = netdev_priv(dev);
1291         struct slave *slave;
1292         int i, err = 0;
1293
1294         read_lock(&bond->lock);
1295         bond_for_each_slave(bond, slave, i) {
1296                 err = slave_enable_netpoll(slave);
1297                 if (err) {
1298                         __bond_netpoll_cleanup(bond);
1299                         break;
1300                 }
1301         }
1302         read_unlock(&bond->lock);
1303         return err;
1304 }
1305
1306 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1307 {
1308         return bond->dev->npinfo;
1309 }
1310
1311 #else
1312 static inline int slave_enable_netpoll(struct slave *slave)
1313 {
1314         return 0;
1315 }
1316 static inline void slave_disable_netpoll(struct slave *slave)
1317 {
1318 }
1319 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1320 {
1321 }
1322 #endif
1323
1324 /*---------------------------------- IOCTL ----------------------------------*/
1325
1326 static int bond_sethwaddr(struct net_device *bond_dev,
1327                           struct net_device *slave_dev)
1328 {
1329         pr_debug("bond_dev=%p\n", bond_dev);
1330         pr_debug("slave_dev=%p\n", slave_dev);
1331         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1332         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1333         return 0;
1334 }
1335
1336 static u32 bond_fix_features(struct net_device *dev, u32 features)
1337 {
1338         struct slave *slave;
1339         struct bonding *bond = netdev_priv(dev);
1340         u32 mask;
1341         int i;
1342
1343         read_lock(&bond->lock);
1344
1345         if (!bond->first_slave) {
1346                 /* Disable adding VLANs to empty bond. But why? --mq */
1347                 features |= NETIF_F_VLAN_CHALLENGED;
1348                 goto out;
1349         }
1350
1351         mask = features;
1352         features &= ~NETIF_F_ONE_FOR_ALL;
1353         features |= NETIF_F_ALL_FOR_ALL;
1354
1355         bond_for_each_slave(bond, slave, i) {
1356                 features = netdev_increment_features(features,
1357                                                      slave->dev->features,
1358                                                      mask);
1359         }
1360
1361 out:
1362         read_unlock(&bond->lock);
1363         return features;
1364 }
1365
1366 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1367                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1368                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1369
1370 static void bond_compute_features(struct bonding *bond)
1371 {
1372         struct slave *slave;
1373         struct net_device *bond_dev = bond->dev;
1374         u32 vlan_features = BOND_VLAN_FEATURES;
1375         unsigned short max_hard_header_len = ETH_HLEN;
1376         int i;
1377
1378         read_lock(&bond->lock);
1379
1380         if (!bond->first_slave)
1381                 goto done;
1382
1383         bond_for_each_slave(bond, slave, i) {
1384                 vlan_features = netdev_increment_features(vlan_features,
1385                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1386
1387                 if (slave->dev->hard_header_len > max_hard_header_len)
1388                         max_hard_header_len = slave->dev->hard_header_len;
1389         }
1390
1391 done:
1392         bond_dev->vlan_features = vlan_features;
1393         bond_dev->hard_header_len = max_hard_header_len;
1394
1395         read_unlock(&bond->lock);
1396
1397         netdev_change_features(bond_dev);
1398 }
1399
1400 static void bond_setup_by_slave(struct net_device *bond_dev,
1401                                 struct net_device *slave_dev)
1402 {
1403         struct bonding *bond = netdev_priv(bond_dev);
1404
1405         bond_dev->header_ops        = slave_dev->header_ops;
1406
1407         bond_dev->type              = slave_dev->type;
1408         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1409         bond_dev->addr_len          = slave_dev->addr_len;
1410
1411         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1412                 slave_dev->addr_len);
1413         bond->setup_by_slave = 1;
1414 }
1415
1416 /* On bonding slaves other than the currently active slave, suppress
1417  * duplicates except for alb non-mcast/bcast.
1418  */
1419 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1420                                             struct slave *slave,
1421                                             struct bonding *bond)
1422 {
1423         if (bond_is_slave_inactive(slave)) {
1424                 if (bond->params.mode == BOND_MODE_ALB &&
1425                     skb->pkt_type != PACKET_BROADCAST &&
1426                     skb->pkt_type != PACKET_MULTICAST)
1427                         return false;
1428                 return true;
1429         }
1430         return false;
1431 }
1432
1433 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1434 {
1435         struct sk_buff *skb = *pskb;
1436         struct slave *slave;
1437         struct bonding *bond;
1438         void (*recv_probe)(struct sk_buff *, struct bonding *,
1439                                 struct slave *);
1440
1441         skb = skb_share_check(skb, GFP_ATOMIC);
1442         if (unlikely(!skb))
1443                 return RX_HANDLER_CONSUMED;
1444
1445         *pskb = skb;
1446
1447         slave = bond_slave_get_rcu(skb->dev);
1448         bond = slave->bond;
1449
1450         if (bond->params.arp_interval)
1451                 slave->dev->last_rx = jiffies;
1452
1453         recv_probe = ACCESS_ONCE(bond->recv_probe);
1454         if (recv_probe) {
1455                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1456
1457                 if (likely(nskb)) {
1458                         recv_probe(nskb, bond, slave);
1459                         dev_kfree_skb(nskb);
1460                 }
1461         }
1462
1463         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1464                 return RX_HANDLER_EXACT;
1465         }
1466
1467         skb->dev = bond->dev;
1468
1469         if (bond->params.mode == BOND_MODE_ALB &&
1470             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1471             skb->pkt_type == PACKET_HOST) {
1472
1473                 if (unlikely(skb_cow_head(skb,
1474                                           skb->data - skb_mac_header(skb)))) {
1475                         kfree_skb(skb);
1476                         return RX_HANDLER_CONSUMED;
1477                 }
1478                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1479         }
1480
1481         return RX_HANDLER_ANOTHER;
1482 }
1483
1484 /* enslave device <slave> to bond device <master> */
1485 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1486 {
1487         struct bonding *bond = netdev_priv(bond_dev);
1488         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1489         struct slave *new_slave = NULL;
1490         struct netdev_hw_addr *ha;
1491         struct sockaddr addr;
1492         int link_reporting;
1493         int res = 0;
1494
1495         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1496                 slave_ops->ndo_do_ioctl == NULL) {
1497                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1498                            bond_dev->name, slave_dev->name);
1499         }
1500
1501         /* already enslaved */
1502         if (slave_dev->flags & IFF_SLAVE) {
1503                 pr_debug("Error, Device was already enslaved\n");
1504                 return -EBUSY;
1505         }
1506
1507         /* vlan challenged mutual exclusion */
1508         /* no need to lock since we're protected by rtnl_lock */
1509         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1510                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1511                 if (bond_vlan_used(bond)) {
1512                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1513                                bond_dev->name, slave_dev->name, bond_dev->name);
1514                         return -EPERM;
1515                 } else {
1516                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1517                                    bond_dev->name, slave_dev->name,
1518                                    slave_dev->name, bond_dev->name);
1519                 }
1520         } else {
1521                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1522         }
1523
1524         /*
1525          * Old ifenslave binaries are no longer supported.  These can
1526          * be identified with moderate accuracy by the state of the slave:
1527          * the current ifenslave will set the interface down prior to
1528          * enslaving it; the old ifenslave will not.
1529          */
1530         if ((slave_dev->flags & IFF_UP)) {
1531                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1532                        slave_dev->name);
1533                 res = -EPERM;
1534                 goto err_undo_flags;
1535         }
1536
1537         /* set bonding device ether type by slave - bonding netdevices are
1538          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1539          * there is a need to override some of the type dependent attribs/funcs.
1540          *
1541          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1542          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1543          */
1544         if (bond->slave_cnt == 0) {
1545                 if (bond_dev->type != slave_dev->type) {
1546                         pr_debug("%s: change device type from %d to %d\n",
1547                                  bond_dev->name,
1548                                  bond_dev->type, slave_dev->type);
1549
1550                         res = netdev_bonding_change(bond_dev,
1551                                                     NETDEV_PRE_TYPE_CHANGE);
1552                         res = notifier_to_errno(res);
1553                         if (res) {
1554                                 pr_err("%s: refused to change device type\n",
1555                                        bond_dev->name);
1556                                 res = -EBUSY;
1557                                 goto err_undo_flags;
1558                         }
1559
1560                         /* Flush unicast and multicast addresses */
1561                         dev_uc_flush(bond_dev);
1562                         dev_mc_flush(bond_dev);
1563
1564                         if (slave_dev->type != ARPHRD_ETHER)
1565                                 bond_setup_by_slave(bond_dev, slave_dev);
1566                         else {
1567                                 ether_setup(bond_dev);
1568                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1569                         }
1570
1571                         netdev_bonding_change(bond_dev,
1572                                               NETDEV_POST_TYPE_CHANGE);
1573                 }
1574         } else if (bond_dev->type != slave_dev->type) {
1575                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1576                        slave_dev->name,
1577                        slave_dev->type, bond_dev->type);
1578                 res = -EINVAL;
1579                 goto err_undo_flags;
1580         }
1581
1582         if (slave_ops->ndo_set_mac_address == NULL) {
1583                 if (bond->slave_cnt == 0) {
1584                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1585                                    bond_dev->name);
1586                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1587                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1588                         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",
1589                                bond_dev->name);
1590                         res = -EOPNOTSUPP;
1591                         goto err_undo_flags;
1592                 }
1593         }
1594
1595         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1596
1597         /* If this is the first slave, then we need to set the master's hardware
1598          * address to be the same as the slave's. */
1599         if (is_zero_ether_addr(bond->dev->dev_addr))
1600                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1601                        slave_dev->addr_len);
1602
1603
1604         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1605         if (!new_slave) {
1606                 res = -ENOMEM;
1607                 goto err_undo_flags;
1608         }
1609
1610         /*
1611          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1612          * is set via sysfs or module option if desired.
1613          */
1614         new_slave->queue_id = 0;
1615
1616         /* Save slave's original mtu and then set it to match the bond */
1617         new_slave->original_mtu = slave_dev->mtu;
1618         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1619         if (res) {
1620                 pr_debug("Error %d calling dev_set_mtu\n", res);
1621                 goto err_free;
1622         }
1623
1624         /*
1625          * Save slave's original ("permanent") mac address for modes
1626          * that need it, and for restoring it upon release, and then
1627          * set it to the master's address
1628          */
1629         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1630
1631         if (!bond->params.fail_over_mac) {
1632                 /*
1633                  * Set slave to master's mac address.  The application already
1634                  * set the master's mac address to that of the first slave
1635                  */
1636                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1637                 addr.sa_family = slave_dev->type;
1638                 res = dev_set_mac_address(slave_dev, &addr);
1639                 if (res) {
1640                         pr_debug("Error %d calling set_mac_address\n", res);
1641                         goto err_restore_mtu;
1642                 }
1643         }
1644
1645         res = netdev_set_bond_master(slave_dev, bond_dev);
1646         if (res) {
1647                 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1648                 goto err_restore_mac;
1649         }
1650
1651         /* open the slave since the application closed it */
1652         res = dev_open(slave_dev);
1653         if (res) {
1654                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1655                 goto err_unset_master;
1656         }
1657
1658         new_slave->bond = bond;
1659         new_slave->dev = slave_dev;
1660         slave_dev->priv_flags |= IFF_BONDING;
1661
1662         if (bond_is_lb(bond)) {
1663                 /* bond_alb_init_slave() must be called before all other stages since
1664                  * it might fail and we do not want to have to undo everything
1665                  */
1666                 res = bond_alb_init_slave(bond, new_slave);
1667                 if (res)
1668                         goto err_close;
1669         }
1670
1671         /* If the mode USES_PRIMARY, then the new slave gets the
1672          * master's promisc (and mc) settings only if it becomes the
1673          * curr_active_slave, and that is taken care of later when calling
1674          * bond_change_active()
1675          */
1676         if (!USES_PRIMARY(bond->params.mode)) {
1677                 /* set promiscuity level to new slave */
1678                 if (bond_dev->flags & IFF_PROMISC) {
1679                         res = dev_set_promiscuity(slave_dev, 1);
1680                         if (res)
1681                                 goto err_close;
1682                 }
1683
1684                 /* set allmulti level to new slave */
1685                 if (bond_dev->flags & IFF_ALLMULTI) {
1686                         res = dev_set_allmulti(slave_dev, 1);
1687                         if (res)
1688                                 goto err_close;
1689                 }
1690
1691                 netif_addr_lock_bh(bond_dev);
1692                 /* upload master's mc_list to new slave */
1693                 netdev_for_each_mc_addr(ha, bond_dev)
1694                         dev_mc_add(slave_dev, ha->addr);
1695                 netif_addr_unlock_bh(bond_dev);
1696         }
1697
1698         if (bond->params.mode == BOND_MODE_8023AD) {
1699                 /* add lacpdu mc addr to mc list */
1700                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1701
1702                 dev_mc_add(slave_dev, lacpdu_multicast);
1703         }
1704
1705         bond_add_vlans_on_slave(bond, slave_dev);
1706
1707         write_lock_bh(&bond->lock);
1708
1709         bond_attach_slave(bond, new_slave);
1710
1711         new_slave->delay = 0;
1712         new_slave->link_failure_count = 0;
1713
1714         write_unlock_bh(&bond->lock);
1715
1716         bond_compute_features(bond);
1717
1718         read_lock(&bond->lock);
1719
1720         new_slave->last_arp_rx = jiffies;
1721
1722         if (bond->params.miimon && !bond->params.use_carrier) {
1723                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1724
1725                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1726                         /*
1727                          * miimon is set but a bonded network driver
1728                          * does not support ETHTOOL/MII and
1729                          * arp_interval is not set.  Note: if
1730                          * use_carrier is enabled, we will never go
1731                          * here (because netif_carrier is always
1732                          * supported); thus, we don't need to change
1733                          * the messages for netif_carrier.
1734                          */
1735                         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",
1736                                bond_dev->name, slave_dev->name);
1737                 } else if (link_reporting == -1) {
1738                         /* unable get link status using mii/ethtool */
1739                         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",
1740                                    bond_dev->name, slave_dev->name);
1741                 }
1742         }
1743
1744         /* check for initial state */
1745         if (!bond->params.miimon ||
1746             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1747                 if (bond->params.updelay) {
1748                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1749                         new_slave->link  = BOND_LINK_BACK;
1750                         new_slave->delay = bond->params.updelay;
1751                 } else {
1752                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1753                         new_slave->link  = BOND_LINK_UP;
1754                 }
1755                 new_slave->jiffies = jiffies;
1756         } else {
1757                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1758                 new_slave->link  = BOND_LINK_DOWN;
1759         }
1760
1761         if (bond_update_speed_duplex(new_slave) &&
1762             (new_slave->link != BOND_LINK_DOWN)) {
1763                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1764                            bond_dev->name, new_slave->dev->name);
1765
1766                 if (bond->params.mode == BOND_MODE_8023AD) {
1767                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1768                                    bond_dev->name);
1769                 }
1770         }
1771
1772         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1773                 /* if there is a primary slave, remember it */
1774                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1775                         bond->primary_slave = new_slave;
1776                         bond->force_primary = true;
1777                 }
1778         }
1779
1780         write_lock_bh(&bond->curr_slave_lock);
1781
1782         switch (bond->params.mode) {
1783         case BOND_MODE_ACTIVEBACKUP:
1784                 bond_set_slave_inactive_flags(new_slave);
1785                 bond_select_active_slave(bond);
1786                 break;
1787         case BOND_MODE_8023AD:
1788                 /* in 802.3ad mode, the internal mechanism
1789                  * will activate the slaves in the selected
1790                  * aggregator
1791                  */
1792                 bond_set_slave_inactive_flags(new_slave);
1793                 /* if this is the first slave */
1794                 if (bond->slave_cnt == 1) {
1795                         SLAVE_AD_INFO(new_slave).id = 1;
1796                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1797                          * can be called only after the mac address of the bond is set
1798                          */
1799                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1800                 } else {
1801                         SLAVE_AD_INFO(new_slave).id =
1802                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1803                 }
1804
1805                 bond_3ad_bind_slave(new_slave);
1806                 break;
1807         case BOND_MODE_TLB:
1808         case BOND_MODE_ALB:
1809                 bond_set_active_slave(new_slave);
1810                 bond_set_slave_inactive_flags(new_slave);
1811                 bond_select_active_slave(bond);
1812                 break;
1813         default:
1814                 pr_debug("This slave is always active in trunk mode\n");
1815
1816                 /* always active in trunk mode */
1817                 bond_set_active_slave(new_slave);
1818
1819                 /* In trunking mode there is little meaning to curr_active_slave
1820                  * anyway (it holds no special properties of the bond device),
1821                  * so we can change it without calling change_active_interface()
1822                  */
1823                 if (!bond->curr_active_slave)
1824                         bond->curr_active_slave = new_slave;
1825
1826                 break;
1827         } /* switch(bond_mode) */
1828
1829         write_unlock_bh(&bond->curr_slave_lock);
1830
1831         bond_set_carrier(bond);
1832
1833 #ifdef CONFIG_NET_POLL_CONTROLLER
1834         slave_dev->npinfo = bond_netpoll_info(bond);
1835         if (slave_dev->npinfo) {
1836                 if (slave_enable_netpoll(new_slave)) {
1837                         read_unlock(&bond->lock);
1838                         pr_info("Error, %s: master_dev is using netpoll, "
1839                                  "but new slave device does not support netpoll.\n",
1840                                  bond_dev->name);
1841                         res = -EBUSY;
1842                         goto err_detach;
1843                 }
1844         }
1845 #endif
1846
1847         read_unlock(&bond->lock);
1848
1849         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1850         if (res)
1851                 goto err_detach;
1852
1853         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1854                                          new_slave);
1855         if (res) {
1856                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1857                 goto err_dest_symlinks;
1858         }
1859
1860         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1861                 bond_dev->name, slave_dev->name,
1862                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1863                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1864
1865         /* enslave is successful */
1866         return 0;
1867
1868 /* Undo stages on error */
1869 err_dest_symlinks:
1870         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1871
1872 err_detach:
1873         write_lock_bh(&bond->lock);
1874         bond_detach_slave(bond, new_slave);
1875         write_unlock_bh(&bond->lock);
1876
1877 err_close:
1878         dev_close(slave_dev);
1879
1880 err_unset_master:
1881         netdev_set_bond_master(slave_dev, NULL);
1882
1883 err_restore_mac:
1884         if (!bond->params.fail_over_mac) {
1885                 /* XXX TODO - fom follow mode needs to change master's
1886                  * MAC if this slave's MAC is in use by the bond, or at
1887                  * least print a warning.
1888                  */
1889                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1890                 addr.sa_family = slave_dev->type;
1891                 dev_set_mac_address(slave_dev, &addr);
1892         }
1893
1894 err_restore_mtu:
1895         dev_set_mtu(slave_dev, new_slave->original_mtu);
1896
1897 err_free:
1898         kfree(new_slave);
1899
1900 err_undo_flags:
1901         bond_compute_features(bond);
1902
1903         return res;
1904 }
1905
1906 /*
1907  * Try to release the slave device <slave> from the bond device <master>
1908  * It is legal to access curr_active_slave without a lock because all the function
1909  * is write-locked.
1910  *
1911  * The rules for slave state should be:
1912  *   for Active/Backup:
1913  *     Active stays on all backups go down
1914  *   for Bonded connections:
1915  *     The first up interface should be left on and all others downed.
1916  */
1917 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1918 {
1919         struct bonding *bond = netdev_priv(bond_dev);
1920         struct slave *slave, *oldcurrent;
1921         struct sockaddr addr;
1922         u32 old_features = bond_dev->features;
1923
1924         /* slave is not a slave or master is not master of this slave */
1925         if (!(slave_dev->flags & IFF_SLAVE) ||
1926             (slave_dev->master != bond_dev)) {
1927                 pr_err("%s: Error: cannot release %s.\n",
1928                        bond_dev->name, slave_dev->name);
1929                 return -EINVAL;
1930         }
1931
1932         block_netpoll_tx();
1933         netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1934         write_lock_bh(&bond->lock);
1935
1936         slave = bond_get_slave_by_dev(bond, slave_dev);
1937         if (!slave) {
1938                 /* not a slave of this bond */
1939                 pr_info("%s: %s not enslaved\n",
1940                         bond_dev->name, slave_dev->name);
1941                 write_unlock_bh(&bond->lock);
1942                 unblock_netpoll_tx();
1943                 return -EINVAL;
1944         }
1945
1946         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1947          * for this slave anymore.
1948          */
1949         netdev_rx_handler_unregister(slave_dev);
1950         write_unlock_bh(&bond->lock);
1951         synchronize_net();
1952         write_lock_bh(&bond->lock);
1953
1954         if (!bond->params.fail_over_mac) {
1955                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1956                     bond->slave_cnt > 1)
1957                         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",
1958                                    bond_dev->name, slave_dev->name,
1959                                    slave->perm_hwaddr,
1960                                    bond_dev->name, slave_dev->name);
1961         }
1962
1963         /* Inform AD package of unbinding of slave. */
1964         if (bond->params.mode == BOND_MODE_8023AD) {
1965                 /* must be called before the slave is
1966                  * detached from the list
1967                  */
1968                 bond_3ad_unbind_slave(slave);
1969         }
1970
1971         pr_info("%s: releasing %s interface %s\n",
1972                 bond_dev->name,
1973                 bond_is_active_slave(slave) ? "active" : "backup",
1974                 slave_dev->name);
1975
1976         oldcurrent = bond->curr_active_slave;
1977
1978         bond->current_arp_slave = NULL;
1979
1980         /* release the slave from its bond */
1981         bond_detach_slave(bond, slave);
1982
1983         if (bond->primary_slave == slave)
1984                 bond->primary_slave = NULL;
1985
1986         if (oldcurrent == slave)
1987                 bond_change_active_slave(bond, NULL);
1988
1989         if (bond_is_lb(bond)) {
1990                 /* Must be called only after the slave has been
1991                  * detached from the list and the curr_active_slave
1992                  * has been cleared (if our_slave == old_current),
1993                  * but before a new active slave is selected.
1994                  */
1995                 write_unlock_bh(&bond->lock);
1996                 bond_alb_deinit_slave(bond, slave);
1997                 write_lock_bh(&bond->lock);
1998         }
1999
2000         if (oldcurrent == slave) {
2001                 /*
2002                  * Note that we hold RTNL over this sequence, so there
2003                  * is no concern that another slave add/remove event
2004                  * will interfere.
2005                  */
2006                 write_unlock_bh(&bond->lock);
2007                 read_lock(&bond->lock);
2008                 write_lock_bh(&bond->curr_slave_lock);
2009
2010                 bond_select_active_slave(bond);
2011
2012                 write_unlock_bh(&bond->curr_slave_lock);
2013                 read_unlock(&bond->lock);
2014                 write_lock_bh(&bond->lock);
2015         }
2016
2017         if (bond->slave_cnt == 0) {
2018                 bond_set_carrier(bond);
2019
2020                 /* if the last slave was removed, zero the mac address
2021                  * of the master so it will be set by the application
2022                  * to the mac address of the first slave
2023                  */
2024                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2025
2026                 if (bond_vlan_used(bond)) {
2027                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2028                                    bond_dev->name, bond_dev->name);
2029                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2030                                    bond_dev->name);
2031                 }
2032         }
2033
2034         write_unlock_bh(&bond->lock);
2035         unblock_netpoll_tx();
2036
2037         bond_compute_features(bond);
2038         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2039             (old_features & NETIF_F_VLAN_CHALLENGED))
2040                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2041                         bond_dev->name, slave_dev->name, bond_dev->name);
2042
2043         /* must do this from outside any spinlocks */
2044         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2045
2046         bond_del_vlans_from_slave(bond, slave_dev);
2047
2048         /* If the mode USES_PRIMARY, then we should only remove its
2049          * promisc and mc settings if it was the curr_active_slave, but that was
2050          * already taken care of above when we detached the slave
2051          */
2052         if (!USES_PRIMARY(bond->params.mode)) {
2053                 /* unset promiscuity level from slave */
2054                 if (bond_dev->flags & IFF_PROMISC)
2055                         dev_set_promiscuity(slave_dev, -1);
2056
2057                 /* unset allmulti level from slave */
2058                 if (bond_dev->flags & IFF_ALLMULTI)
2059                         dev_set_allmulti(slave_dev, -1);
2060
2061                 /* flush master's mc_list from slave */
2062                 netif_addr_lock_bh(bond_dev);
2063                 bond_mc_list_flush(bond_dev, slave_dev);
2064                 netif_addr_unlock_bh(bond_dev);
2065         }
2066
2067         netdev_set_bond_master(slave_dev, NULL);
2068
2069         slave_disable_netpoll(slave);
2070
2071         /* close slave before restoring its mac address */
2072         dev_close(slave_dev);
2073
2074         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2075                 /* restore original ("permanent") mac address */
2076                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2077                 addr.sa_family = slave_dev->type;
2078                 dev_set_mac_address(slave_dev, &addr);
2079         }
2080
2081         dev_set_mtu(slave_dev, slave->original_mtu);
2082
2083         slave_dev->priv_flags &= ~IFF_BONDING;
2084
2085         kfree(slave);
2086
2087         return 0;  /* deletion OK */
2088 }
2089
2090 /*
2091 * First release a slave and then destroy the bond if no more slaves are left.
2092 * Must be under rtnl_lock when this function is called.
2093 */
2094 static int  bond_release_and_destroy(struct net_device *bond_dev,
2095                                      struct net_device *slave_dev)
2096 {
2097         struct bonding *bond = netdev_priv(bond_dev);
2098         int ret;
2099
2100         ret = bond_release(bond_dev, slave_dev);
2101         if ((ret == 0) && (bond->slave_cnt == 0)) {
2102                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2103                 pr_info("%s: destroying bond %s.\n",
2104                         bond_dev->name, bond_dev->name);
2105                 unregister_netdevice(bond_dev);
2106         }
2107         return ret;
2108 }
2109
2110 /*
2111  * This function releases all slaves.
2112  */
2113 static int bond_release_all(struct net_device *bond_dev)
2114 {
2115         struct bonding *bond = netdev_priv(bond_dev);
2116         struct slave *slave;
2117         struct net_device *slave_dev;
2118         struct sockaddr addr;
2119
2120         write_lock_bh(&bond->lock);
2121
2122         netif_carrier_off(bond_dev);
2123
2124         if (bond->slave_cnt == 0)
2125                 goto out;
2126
2127         bond->current_arp_slave = NULL;
2128         bond->primary_slave = NULL;
2129         bond_change_active_slave(bond, NULL);
2130
2131         while ((slave = bond->first_slave) != NULL) {
2132                 /* Inform AD package of unbinding of slave
2133                  * before slave is detached from the list.
2134                  */
2135                 if (bond->params.mode == BOND_MODE_8023AD)
2136                         bond_3ad_unbind_slave(slave);
2137
2138                 slave_dev = slave->dev;
2139                 bond_detach_slave(bond, slave);
2140
2141                 /* now that the slave is detached, unlock and perform
2142                  * all the undo steps that should not be called from
2143                  * within a lock.
2144                  */
2145                 write_unlock_bh(&bond->lock);
2146
2147                 /* unregister rx_handler early so bond_handle_frame wouldn't
2148                  * be called for this slave anymore.
2149                  */
2150                 netdev_rx_handler_unregister(slave_dev);
2151                 synchronize_net();
2152
2153                 if (bond_is_lb(bond)) {
2154                         /* must be called only after the slave
2155                          * has been detached from the list
2156                          */
2157                         bond_alb_deinit_slave(bond, slave);
2158                 }
2159
2160                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2161                 bond_del_vlans_from_slave(bond, slave_dev);
2162
2163                 /* If the mode USES_PRIMARY, then we should only remove its
2164                  * promisc and mc settings if it was the curr_active_slave, but that was
2165                  * already taken care of above when we detached the slave
2166                  */
2167                 if (!USES_PRIMARY(bond->params.mode)) {
2168                         /* unset promiscuity level from slave */
2169                         if (bond_dev->flags & IFF_PROMISC)
2170                                 dev_set_promiscuity(slave_dev, -1);
2171
2172                         /* unset allmulti level from slave */
2173                         if (bond_dev->flags & IFF_ALLMULTI)
2174                                 dev_set_allmulti(slave_dev, -1);
2175
2176                         /* flush master's mc_list from slave */
2177                         netif_addr_lock_bh(bond_dev);
2178                         bond_mc_list_flush(bond_dev, slave_dev);
2179                         netif_addr_unlock_bh(bond_dev);
2180                 }
2181
2182                 netdev_set_bond_master(slave_dev, NULL);
2183
2184                 slave_disable_netpoll(slave);
2185
2186                 /* close slave before restoring its mac address */
2187                 dev_close(slave_dev);
2188
2189                 if (!bond->params.fail_over_mac) {
2190                         /* restore original ("permanent") mac address*/
2191                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2192                         addr.sa_family = slave_dev->type;
2193                         dev_set_mac_address(slave_dev, &addr);
2194                 }
2195
2196                 kfree(slave);
2197
2198                 /* re-acquire the lock before getting the next slave */
2199                 write_lock_bh(&bond->lock);
2200         }
2201
2202         /* zero the mac address of the master so it will be
2203          * set by the application to the mac address of the
2204          * first slave
2205          */
2206         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2207
2208         if (bond_vlan_used(bond)) {
2209                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2210                            bond_dev->name, bond_dev->name);
2211                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2212                            bond_dev->name);
2213         }
2214
2215         pr_info("%s: released all slaves\n", bond_dev->name);
2216
2217 out:
2218         write_unlock_bh(&bond->lock);
2219
2220         bond_compute_features(bond);
2221
2222         return 0;
2223 }
2224
2225 /*
2226  * This function changes the active slave to slave <slave_dev>.
2227  * It returns -EINVAL in the following cases.
2228  *  - <slave_dev> is not found in the list.
2229  *  - There is not active slave now.
2230  *  - <slave_dev> is already active.
2231  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2232  *  - <slave_dev> is not running.
2233  * In these cases, this function does nothing.
2234  * In the other cases, current_slave pointer is changed and 0 is returned.
2235  */
2236 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2237 {
2238         struct bonding *bond = netdev_priv(bond_dev);
2239         struct slave *old_active = NULL;
2240         struct slave *new_active = NULL;
2241         int res = 0;
2242
2243         if (!USES_PRIMARY(bond->params.mode))
2244                 return -EINVAL;
2245
2246         /* Verify that master_dev is indeed the master of slave_dev */
2247         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2248                 return -EINVAL;
2249
2250         read_lock(&bond->lock);
2251
2252         read_lock(&bond->curr_slave_lock);
2253         old_active = bond->curr_active_slave;
2254         read_unlock(&bond->curr_slave_lock);
2255
2256         new_active = bond_get_slave_by_dev(bond, slave_dev);
2257
2258         /*
2259          * Changing to the current active: do nothing; return success.
2260          */
2261         if (new_active && (new_active == old_active)) {
2262                 read_unlock(&bond->lock);
2263                 return 0;
2264         }
2265
2266         if ((new_active) &&
2267             (old_active) &&
2268             (new_active->link == BOND_LINK_UP) &&
2269             IS_UP(new_active->dev)) {
2270                 block_netpoll_tx();
2271                 write_lock_bh(&bond->curr_slave_lock);
2272                 bond_change_active_slave(bond, new_active);
2273                 write_unlock_bh(&bond->curr_slave_lock);
2274                 unblock_netpoll_tx();
2275         } else
2276                 res = -EINVAL;
2277
2278         read_unlock(&bond->lock);
2279
2280         return res;
2281 }
2282
2283 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2284 {
2285         struct bonding *bond = netdev_priv(bond_dev);
2286
2287         info->bond_mode = bond->params.mode;
2288         info->miimon = bond->params.miimon;
2289
2290         read_lock(&bond->lock);
2291         info->num_slaves = bond->slave_cnt;
2292         read_unlock(&bond->lock);
2293
2294         return 0;
2295 }
2296
2297 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2298 {
2299         struct bonding *bond = netdev_priv(bond_dev);
2300         struct slave *slave;
2301         int i, res = -ENODEV;
2302
2303         read_lock(&bond->lock);
2304
2305         bond_for_each_slave(bond, slave, i) {
2306                 if (i == (int)info->slave_id) {
2307                         res = 0;
2308                         strcpy(info->slave_name, slave->dev->name);
2309                         info->link = slave->link;
2310                         info->state = bond_slave_state(slave);
2311                         info->link_failure_count = slave->link_failure_count;
2312                         break;
2313                 }
2314         }
2315
2316         read_unlock(&bond->lock);
2317
2318         return res;
2319 }
2320
2321 /*-------------------------------- Monitoring -------------------------------*/
2322
2323
2324 static int bond_miimon_inspect(struct bonding *bond)
2325 {
2326         struct slave *slave;
2327         int i, link_state, commit = 0;
2328         bool ignore_updelay;
2329
2330         ignore_updelay = !bond->curr_active_slave ? true : false;
2331
2332         bond_for_each_slave(bond, slave, i) {
2333                 slave->new_link = BOND_LINK_NOCHANGE;
2334
2335                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2336
2337                 switch (slave->link) {
2338                 case BOND_LINK_UP:
2339                         if (link_state)
2340                                 continue;
2341
2342                         slave->link = BOND_LINK_FAIL;
2343                         slave->delay = bond->params.downdelay;
2344                         if (slave->delay) {
2345                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2346                                         bond->dev->name,
2347                                         (bond->params.mode ==
2348                                          BOND_MODE_ACTIVEBACKUP) ?
2349                                         (bond_is_active_slave(slave) ?
2350                                          "active " : "backup ") : "",
2351                                         slave->dev->name,
2352                                         bond->params.downdelay * bond->params.miimon);
2353                         }
2354                         /*FALLTHRU*/
2355                 case BOND_LINK_FAIL:
2356                         if (link_state) {
2357                                 /*
2358                                  * recovered before downdelay expired
2359                                  */
2360                                 slave->link = BOND_LINK_UP;
2361                                 slave->jiffies = jiffies;
2362                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2363                                         bond->dev->name,
2364                                         (bond->params.downdelay - slave->delay) *
2365                                         bond->params.miimon,
2366                                         slave->dev->name);
2367                                 continue;
2368                         }
2369
2370                         if (slave->delay <= 0) {
2371                                 slave->new_link = BOND_LINK_DOWN;
2372                                 commit++;
2373                                 continue;
2374                         }
2375
2376                         slave->delay--;
2377                         break;
2378
2379                 case BOND_LINK_DOWN:
2380                         if (!link_state)
2381                                 continue;
2382
2383                         slave->link = BOND_LINK_BACK;
2384                         slave->delay = bond->params.updelay;
2385
2386                         if (slave->delay) {
2387                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2388                                         bond->dev->name, slave->dev->name,
2389                                         ignore_updelay ? 0 :
2390                                         bond->params.updelay *
2391                                         bond->params.miimon);
2392                         }
2393                         /*FALLTHRU*/
2394                 case BOND_LINK_BACK:
2395                         if (!link_state) {
2396                                 slave->link = BOND_LINK_DOWN;
2397                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2398                                         bond->dev->name,
2399                                         (bond->params.updelay - slave->delay) *
2400                                         bond->params.miimon,
2401                                         slave->dev->name);
2402
2403                                 continue;
2404                         }
2405
2406                         if (ignore_updelay)
2407                                 slave->delay = 0;
2408
2409                         if (slave->delay <= 0) {
2410                                 slave->new_link = BOND_LINK_UP;
2411                                 commit++;
2412                                 ignore_updelay = false;
2413                                 continue;
2414                         }
2415
2416                         slave->delay--;
2417                         break;
2418                 }
2419         }
2420
2421         return commit;
2422 }
2423
2424 static void bond_miimon_commit(struct bonding *bond)
2425 {
2426         struct slave *slave;
2427         int i;
2428
2429         bond_for_each_slave(bond, slave, i) {
2430                 switch (slave->new_link) {
2431                 case BOND_LINK_NOCHANGE:
2432                         continue;
2433
2434                 case BOND_LINK_UP:
2435                         slave->link = BOND_LINK_UP;
2436                         slave->jiffies = jiffies;
2437
2438                         if (bond->params.mode == BOND_MODE_8023AD) {
2439                                 /* prevent it from being the active one */
2440                                 bond_set_backup_slave(slave);
2441                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2442                                 /* make it immediately active */
2443                                 bond_set_active_slave(slave);
2444                         } else if (slave != bond->primary_slave) {
2445                                 /* prevent it from being the active one */
2446                                 bond_set_backup_slave(slave);
2447                         }
2448
2449                         bond_update_speed_duplex(slave);
2450
2451                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2452                                 bond->dev->name, slave->dev->name,
2453                                 slave->speed, slave->duplex ? "full" : "half");
2454
2455                         /* notify ad that the link status has changed */
2456                         if (bond->params.mode == BOND_MODE_8023AD)
2457                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2458
2459                         if (bond_is_lb(bond))
2460                                 bond_alb_handle_link_change(bond, slave,
2461                                                             BOND_LINK_UP);
2462
2463                         if (!bond->curr_active_slave ||
2464                             (slave == bond->primary_slave))
2465                                 goto do_failover;
2466
2467                         continue;
2468
2469                 case BOND_LINK_DOWN:
2470                         if (slave->link_failure_count < UINT_MAX)
2471                                 slave->link_failure_count++;
2472
2473                         slave->link = BOND_LINK_DOWN;
2474
2475                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2476                             bond->params.mode == BOND_MODE_8023AD)
2477                                 bond_set_slave_inactive_flags(slave);
2478
2479                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2480                                 bond->dev->name, slave->dev->name);
2481
2482                         if (bond->params.mode == BOND_MODE_8023AD)
2483                                 bond_3ad_handle_link_change(slave,
2484                                                             BOND_LINK_DOWN);
2485
2486                         if (bond_is_lb(bond))
2487                                 bond_alb_handle_link_change(bond, slave,
2488                                                             BOND_LINK_DOWN);
2489
2490                         if (slave == bond->curr_active_slave)
2491                                 goto do_failover;
2492
2493                         continue;
2494
2495                 default:
2496                         pr_err("%s: invalid new link %d on slave %s\n",
2497                                bond->dev->name, slave->new_link,
2498                                slave->dev->name);
2499                         slave->new_link = BOND_LINK_NOCHANGE;
2500
2501                         continue;
2502                 }
2503
2504 do_failover:
2505                 ASSERT_RTNL();
2506                 block_netpoll_tx();
2507                 write_lock_bh(&bond->curr_slave_lock);
2508                 bond_select_active_slave(bond);
2509                 write_unlock_bh(&bond->curr_slave_lock);
2510                 unblock_netpoll_tx();
2511         }
2512
2513         bond_set_carrier(bond);
2514 }
2515
2516 /*
2517  * bond_mii_monitor
2518  *
2519  * Really a wrapper that splits the mii monitor into two phases: an
2520  * inspection, then (if inspection indicates something needs to be done)
2521  * an acquisition of appropriate locks followed by a commit phase to
2522  * implement whatever link state changes are indicated.
2523  */
2524 void bond_mii_monitor(struct work_struct *work)
2525 {
2526         struct bonding *bond = container_of(work, struct bonding,
2527                                             mii_work.work);
2528         bool should_notify_peers = false;
2529
2530         read_lock(&bond->lock);
2531         if (bond->kill_timers)
2532                 goto out;
2533
2534         if (bond->slave_cnt == 0)
2535                 goto re_arm;
2536
2537         should_notify_peers = bond_should_notify_peers(bond);
2538
2539         if (bond_miimon_inspect(bond)) {
2540                 read_unlock(&bond->lock);
2541                 rtnl_lock();
2542                 read_lock(&bond->lock);
2543
2544                 bond_miimon_commit(bond);
2545
2546                 read_unlock(&bond->lock);
2547                 rtnl_unlock();  /* might sleep, hold no other locks */
2548                 read_lock(&bond->lock);
2549         }
2550
2551 re_arm:
2552         if (bond->params.miimon && !bond->kill_timers)
2553                 queue_delayed_work(bond->wq, &bond->mii_work,
2554                                    msecs_to_jiffies(bond->params.miimon));
2555 out:
2556         read_unlock(&bond->lock);
2557
2558         if (should_notify_peers) {
2559                 rtnl_lock();
2560                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2561                 rtnl_unlock();
2562         }
2563 }
2564
2565 static __be32 bond_glean_dev_ip(struct net_device *dev)
2566 {
2567         struct in_device *idev;
2568         struct in_ifaddr *ifa;
2569         __be32 addr = 0;
2570
2571         if (!dev)
2572                 return 0;
2573
2574         rcu_read_lock();
2575         idev = __in_dev_get_rcu(dev);
2576         if (!idev)
2577                 goto out;
2578
2579         ifa = idev->ifa_list;
2580         if (!ifa)
2581                 goto out;
2582
2583         addr = ifa->ifa_local;
2584 out:
2585         rcu_read_unlock();
2586         return addr;
2587 }
2588
2589 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2590 {
2591         struct vlan_entry *vlan;
2592
2593         if (ip == bond->master_ip)
2594                 return 1;
2595
2596         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2597                 if (ip == vlan->vlan_ip)
2598                         return 1;
2599         }
2600
2601         return 0;
2602 }
2603
2604 /*
2605  * We go to the (large) trouble of VLAN tagging ARP frames because
2606  * switches in VLAN mode (especially if ports are configured as
2607  * "native" to a VLAN) might not pass non-tagged frames.
2608  */
2609 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2610 {
2611         struct sk_buff *skb;
2612
2613         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2614                  slave_dev->name, dest_ip, src_ip, vlan_id);
2615
2616         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2617                          NULL, slave_dev->dev_addr, NULL);
2618
2619         if (!skb) {
2620                 pr_err("ARP packet allocation failed\n");
2621                 return;
2622         }
2623         if (vlan_id) {
2624                 skb = vlan_put_tag(skb, vlan_id);
2625                 if (!skb) {
2626                         pr_err("failed to insert VLAN tag\n");
2627                         return;
2628                 }
2629         }
2630         arp_xmit(skb);
2631 }
2632
2633
2634 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2635 {
2636         int i, vlan_id;
2637         __be32 *targets = bond->params.arp_targets;
2638         struct vlan_entry *vlan;
2639         struct net_device *vlan_dev;
2640         struct rtable *rt;
2641
2642         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2643                 if (!targets[i])
2644                         break;
2645                 pr_debug("basa: target %x\n", targets[i]);
2646                 if (!bond_vlan_used(bond)) {
2647                         pr_debug("basa: empty vlan: arp_send\n");
2648                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2649                                       bond->master_ip, 0);
2650                         continue;
2651                 }
2652
2653                 /*
2654                  * If VLANs are configured, we do a route lookup to
2655                  * determine which VLAN interface would be used, so we
2656                  * can tag the ARP with the proper VLAN tag.
2657                  */
2658                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2659                                      RTO_ONLINK, 0);
2660                 if (IS_ERR(rt)) {
2661                         if (net_ratelimit()) {
2662                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2663                                            bond->dev->name, &targets[i]);
2664                         }
2665                         continue;
2666                 }
2667
2668                 /*
2669                  * This target is not on a VLAN
2670                  */
2671                 if (rt->dst.dev == bond->dev) {
2672                         ip_rt_put(rt);
2673                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2674                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2675                                       bond->master_ip, 0);
2676                         continue;
2677                 }
2678
2679                 vlan_id = 0;
2680                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2681                         rcu_read_lock();
2682                         vlan_dev = __vlan_find_dev_deep(bond->dev,
2683                                                         vlan->vlan_id);
2684                         rcu_read_unlock();
2685                         if (vlan_dev == rt->dst.dev) {
2686                                 vlan_id = vlan->vlan_id;
2687                                 pr_debug("basa: vlan match on %s %d\n",
2688                                        vlan_dev->name, vlan_id);
2689                                 break;
2690                         }
2691                 }
2692
2693                 if (vlan_id) {
2694                         ip_rt_put(rt);
2695                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2696                                       vlan->vlan_ip, vlan_id);
2697                         continue;
2698                 }
2699
2700                 if (net_ratelimit()) {
2701                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2702                                    bond->dev->name, &targets[i],
2703                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2704                 }
2705                 ip_rt_put(rt);
2706         }
2707 }
2708
2709 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2710 {
2711         int i;
2712         __be32 *targets = bond->params.arp_targets;
2713
2714         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2715                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2716                          &sip, &tip, i, &targets[i],
2717                          bond_has_this_ip(bond, tip));
2718                 if (sip == targets[i]) {
2719                         if (bond_has_this_ip(bond, tip))
2720                                 slave->last_arp_rx = jiffies;
2721                         return;
2722                 }
2723         }
2724 }
2725
2726 static void bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2727                          struct slave *slave)
2728 {
2729         struct arphdr *arp;
2730         unsigned char *arp_ptr;
2731         __be32 sip, tip;
2732
2733         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2734                 return;
2735
2736         read_lock(&bond->lock);
2737
2738         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2739                  bond->dev->name, skb->dev->name);
2740
2741         if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
2742                 goto out_unlock;
2743
2744         arp = arp_hdr(skb);
2745         if (arp->ar_hln != bond->dev->addr_len ||
2746             skb->pkt_type == PACKET_OTHERHOST ||
2747             skb->pkt_type == PACKET_LOOPBACK ||
2748             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2749             arp->ar_pro != htons(ETH_P_IP) ||
2750             arp->ar_pln != 4)
2751                 goto out_unlock;
2752
2753         arp_ptr = (unsigned char *)(arp + 1);
2754         arp_ptr += bond->dev->addr_len;
2755         memcpy(&sip, arp_ptr, 4);
2756         arp_ptr += 4 + bond->dev->addr_len;
2757         memcpy(&tip, arp_ptr, 4);
2758
2759         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2760                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2761                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2762                  &sip, &tip);
2763
2764         /*
2765          * Backup slaves won't see the ARP reply, but do come through
2766          * here for each ARP probe (so we swap the sip/tip to validate
2767          * the probe).  In a "redundant switch, common router" type of
2768          * configuration, the ARP probe will (hopefully) travel from
2769          * the active, through one switch, the router, then the other
2770          * switch before reaching the backup.
2771          */
2772         if (bond_is_active_slave(slave))
2773                 bond_validate_arp(bond, slave, sip, tip);
2774         else
2775                 bond_validate_arp(bond, slave, tip, sip);
2776
2777 out_unlock:
2778         read_unlock(&bond->lock);
2779 }
2780
2781 /*
2782  * this function is called regularly to monitor each slave's link
2783  * ensuring that traffic is being sent and received when arp monitoring
2784  * is used in load-balancing mode. if the adapter has been dormant, then an
2785  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2786  * arp monitoring in active backup mode.
2787  */
2788 void bond_loadbalance_arp_mon(struct work_struct *work)
2789 {
2790         struct bonding *bond = container_of(work, struct bonding,
2791                                             arp_work.work);
2792         struct slave *slave, *oldcurrent;
2793         int do_failover = 0;
2794         int delta_in_ticks;
2795         int i;
2796
2797         read_lock(&bond->lock);
2798
2799         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2800
2801         if (bond->kill_timers)
2802                 goto out;
2803
2804         if (bond->slave_cnt == 0)
2805                 goto re_arm;
2806
2807         read_lock(&bond->curr_slave_lock);
2808         oldcurrent = bond->curr_active_slave;
2809         read_unlock(&bond->curr_slave_lock);
2810
2811         /* see if any of the previous devices are up now (i.e. they have
2812          * xmt and rcv traffic). the curr_active_slave does not come into
2813          * the picture unless it is null. also, slave->jiffies is not needed
2814          * here because we send an arp on each slave and give a slave as
2815          * long as it needs to get the tx/rx within the delta.
2816          * TODO: what about up/down delay in arp mode? it wasn't here before
2817          *       so it can wait
2818          */
2819         bond_for_each_slave(bond, slave, i) {
2820                 unsigned long trans_start = dev_trans_start(slave->dev);
2821
2822                 if (slave->link != BOND_LINK_UP) {
2823                         if (time_in_range(jiffies,
2824                                 trans_start - delta_in_ticks,
2825                                 trans_start + delta_in_ticks) &&
2826                             time_in_range(jiffies,
2827                                 slave->dev->last_rx - delta_in_ticks,
2828                                 slave->dev->last_rx + delta_in_ticks)) {
2829
2830                                 slave->link  = BOND_LINK_UP;
2831                                 bond_set_active_slave(slave);
2832
2833                                 /* primary_slave has no meaning in round-robin
2834                                  * mode. the window of a slave being up and
2835                                  * curr_active_slave being null after enslaving
2836                                  * is closed.
2837                                  */
2838                                 if (!oldcurrent) {
2839                                         pr_info("%s: link status definitely up for interface %s, ",
2840                                                 bond->dev->name,
2841                                                 slave->dev->name);
2842                                         do_failover = 1;
2843                                 } else {
2844                                         pr_info("%s: interface %s is now up\n",
2845                                                 bond->dev->name,
2846                                                 slave->dev->name);
2847                                 }
2848                         }
2849                 } else {
2850                         /* slave->link == BOND_LINK_UP */
2851
2852                         /* not all switches will respond to an arp request
2853                          * when the source ip is 0, so don't take the link down
2854                          * if we don't know our ip yet
2855                          */
2856                         if (!time_in_range(jiffies,
2857                                 trans_start - delta_in_ticks,
2858                                 trans_start + 2 * delta_in_ticks) ||
2859                             !time_in_range(jiffies,
2860                                 slave->dev->last_rx - delta_in_ticks,
2861                                 slave->dev->last_rx + 2 * delta_in_ticks)) {
2862
2863                                 slave->link  = BOND_LINK_DOWN;
2864                                 bond_set_backup_slave(slave);
2865
2866                                 if (slave->link_failure_count < UINT_MAX)
2867                                         slave->link_failure_count++;
2868
2869                                 pr_info("%s: interface %s is now down.\n",
2870                                         bond->dev->name,
2871                                         slave->dev->name);
2872
2873                                 if (slave == oldcurrent)
2874                                         do_failover = 1;
2875                         }
2876                 }
2877
2878                 /* note: if switch is in round-robin mode, all links
2879                  * must tx arp to ensure all links rx an arp - otherwise
2880                  * links may oscillate or not come up at all; if switch is
2881                  * in something like xor mode, there is nothing we can
2882                  * do - all replies will be rx'ed on same link causing slaves
2883                  * to be unstable during low/no traffic periods
2884                  */
2885                 if (IS_UP(slave->dev))
2886                         bond_arp_send_all(bond, slave);
2887         }
2888
2889         if (do_failover) {
2890                 block_netpoll_tx();
2891                 write_lock_bh(&bond->curr_slave_lock);
2892
2893                 bond_select_active_slave(bond);
2894
2895                 write_unlock_bh(&bond->curr_slave_lock);
2896                 unblock_netpoll_tx();
2897         }
2898
2899 re_arm:
2900         if (bond->params.arp_interval && !bond->kill_timers)
2901                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2902 out:
2903         read_unlock(&bond->lock);
2904 }
2905
2906 /*
2907  * Called to inspect slaves for active-backup mode ARP monitor link state
2908  * changes.  Sets new_link in slaves to specify what action should take
2909  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2910  * to link states must be committed.
2911  *
2912  * Called with bond->lock held for read.
2913  */
2914 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2915 {
2916         struct slave *slave;
2917         int i, commit = 0;
2918         unsigned long trans_start;
2919
2920         bond_for_each_slave(bond, slave, i) {
2921                 slave->new_link = BOND_LINK_NOCHANGE;
2922
2923                 if (slave->link != BOND_LINK_UP) {
2924                         if (time_in_range(jiffies,
2925                                 slave_last_rx(bond, slave) - delta_in_ticks,
2926                                 slave_last_rx(bond, slave) + delta_in_ticks)) {
2927
2928                                 slave->new_link = BOND_LINK_UP;
2929                                 commit++;
2930                         }
2931
2932                         continue;
2933                 }
2934
2935                 /*
2936                  * Give slaves 2*delta after being enslaved or made
2937                  * active.  This avoids bouncing, as the last receive
2938                  * times need a full ARP monitor cycle to be updated.
2939                  */
2940                 if (time_in_range(jiffies,
2941                                   slave->jiffies - delta_in_ticks,
2942                                   slave->jiffies + 2 * delta_in_ticks))
2943                         continue;
2944
2945                 /*
2946                  * Backup slave is down if:
2947                  * - No current_arp_slave AND
2948                  * - more than 3*delta since last receive AND
2949                  * - the bond has an IP address
2950                  *
2951                  * Note: a non-null current_arp_slave indicates
2952                  * the curr_active_slave went down and we are
2953                  * searching for a new one; under this condition
2954                  * we only take the curr_active_slave down - this
2955                  * gives each slave a chance to tx/rx traffic
2956                  * before being taken out
2957                  */
2958                 if (!bond_is_active_slave(slave) &&
2959                     !bond->current_arp_slave &&
2960                     !time_in_range(jiffies,
2961                         slave_last_rx(bond, slave) - delta_in_ticks,
2962                         slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2963
2964                         slave->new_link = BOND_LINK_DOWN;
2965                         commit++;
2966                 }
2967
2968                 /*
2969                  * Active slave is down if:
2970                  * - more than 2*delta since transmitting OR
2971                  * - (more than 2*delta since receive AND
2972                  *    the bond has an IP address)
2973                  */
2974                 trans_start = dev_trans_start(slave->dev);
2975                 if (bond_is_active_slave(slave) &&
2976                     (!time_in_range(jiffies,
2977                         trans_start - delta_in_ticks,
2978                         trans_start + 2 * delta_in_ticks) ||
2979                      !time_in_range(jiffies,
2980                         slave_last_rx(bond, slave) - delta_in_ticks,
2981                         slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2982
2983                         slave->new_link = BOND_LINK_DOWN;
2984                         commit++;
2985                 }
2986         }
2987
2988         return commit;
2989 }
2990
2991 /*
2992  * Called to commit link state changes noted by inspection step of
2993  * active-backup mode ARP monitor.
2994  *
2995  * Called with RTNL and bond->lock for read.
2996  */
2997 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2998 {
2999         struct slave *slave;
3000         int i;
3001         unsigned long trans_start;
3002
3003         bond_for_each_slave(bond, slave, i) {
3004                 switch (slave->new_link) {
3005                 case BOND_LINK_NOCHANGE:
3006                         continue;
3007
3008                 case BOND_LINK_UP:
3009                         trans_start = dev_trans_start(slave->dev);
3010                         if ((!bond->curr_active_slave &&
3011                              time_in_range(jiffies,
3012                                            trans_start - delta_in_ticks,
3013                                            trans_start + delta_in_ticks)) ||
3014                             bond->curr_active_slave != slave) {
3015                                 slave->link = BOND_LINK_UP;
3016                                 bond->current_arp_slave = NULL;
3017
3018                                 pr_info("%s: link status definitely up for interface %s.\n",
3019                                         bond->dev->name, slave->dev->name);
3020
3021                                 if (!bond->curr_active_slave ||
3022                                     (slave == bond->primary_slave))
3023                                         goto do_failover;
3024
3025                         }
3026
3027                         continue;
3028
3029                 case BOND_LINK_DOWN:
3030                         if (slave->link_failure_count < UINT_MAX)
3031                                 slave->link_failure_count++;
3032
3033                         slave->link = BOND_LINK_DOWN;
3034                         bond_set_slave_inactive_flags(slave);
3035
3036                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3037                                 bond->dev->name, slave->dev->name);
3038
3039                         if (slave == bond->curr_active_slave) {
3040                                 bond->current_arp_slave = NULL;
3041                                 goto do_failover;
3042                         }
3043
3044                         continue;
3045
3046                 default:
3047                         pr_err("%s: impossible: new_link %d on slave %s\n",
3048                                bond->dev->name, slave->new_link,
3049                                slave->dev->name);
3050                         continue;
3051                 }
3052
3053 do_failover:
3054                 ASSERT_RTNL();
3055                 block_netpoll_tx();
3056                 write_lock_bh(&bond->curr_slave_lock);
3057                 bond_select_active_slave(bond);
3058                 write_unlock_bh(&bond->curr_slave_lock);
3059                 unblock_netpoll_tx();
3060         }
3061
3062         bond_set_carrier(bond);
3063 }
3064
3065 /*
3066  * Send ARP probes for active-backup mode ARP monitor.
3067  *
3068  * Called with bond->lock held for read.
3069  */
3070 static void bond_ab_arp_probe(struct bonding *bond)
3071 {
3072         struct slave *slave;
3073         int i;
3074
3075         read_lock(&bond->curr_slave_lock);
3076
3077         if (bond->current_arp_slave && bond->curr_active_slave)
3078                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3079                         bond->current_arp_slave->dev->name,
3080                         bond->curr_active_slave->dev->name);
3081
3082         if (bond->curr_active_slave) {
3083                 bond_arp_send_all(bond, bond->curr_active_slave);
3084                 read_unlock(&bond->curr_slave_lock);
3085                 return;
3086         }
3087
3088         read_unlock(&bond->curr_slave_lock);
3089
3090         /* if we don't have a curr_active_slave, search for the next available
3091          * backup slave from the current_arp_slave and make it the candidate
3092          * for becoming the curr_active_slave
3093          */
3094
3095         if (!bond->current_arp_slave) {
3096                 bond->current_arp_slave = bond->first_slave;
3097                 if (!bond->current_arp_slave)
3098                         return;
3099         }
3100
3101         bond_set_slave_inactive_flags(bond->current_arp_slave);
3102
3103         /* search for next candidate */
3104         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3105                 if (IS_UP(slave->dev)) {
3106                         slave->link = BOND_LINK_BACK;
3107                         bond_set_slave_active_flags(slave);
3108                         bond_arp_send_all(bond, slave);
3109                         slave->jiffies = jiffies;
3110                         bond->current_arp_slave = slave;
3111                         break;
3112                 }
3113
3114                 /* if the link state is up at this point, we
3115                  * mark it down - this can happen if we have
3116                  * simultaneous link failures and
3117                  * reselect_active_interface doesn't make this
3118                  * one the current slave so it is still marked
3119                  * up when it is actually down
3120                  */
3121                 if (slave->link == BOND_LINK_UP) {
3122                         slave->link = BOND_LINK_DOWN;
3123                         if (slave->link_failure_count < UINT_MAX)
3124                                 slave->link_failure_count++;
3125
3126                         bond_set_slave_inactive_flags(slave);
3127
3128                         pr_info("%s: backup interface %s is now down.\n",
3129                                 bond->dev->name, slave->dev->name);
3130                 }
3131         }
3132 }
3133
3134 void bond_activebackup_arp_mon(struct work_struct *work)
3135 {
3136         struct bonding *bond = container_of(work, struct bonding,
3137                                             arp_work.work);
3138         bool should_notify_peers = false;
3139         int delta_in_ticks;
3140
3141         read_lock(&bond->lock);
3142
3143         if (bond->kill_timers)
3144                 goto out;
3145
3146         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3147
3148         if (bond->slave_cnt == 0)
3149                 goto re_arm;
3150
3151         should_notify_peers = bond_should_notify_peers(bond);
3152
3153         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3154                 read_unlock(&bond->lock);
3155                 rtnl_lock();
3156                 read_lock(&bond->lock);
3157
3158                 bond_ab_arp_commit(bond, delta_in_ticks);
3159
3160                 read_unlock(&bond->lock);
3161                 rtnl_unlock();
3162                 read_lock(&bond->lock);
3163         }
3164
3165         bond_ab_arp_probe(bond);
3166
3167 re_arm:
3168         if (bond->params.arp_interval && !bond->kill_timers)
3169                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3170 out:
3171         read_unlock(&bond->lock);
3172
3173         if (should_notify_peers) {
3174                 rtnl_lock();
3175                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3176                 rtnl_unlock();
3177         }
3178 }
3179
3180 /*-------------------------- netdev event handling --------------------------*/
3181
3182 /*
3183  * Change device name
3184  */
3185 static int bond_event_changename(struct bonding *bond)
3186 {
3187         bond_remove_proc_entry(bond);
3188         bond_create_proc_entry(bond);
3189
3190         bond_debug_reregister(bond);
3191
3192         return NOTIFY_DONE;
3193 }
3194
3195 static int bond_master_netdev_event(unsigned long event,
3196                                     struct net_device *bond_dev)
3197 {
3198         struct bonding *event_bond = netdev_priv(bond_dev);
3199
3200         switch (event) {
3201         case NETDEV_CHANGENAME:
3202                 return bond_event_changename(event_bond);
3203         default:
3204                 break;
3205         }
3206
3207         return NOTIFY_DONE;
3208 }
3209
3210 static int bond_slave_netdev_event(unsigned long event,
3211                                    struct net_device *slave_dev)
3212 {
3213         struct net_device *bond_dev = slave_dev->master;
3214         struct bonding *bond = netdev_priv(bond_dev);
3215
3216         switch (event) {
3217         case NETDEV_UNREGISTER:
3218                 if (bond_dev) {
3219                         if (bond->setup_by_slave)
3220                                 bond_release_and_destroy(bond_dev, slave_dev);
3221                         else
3222                                 bond_release(bond_dev, slave_dev);
3223                 }
3224                 break;
3225         case NETDEV_CHANGE:
3226                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3227                         struct slave *slave;
3228
3229                         slave = bond_get_slave_by_dev(bond, slave_dev);
3230                         if (slave) {
3231                                 u32 old_speed = slave->speed;
3232                                 u8  old_duplex = slave->duplex;
3233
3234                                 bond_update_speed_duplex(slave);
3235
3236                                 if (bond_is_lb(bond))
3237                                         break;
3238
3239                                 if (old_speed != slave->speed)
3240                                         bond_3ad_adapter_speed_changed(slave);
3241                                 if (old_duplex != slave->duplex)
3242                                         bond_3ad_adapter_duplex_changed(slave);
3243                         }
3244                 }
3245
3246                 break;
3247         case NETDEV_DOWN:
3248                 /*
3249                  * ... Or is it this?
3250                  */
3251                 break;
3252         case NETDEV_CHANGEMTU:
3253                 /*
3254                  * TODO: Should slaves be allowed to
3255                  * independently alter their MTU?  For
3256                  * an active-backup bond, slaves need
3257                  * not be the same type of device, so
3258                  * MTUs may vary.  For other modes,
3259                  * slaves arguably should have the
3260                  * same MTUs. To do this, we'd need to
3261                  * take over the slave's change_mtu
3262                  * function for the duration of their
3263                  * servitude.
3264                  */
3265                 break;
3266         case NETDEV_CHANGENAME:
3267                 /*
3268                  * TODO: handle changing the primary's name
3269                  */
3270                 break;
3271         case NETDEV_FEAT_CHANGE:
3272                 bond_compute_features(bond);
3273                 break;
3274         default:
3275                 break;
3276         }
3277
3278         return NOTIFY_DONE;
3279 }
3280
3281 /*
3282  * bond_netdev_event: handle netdev notifier chain events.
3283  *
3284  * This function receives events for the netdev chain.  The caller (an
3285  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3286  * locks for us to safely manipulate the slave devices (RTNL lock,
3287  * dev_probe_lock).
3288  */
3289 static int bond_netdev_event(struct notifier_block *this,
3290                              unsigned long event, void *ptr)
3291 {
3292         struct net_device *event_dev = (struct net_device *)ptr;
3293
3294         pr_debug("event_dev: %s, event: %lx\n",
3295                  event_dev ? event_dev->name : "None",
3296                  event);
3297
3298         if (!(event_dev->priv_flags & IFF_BONDING))
3299                 return NOTIFY_DONE;
3300
3301         if (event_dev->flags & IFF_MASTER) {
3302                 pr_debug("IFF_MASTER\n");
3303                 return bond_master_netdev_event(event, event_dev);
3304         }
3305
3306         if (event_dev->flags & IFF_SLAVE) {
3307                 pr_debug("IFF_SLAVE\n");
3308                 return bond_slave_netdev_event(event, event_dev);
3309         }
3310
3311         return NOTIFY_DONE;
3312 }
3313
3314 /*
3315  * bond_inetaddr_event: handle inetaddr notifier chain events.
3316  *
3317  * We keep track of device IPs primarily to use as source addresses in
3318  * ARP monitor probes (rather than spewing out broadcasts all the time).
3319  *
3320  * We track one IP for the main device (if it has one), plus one per VLAN.
3321  */
3322 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3323 {
3324         struct in_ifaddr *ifa = ptr;
3325         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3326         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3327         struct bonding *bond;
3328         struct vlan_entry *vlan;
3329
3330         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3331                 if (bond->dev == event_dev) {
3332                         switch (event) {
3333                         case NETDEV_UP:
3334                                 bond->master_ip = ifa->ifa_local;
3335                                 return NOTIFY_OK;
3336                         case NETDEV_DOWN:
3337                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3338                                 return NOTIFY_OK;
3339                         default:
3340                                 return NOTIFY_DONE;
3341                         }
3342                 }
3343
3344                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3345                         vlan_dev = __vlan_find_dev_deep(bond->dev,
3346                                                         vlan->vlan_id);
3347                         if (vlan_dev == event_dev) {
3348                                 switch (event) {
3349                                 case NETDEV_UP:
3350                                         vlan->vlan_ip = ifa->ifa_local;
3351                                         return NOTIFY_OK;
3352                                 case NETDEV_DOWN:
3353                                         vlan->vlan_ip =
3354                                                 bond_glean_dev_ip(vlan_dev);
3355                                         return NOTIFY_OK;
3356                                 default:
3357                                         return NOTIFY_DONE;
3358                                 }
3359                         }
3360                 }
3361         }
3362         return NOTIFY_DONE;
3363 }
3364
3365 static struct notifier_block bond_netdev_notifier = {
3366         .notifier_call = bond_netdev_event,
3367 };
3368
3369 static struct notifier_block bond_inetaddr_notifier = {
3370         .notifier_call = bond_inetaddr_event,
3371 };
3372
3373 /*---------------------------- Hashing Policies -----------------------------*/
3374
3375 /*
3376  * Hash for the output device based upon layer 2 and layer 3 data. If
3377  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3378  */
3379 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3380 {
3381         struct ethhdr *data = (struct ethhdr *)skb->data;
3382         struct iphdr *iph = ip_hdr(skb);
3383
3384         if (skb->protocol == htons(ETH_P_IP)) {
3385                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3386                         (data->h_dest[5] ^ data->h_source[5])) % count;
3387         }
3388
3389         return (data->h_dest[5] ^ data->h_source[5]) % count;
3390 }
3391
3392 /*
3393  * Hash for the output device based upon layer 3 and layer 4 data. If
3394  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3395  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3396  */
3397 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3398 {
3399         struct ethhdr *data = (struct ethhdr *)skb->data;
3400         struct iphdr *iph = ip_hdr(skb);
3401         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3402         int layer4_xor = 0;
3403
3404         if (skb->protocol == htons(ETH_P_IP)) {
3405                 if (!ip_is_fragment(iph) &&
3406                     (iph->protocol == IPPROTO_TCP ||
3407                      iph->protocol == IPPROTO_UDP)) {
3408                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3409                 }
3410                 return (layer4_xor ^
3411                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3412
3413         }
3414
3415         return (data->h_dest[5] ^ data->h_source[5]) % count;
3416 }
3417
3418 /*
3419  * Hash for the output device based upon layer 2 data
3420  */
3421 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3422 {
3423         struct ethhdr *data = (struct ethhdr *)skb->data;
3424
3425         return (data->h_dest[5] ^ data->h_source[5]) % count;
3426 }
3427
3428 /*-------------------------- Device entry points ----------------------------*/
3429
3430 static int bond_open(struct net_device *bond_dev)
3431 {
3432         struct bonding *bond = netdev_priv(bond_dev);
3433         struct slave *slave;
3434         int i;
3435
3436         bond->kill_timers = 0;
3437
3438         /* reset slave->backup and slave->inactive */
3439         read_lock(&bond->lock);
3440         if (bond->slave_cnt > 0) {
3441                 read_lock(&bond->curr_slave_lock);
3442                 bond_for_each_slave(bond, slave, i) {
3443                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3444                                 && (slave != bond->curr_active_slave)) {
3445                                 bond_set_slave_inactive_flags(slave);
3446                         } else {
3447                                 bond_set_slave_active_flags(slave);
3448                         }
3449                 }
3450                 read_unlock(&bond->curr_slave_lock);
3451         }
3452         read_unlock(&bond->lock);
3453
3454         INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3455
3456         if (bond_is_lb(bond)) {
3457                 /* bond_alb_initialize must be called before the timer
3458                  * is started.
3459                  */
3460                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3461                         /* something went wrong - fail the open operation */
3462                         return -ENOMEM;
3463                 }
3464
3465                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3466                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3467         }
3468
3469         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3470                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3471                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3472         }
3473
3474         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3475                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3476                         INIT_DELAYED_WORK(&bond->arp_work,
3477                                           bond_activebackup_arp_mon);
3478                 else
3479                         INIT_DELAYED_WORK(&bond->arp_work,
3480                                           bond_loadbalance_arp_mon);
3481
3482                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3483                 if (bond->params.arp_validate)
3484                         bond->recv_probe = bond_arp_rcv;
3485         }
3486
3487         if (bond->params.mode == BOND_MODE_8023AD) {
3488                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3489                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3490                 /* register to receive LACPDUs */
3491                 bond->recv_probe = bond_3ad_lacpdu_recv;
3492                 bond_3ad_initiate_agg_selection(bond, 1);
3493         }
3494
3495         return 0;
3496 }
3497
3498 static int bond_close(struct net_device *bond_dev)
3499 {
3500         struct bonding *bond = netdev_priv(bond_dev);
3501
3502         write_lock_bh(&bond->lock);
3503
3504         bond->send_peer_notif = 0;
3505
3506         /* signal timers not to re-arm */
3507         bond->kill_timers = 1;
3508
3509         write_unlock_bh(&bond->lock);
3510
3511         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3512                 cancel_delayed_work(&bond->mii_work);
3513         }
3514
3515         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3516                 cancel_delayed_work(&bond->arp_work);
3517         }
3518
3519         switch (bond->params.mode) {
3520         case BOND_MODE_8023AD:
3521                 cancel_delayed_work(&bond->ad_work);
3522                 break;
3523         case BOND_MODE_TLB:
3524         case BOND_MODE_ALB:
3525                 cancel_delayed_work(&bond->alb_work);
3526                 break;
3527         default:
3528                 break;
3529         }
3530
3531         if (delayed_work_pending(&bond->mcast_work))
3532                 cancel_delayed_work(&bond->mcast_work);
3533
3534         if (bond_is_lb(bond)) {
3535                 /* Must be called only after all
3536                  * slaves have been released
3537                  */
3538                 bond_alb_deinitialize(bond);
3539         }
3540         bond->recv_probe = NULL;
3541
3542         return 0;
3543 }
3544
3545 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3546                                                 struct rtnl_link_stats64 *stats)
3547 {
3548         struct bonding *bond = netdev_priv(bond_dev);
3549         struct rtnl_link_stats64 temp;
3550         struct slave *slave;
3551         int i;
3552
3553         memset(stats, 0, sizeof(*stats));
3554
3555         read_lock_bh(&bond->lock);
3556
3557         bond_for_each_slave(bond, slave, i) {
3558                 const struct rtnl_link_stats64 *sstats =
3559                         dev_get_stats(slave->dev, &temp);
3560
3561                 stats->rx_packets += sstats->rx_packets;
3562                 stats->rx_bytes += sstats->rx_bytes;
3563                 stats->rx_errors += sstats->rx_errors;
3564                 stats->rx_dropped += sstats->rx_dropped;
3565
3566                 stats->tx_packets += sstats->tx_packets;
3567                 stats->tx_bytes += sstats->tx_bytes;
3568                 stats->tx_errors += sstats->tx_errors;
3569                 stats->tx_dropped += sstats->tx_dropped;
3570
3571                 stats->multicast += sstats->multicast;
3572                 stats->collisions += sstats->collisions;
3573
3574                 stats->rx_length_errors += sstats->rx_length_errors;
3575                 stats->rx_over_errors += sstats->rx_over_errors;
3576                 stats->rx_crc_errors += sstats->rx_crc_errors;
3577                 stats->rx_frame_errors += sstats->rx_frame_errors;
3578                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3579                 stats->rx_missed_errors += sstats->rx_missed_errors;
3580
3581                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3582                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3583                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3584                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3585                 stats->tx_window_errors += sstats->tx_window_errors;
3586         }
3587
3588         read_unlock_bh(&bond->lock);
3589
3590         return stats;
3591 }
3592
3593 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3594 {
3595         struct net_device *slave_dev = NULL;
3596         struct ifbond k_binfo;
3597         struct ifbond __user *u_binfo = NULL;
3598         struct ifslave k_sinfo;
3599         struct ifslave __user *u_sinfo = NULL;
3600         struct mii_ioctl_data *mii = NULL;
3601         int res = 0;
3602
3603         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3604
3605         switch (cmd) {
3606         case SIOCGMIIPHY:
3607                 mii = if_mii(ifr);
3608                 if (!mii)
3609                         return -EINVAL;
3610
3611                 mii->phy_id = 0;
3612                 /* Fall Through */
3613         case SIOCGMIIREG:
3614                 /*
3615                  * We do this again just in case we were called by SIOCGMIIREG
3616                  * instead of SIOCGMIIPHY.
3617                  */
3618                 mii = if_mii(ifr);
3619                 if (!mii)
3620                         return -EINVAL;
3621
3622
3623                 if (mii->reg_num == 1) {
3624                         struct bonding *bond = netdev_priv(bond_dev);
3625                         mii->val_out = 0;
3626                         read_lock(&bond->lock);
3627                         read_lock(&bond->curr_slave_lock);
3628                         if (netif_carrier_ok(bond->dev))
3629                                 mii->val_out = BMSR_LSTATUS;
3630
3631                         read_unlock(&bond->curr_slave_lock);
3632                         read_unlock(&bond->lock);
3633                 }
3634
3635                 return 0;
3636         case BOND_INFO_QUERY_OLD:
3637         case SIOCBONDINFOQUERY:
3638                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3639
3640                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3641                         return -EFAULT;
3642
3643                 res = bond_info_query(bond_dev, &k_binfo);
3644                 if (res == 0 &&
3645                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3646                         return -EFAULT;
3647
3648                 return res;
3649         case BOND_SLAVE_INFO_QUERY_OLD:
3650         case SIOCBONDSLAVEINFOQUERY:
3651                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3652
3653                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3654                         return -EFAULT;
3655
3656                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3657                 if (res == 0 &&
3658                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3659                         return -EFAULT;
3660
3661                 return res;
3662         default:
3663                 /* Go on */
3664                 break;
3665         }
3666
3667         if (!capable(CAP_NET_ADMIN))
3668                 return -EPERM;
3669
3670         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3671
3672         pr_debug("slave_dev=%p:\n", slave_dev);
3673
3674         if (!slave_dev)
3675                 res = -ENODEV;
3676         else {
3677                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3678                 switch (cmd) {
3679                 case BOND_ENSLAVE_OLD:
3680                 case SIOCBONDENSLAVE:
3681                         res = bond_enslave(bond_dev, slave_dev);
3682                         break;
3683                 case BOND_RELEASE_OLD:
3684                 case SIOCBONDRELEASE:
3685                         res = bond_release(bond_dev, slave_dev);
3686                         break;
3687                 case BOND_SETHWADDR_OLD:
3688                 case SIOCBONDSETHWADDR:
3689                         res = bond_sethwaddr(bond_dev, slave_dev);
3690                         break;
3691                 case BOND_CHANGE_ACTIVE_OLD:
3692                 case SIOCBONDCHANGEACTIVE:
3693                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3694                         break;
3695                 default:
3696                         res = -EOPNOTSUPP;
3697                 }
3698
3699                 dev_put(slave_dev);
3700         }
3701
3702         return res;
3703 }
3704
3705 static bool bond_addr_in_mc_list(unsigned char *addr,
3706                                  struct netdev_hw_addr_list *list,
3707                                  int addrlen)
3708 {
3709         struct netdev_hw_addr *ha;
3710
3711         netdev_hw_addr_list_for_each(ha, list)
3712                 if (!memcmp(ha->addr, addr, addrlen))
3713                         return true;
3714
3715         return false;
3716 }
3717
3718 static void bond_set_multicast_list(struct net_device *bond_dev)
3719 {
3720         struct bonding *bond = netdev_priv(bond_dev);
3721         struct netdev_hw_addr *ha;
3722         bool found;
3723
3724         /*
3725          * Do promisc before checking multicast_mode
3726          */
3727         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
3728                 /*
3729                  * FIXME: Need to handle the error when one of the multi-slaves
3730                  * encounters error.
3731                  */
3732                 bond_set_promiscuity(bond, 1);
3733
3734
3735         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
3736                 bond_set_promiscuity(bond, -1);
3737
3738
3739         /* set allmulti flag to slaves */
3740         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
3741                 /*
3742                  * FIXME: Need to handle the error when one of the multi-slaves
3743                  * encounters error.
3744                  */
3745                 bond_set_allmulti(bond, 1);
3746
3747
3748         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
3749                 bond_set_allmulti(bond, -1);
3750
3751
3752         read_lock(&bond->lock);
3753
3754         bond->flags = bond_dev->flags;
3755
3756         /* looking for addresses to add to slaves' mc list */
3757         netdev_for_each_mc_addr(ha, bond_dev) {
3758                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3759                                              bond_dev->addr_len);
3760                 if (!found)
3761                         bond_mc_add(bond, ha->addr);
3762         }
3763
3764         /* looking for addresses to delete from slaves' list */
3765         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3766                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3767                                              bond_dev->addr_len);
3768                 if (!found)
3769                         bond_mc_del(bond, ha->addr);
3770         }
3771
3772         /* save master's multicast list */
3773         __hw_addr_flush(&bond->mc_list);
3774         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3775                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3776
3777         read_unlock(&bond->lock);
3778 }
3779
3780 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3781 {
3782         struct bonding *bond = netdev_priv(dev);
3783         struct slave *slave = bond->first_slave;
3784
3785         if (slave) {
3786                 const struct net_device_ops *slave_ops
3787                         = slave->dev->netdev_ops;
3788                 if (slave_ops->ndo_neigh_setup)
3789                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
3790         }
3791         return 0;
3792 }
3793
3794 /*
3795  * Change the MTU of all of a master's slaves to match the master
3796  */
3797 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3798 {
3799         struct bonding *bond = netdev_priv(bond_dev);
3800         struct slave *slave, *stop_at;
3801         int res = 0;
3802         int i;
3803
3804         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3805                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3806
3807         /* Can't hold bond->lock with bh disabled here since
3808          * some base drivers panic. On the other hand we can't
3809          * hold bond->lock without bh disabled because we'll
3810          * deadlock. The only solution is to rely on the fact
3811          * that we're under rtnl_lock here, and the slaves
3812          * list won't change. This doesn't solve the problem
3813          * of setting the slave's MTU while it is
3814          * transmitting, but the assumption is that the base
3815          * driver can handle that.
3816          *
3817          * TODO: figure out a way to safely iterate the slaves
3818          * list, but without holding a lock around the actual
3819          * call to the base driver.
3820          */
3821
3822         bond_for_each_slave(bond, slave, i) {
3823                 pr_debug("s %p s->p %p c_m %p\n",
3824                          slave,
3825                          slave->prev,
3826                          slave->dev->netdev_ops->ndo_change_mtu);
3827
3828                 res = dev_set_mtu(slave->dev, new_mtu);
3829
3830                 if (res) {
3831                         /* If we failed to set the slave's mtu to the new value
3832                          * we must abort the operation even in ACTIVE_BACKUP
3833                          * mode, because if we allow the backup slaves to have
3834                          * different mtu values than the active slave we'll
3835                          * need to change their mtu when doing a failover. That
3836                          * means changing their mtu from timer context, which
3837                          * is probably not a good idea.
3838                          */
3839                         pr_debug("err %d %s\n", res, slave->dev->name);
3840                         goto unwind;
3841                 }
3842         }
3843
3844         bond_dev->mtu = new_mtu;
3845
3846         return 0;
3847
3848 unwind:
3849         /* unwind from head to the slave that failed */
3850         stop_at = slave;
3851         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3852                 int tmp_res;
3853
3854                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3855                 if (tmp_res) {
3856                         pr_debug("unwind err %d dev %s\n",
3857                                  tmp_res, slave->dev->name);
3858                 }
3859         }
3860
3861         return res;
3862 }
3863
3864 /*
3865  * Change HW address
3866  *
3867  * Note that many devices must be down to change the HW address, and
3868  * downing the master releases all slaves.  We can make bonds full of
3869  * bonding devices to test this, however.
3870  */
3871 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3872 {
3873         struct bonding *bond = netdev_priv(bond_dev);
3874         struct sockaddr *sa = addr, tmp_sa;
3875         struct slave *slave, *stop_at;
3876         int res = 0;
3877         int i;
3878
3879         if (bond->params.mode == BOND_MODE_ALB)
3880                 return bond_alb_set_mac_address(bond_dev, addr);
3881
3882
3883         pr_debug("bond=%p, name=%s\n",
3884                  bond, bond_dev ? bond_dev->name : "None");
3885
3886         /*
3887          * If fail_over_mac is set to active, do nothing and return
3888          * success.  Returning an error causes ifenslave to fail.
3889          */
3890         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3891                 return 0;
3892
3893         if (!is_valid_ether_addr(sa->sa_data))
3894                 return -EADDRNOTAVAIL;
3895
3896         /* Can't hold bond->lock with bh disabled here since
3897          * some base drivers panic. On the other hand we can't
3898          * hold bond->lock without bh disabled because we'll
3899          * deadlock. The only solution is to rely on the fact
3900          * that we're under rtnl_lock here, and the slaves
3901          * list won't change. This doesn't solve the problem
3902          * of setting the slave's hw address while it is
3903          * transmitting, but the assumption is that the base
3904          * driver can handle that.
3905          *
3906          * TODO: figure out a way to safely iterate the slaves
3907          * list, but without holding a lock around the actual
3908          * call to the base driver.
3909          */
3910
3911         bond_for_each_slave(bond, slave, i) {
3912                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3913                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3914
3915                 if (slave_ops->ndo_set_mac_address == NULL) {
3916                         res = -EOPNOTSUPP;
3917                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3918                         goto unwind;
3919                 }
3920
3921                 res = dev_set_mac_address(slave->dev, addr);
3922                 if (res) {
3923                         /* TODO: consider downing the slave
3924                          * and retry ?
3925                          * User should expect communications
3926                          * breakage anyway until ARP finish
3927                          * updating, so...
3928                          */
3929                         pr_debug("err %d %s\n", res, slave->dev->name);
3930                         goto unwind;
3931                 }
3932         }
3933
3934         /* success */
3935         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3936         return 0;
3937
3938 unwind:
3939         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3940         tmp_sa.sa_family = bond_dev->type;
3941
3942         /* unwind from head to the slave that failed */
3943         stop_at = slave;
3944         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3945                 int tmp_res;
3946
3947                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3948                 if (tmp_res) {
3949                         pr_debug("unwind err %d dev %s\n",
3950                                  tmp_res, slave->dev->name);
3951                 }
3952         }
3953
3954         return res;
3955 }
3956
3957 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3958 {
3959         struct bonding *bond = netdev_priv(bond_dev);
3960         struct slave *slave, *start_at;
3961         int i, slave_no, res = 1;
3962         struct iphdr *iph = ip_hdr(skb);
3963
3964         /*
3965          * Start with the curr_active_slave that joined the bond as the
3966          * default for sending IGMP traffic.  For failover purposes one
3967          * needs to maintain some consistency for the interface that will
3968          * send the join/membership reports.  The curr_active_slave found
3969          * will send all of this type of traffic.
3970          */
3971         if ((iph->protocol == IPPROTO_IGMP) &&
3972             (skb->protocol == htons(ETH_P_IP))) {
3973
3974                 read_lock(&bond->curr_slave_lock);
3975                 slave = bond->curr_active_slave;
3976                 read_unlock(&bond->curr_slave_lock);
3977
3978                 if (!slave)
3979                         goto out;
3980         } else {
3981                 /*
3982                  * Concurrent TX may collide on rr_tx_counter; we accept
3983                  * that as being rare enough not to justify using an
3984                  * atomic op here.
3985                  */
3986                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3987
3988                 bond_for_each_slave(bond, slave, i) {
3989                         slave_no--;
3990                         if (slave_no < 0)
3991                                 break;
3992                 }
3993         }
3994
3995         start_at = slave;
3996         bond_for_each_slave_from(bond, slave, i, start_at) {
3997                 if (IS_UP(slave->dev) &&
3998                     (slave->link == BOND_LINK_UP) &&
3999                     bond_is_active_slave(slave)) {
4000                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4001                         break;
4002                 }
4003         }
4004
4005 out:
4006         if (res) {
4007                 /* no suitable interface, frame not sent */
4008                 dev_kfree_skb(skb);
4009         }
4010
4011         return NETDEV_TX_OK;
4012 }
4013
4014
4015 /*
4016  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4017  * the bond has a usable interface.
4018  */
4019 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4020 {
4021         struct bonding *bond = netdev_priv(bond_dev);
4022         int res = 1;
4023
4024         read_lock(&bond->curr_slave_lock);
4025
4026         if (bond->curr_active_slave)
4027                 res = bond_dev_queue_xmit(bond, skb,
4028                         bond->curr_active_slave->dev);
4029
4030         if (res)
4031                 /* no suitable interface, frame not sent */
4032                 dev_kfree_skb(skb);
4033
4034         read_unlock(&bond->curr_slave_lock);
4035
4036         return NETDEV_TX_OK;
4037 }
4038
4039 /*
4040  * In bond_xmit_xor() , we determine the output device by using a pre-
4041  * determined xmit_hash_policy(), If the selected device is not enabled,
4042  * find the next active slave.
4043  */
4044 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4045 {
4046         struct bonding *bond = netdev_priv(bond_dev);
4047         struct slave *slave, *start_at;
4048         int slave_no;
4049         int i;
4050         int res = 1;
4051
4052         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4053
4054         bond_for_each_slave(bond, slave, i) {
4055                 slave_no--;
4056                 if (slave_no < 0)
4057                         break;
4058         }
4059
4060         start_at = slave;
4061
4062         bond_for_each_slave_from(bond, slave, i, start_at) {
4063                 if (IS_UP(slave->dev) &&
4064                     (slave->link == BOND_LINK_UP) &&
4065                     bond_is_active_slave(slave)) {
4066                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4067                         break;
4068                 }
4069         }
4070
4071         if (res) {
4072                 /* no suitable interface, frame not sent */
4073                 dev_kfree_skb(skb);
4074         }
4075
4076         return NETDEV_TX_OK;
4077 }
4078
4079 /*
4080  * in broadcast mode, we send everything to all usable interfaces.
4081  */
4082 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4083 {
4084         struct bonding *bond = netdev_priv(bond_dev);
4085         struct slave *slave, *start_at;
4086         struct net_device *tx_dev = NULL;
4087         int i;
4088         int res = 1;
4089
4090         read_lock(&bond->curr_slave_lock);
4091         start_at = bond->curr_active_slave;
4092         read_unlock(&bond->curr_slave_lock);
4093
4094         if (!start_at)
4095                 goto out;
4096
4097         bond_for_each_slave_from(bond, slave, i, start_at) {
4098                 if (IS_UP(slave->dev) &&
4099                     (slave->link == BOND_LINK_UP) &&
4100                     bond_is_active_slave(slave)) {
4101                         if (tx_dev) {
4102                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4103                                 if (!skb2) {
4104                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4105                                                bond_dev->name);
4106                                         continue;
4107                                 }
4108
4109                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4110                                 if (res) {
4111                                         dev_kfree_skb(skb2);
4112                                         continue;
4113                                 }
4114                         }
4115                         tx_dev = slave->dev;
4116                 }
4117         }
4118
4119         if (tx_dev)
4120                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4121
4122 out:
4123         if (res)
4124                 /* no suitable interface, frame not sent */
4125                 dev_kfree_skb(skb);
4126
4127         /* frame sent to all suitable interfaces */
4128         return NETDEV_TX_OK;
4129 }
4130
4131 /*------------------------- Device initialization ---------------------------*/
4132
4133 static void bond_set_xmit_hash_policy(struct bonding *bond)
4134 {
4135         switch (bond->params.xmit_policy) {
4136         case BOND_XMIT_POLICY_LAYER23:
4137                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4138                 break;
4139         case BOND_XMIT_POLICY_LAYER34:
4140                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4141                 break;
4142         case BOND_XMIT_POLICY_LAYER2:
4143         default:
4144                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4145                 break;
4146         }
4147 }
4148
4149 /*
4150  * Lookup the slave that corresponds to a qid
4151  */
4152 static inline int bond_slave_override(struct bonding *bond,
4153                                       struct sk_buff *skb)
4154 {
4155         int i, res = 1;
4156         struct slave *slave = NULL;
4157         struct slave *check_slave;
4158
4159         if (!skb->queue_mapping)
4160                 return 1;
4161
4162         /* Find out if any slaves have the same mapping as this skb. */
4163         bond_for_each_slave(bond, check_slave, i) {
4164                 if (check_slave->queue_id == skb->queue_mapping) {
4165                         slave = check_slave;
4166                         break;
4167                 }
4168         }
4169
4170         /* If the slave isn't UP, use default transmit policy. */
4171         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4172             (slave->link == BOND_LINK_UP)) {
4173                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4174         }
4175
4176         return res;
4177 }
4178
4179
4180 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4181 {
4182         /*
4183          * This helper function exists to help dev_pick_tx get the correct
4184          * destination queue.  Using a helper function skips a call to
4185          * skb_tx_hash and will put the skbs in the queue we expect on their
4186          * way down to the bonding driver.
4187          */
4188         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4189
4190         /*
4191          * Save the original txq to restore before passing to the driver
4192          */
4193         bond_queue_mapping(skb) = skb->queue_mapping;
4194
4195         if (unlikely(txq >= dev->real_num_tx_queues)) {
4196                 do {
4197                         txq -= dev->real_num_tx_queues;
4198                 } while (txq >= dev->real_num_tx_queues);
4199         }
4200         return txq;
4201 }
4202
4203 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4204 {
4205         struct bonding *bond = netdev_priv(dev);
4206
4207         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4208                 if (!bond_slave_override(bond, skb))
4209                         return NETDEV_TX_OK;
4210         }
4211
4212         switch (bond->params.mode) {
4213         case BOND_MODE_ROUNDROBIN:
4214                 return bond_xmit_roundrobin(skb, dev);
4215         case BOND_MODE_ACTIVEBACKUP:
4216                 return bond_xmit_activebackup(skb, dev);
4217         case BOND_MODE_XOR:
4218                 return bond_xmit_xor(skb, dev);
4219         case BOND_MODE_BROADCAST:
4220                 return bond_xmit_broadcast(skb, dev);
4221         case BOND_MODE_8023AD:
4222                 return bond_3ad_xmit_xor(skb, dev);
4223         case BOND_MODE_ALB:
4224         case BOND_MODE_TLB:
4225                 return bond_alb_xmit(skb, dev);
4226         default:
4227                 /* Should never happen, mode already checked */
4228                 pr_err("%s: Error: Unknown bonding mode %d\n",
4229                        dev->name, bond->params.mode);
4230                 WARN_ON_ONCE(1);
4231                 dev_kfree_skb(skb);
4232                 return NETDEV_TX_OK;
4233         }
4234 }
4235
4236 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4237 {
4238         struct bonding *bond = netdev_priv(dev);
4239         netdev_tx_t ret = NETDEV_TX_OK;
4240
4241         /*
4242          * If we risk deadlock from transmitting this in the
4243          * netpoll path, tell netpoll to queue the frame for later tx
4244          */
4245         if (is_netpoll_tx_blocked(dev))
4246                 return NETDEV_TX_BUSY;
4247
4248         read_lock(&bond->lock);
4249
4250         if (bond->slave_cnt)
4251                 ret = __bond_start_xmit(skb, dev);
4252         else
4253                 dev_kfree_skb(skb);
4254
4255         read_unlock(&bond->lock);
4256
4257         return ret;
4258 }
4259
4260 /*
4261  * set bond mode specific net device operations
4262  */
4263 void bond_set_mode_ops(struct bonding *bond, int mode)
4264 {
4265         struct net_device *bond_dev = bond->dev;
4266
4267         switch (mode) {
4268         case BOND_MODE_ROUNDROBIN:
4269                 break;
4270         case BOND_MODE_ACTIVEBACKUP:
4271                 break;
4272         case BOND_MODE_XOR:
4273                 bond_set_xmit_hash_policy(bond);
4274                 break;
4275         case BOND_MODE_BROADCAST:
4276                 break;
4277         case BOND_MODE_8023AD:
4278                 bond_set_xmit_hash_policy(bond);
4279                 break;
4280         case BOND_MODE_ALB:
4281                 /* FALLTHRU */
4282         case BOND_MODE_TLB:
4283                 break;
4284         default:
4285                 /* Should never happen, mode already checked */
4286                 pr_err("%s: Error: Unknown bonding mode %d\n",
4287                        bond_dev->name, mode);
4288                 break;
4289         }
4290 }
4291
4292 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4293                                     struct ethtool_drvinfo *drvinfo)
4294 {
4295         strncpy(drvinfo->driver, DRV_NAME, 32);
4296         strncpy(drvinfo->version, DRV_VERSION, 32);
4297         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4298 }
4299
4300 static const struct ethtool_ops bond_ethtool_ops = {
4301         .get_drvinfo            = bond_ethtool_get_drvinfo,
4302         .get_link               = ethtool_op_get_link,
4303 };
4304
4305 static const struct net_device_ops bond_netdev_ops = {
4306         .ndo_init               = bond_init,
4307         .ndo_uninit             = bond_uninit,
4308         .ndo_open               = bond_open,
4309         .ndo_stop               = bond_close,
4310         .ndo_start_xmit         = bond_start_xmit,
4311         .ndo_select_queue       = bond_select_queue,
4312         .ndo_get_stats64        = bond_get_stats,
4313         .ndo_do_ioctl           = bond_do_ioctl,
4314         .ndo_set_multicast_list = bond_set_multicast_list,
4315         .ndo_change_mtu         = bond_change_mtu,
4316         .ndo_set_mac_address    = bond_set_mac_address,
4317         .ndo_neigh_setup        = bond_neigh_setup,
4318         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4319         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4320 #ifdef CONFIG_NET_POLL_CONTROLLER
4321         .ndo_netpoll_setup      = bond_netpoll_setup,
4322         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4323         .ndo_poll_controller    = bond_poll_controller,
4324 #endif
4325         .ndo_add_slave          = bond_enslave,
4326         .ndo_del_slave          = bond_release,
4327         .ndo_fix_features       = bond_fix_features,
4328 };
4329
4330 static void bond_destructor(struct net_device *bond_dev)
4331 {
4332         struct bonding *bond = netdev_priv(bond_dev);
4333         if (bond->wq)
4334                 destroy_workqueue(bond->wq);
4335         free_netdev(bond_dev);
4336 }
4337
4338 static void bond_setup(struct net_device *bond_dev)
4339 {
4340         struct bonding *bond = netdev_priv(bond_dev);
4341
4342         /* initialize rwlocks */
4343         rwlock_init(&bond->lock);
4344         rwlock_init(&bond->curr_slave_lock);
4345
4346         bond->params = bonding_defaults;
4347
4348         /* Initialize pointers */
4349         bond->dev = bond_dev;
4350         INIT_LIST_HEAD(&bond->vlan_list);
4351
4352         /* Initialize the device entry points */
4353         ether_setup(bond_dev);
4354         bond_dev->netdev_ops = &bond_netdev_ops;
4355         bond_dev->ethtool_ops = &bond_ethtool_ops;
4356         bond_set_mode_ops(bond, bond->params.mode);
4357
4358         bond_dev->destructor = bond_destructor;
4359
4360         /* Initialize the device options */
4361         bond_dev->tx_queue_len = 0;
4362         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4363         bond_dev->priv_flags |= IFF_BONDING;
4364         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4365
4366         /* At first, we block adding VLANs. That's the only way to
4367          * prevent problems that occur when adding VLANs over an
4368          * empty bond. The block will be removed once non-challenged
4369          * slaves are enslaved.
4370          */
4371         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4372
4373         /* don't acquire bond device's netif_tx_lock when
4374          * transmitting */
4375         bond_dev->features |= NETIF_F_LLTX;
4376
4377         /* By default, we declare the bond to be fully
4378          * VLAN hardware accelerated capable. Special
4379          * care is taken in the various xmit functions
4380          * when there are slaves that are not hw accel
4381          * capable
4382          */
4383
4384         bond_dev->hw_features = BOND_VLAN_FEATURES |
4385                                 NETIF_F_HW_VLAN_TX |
4386                                 NETIF_F_HW_VLAN_RX |
4387                                 NETIF_F_HW_VLAN_FILTER;
4388
4389         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM);
4390         bond_dev->features |= bond_dev->hw_features;
4391 }
4392
4393 static void bond_work_cancel_all(struct bonding *bond)
4394 {
4395         write_lock_bh(&bond->lock);
4396         bond->kill_timers = 1;
4397         write_unlock_bh(&bond->lock);
4398
4399         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4400                 cancel_delayed_work(&bond->mii_work);
4401
4402         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4403                 cancel_delayed_work(&bond->arp_work);
4404
4405         if (bond->params.mode == BOND_MODE_ALB &&
4406             delayed_work_pending(&bond->alb_work))
4407                 cancel_delayed_work(&bond->alb_work);
4408
4409         if (bond->params.mode == BOND_MODE_8023AD &&
4410             delayed_work_pending(&bond->ad_work))
4411                 cancel_delayed_work(&bond->ad_work);
4412
4413         if (delayed_work_pending(&bond->mcast_work))
4414                 cancel_delayed_work(&bond->mcast_work);
4415 }
4416
4417 /*
4418 * Destroy a bonding device.
4419 * Must be under rtnl_lock when this function is called.
4420 */
4421 static void bond_uninit(struct net_device *bond_dev)
4422 {
4423         struct bonding *bond = netdev_priv(bond_dev);
4424         struct vlan_entry *vlan, *tmp;
4425
4426         bond_netpoll_cleanup(bond_dev);
4427
4428         /* Release the bonded slaves */
4429         bond_release_all(bond_dev);
4430
4431         list_del(&bond->bond_list);
4432
4433         bond_work_cancel_all(bond);
4434
4435         bond_remove_proc_entry(bond);
4436
4437         bond_debug_unregister(bond);
4438
4439         __hw_addr_flush(&bond->mc_list);
4440
4441         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4442                 list_del(&vlan->vlan_list);
4443                 kfree(vlan);
4444         }
4445 }
4446
4447 /*------------------------- Module initialization ---------------------------*/
4448
4449 /*
4450  * Convert string input module parms.  Accept either the
4451  * number of the mode or its string name.  A bit complicated because
4452  * some mode names are substrings of other names, and calls from sysfs
4453  * may have whitespace in the name (trailing newlines, for example).
4454  */
4455 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4456 {
4457         int modeint = -1, i, rv;
4458         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4459
4460         for (p = (char *)buf; *p; p++)
4461                 if (!(isdigit(*p) || isspace(*p)))
4462                         break;
4463
4464         if (*p)
4465                 rv = sscanf(buf, "%20s", modestr);
4466         else
4467                 rv = sscanf(buf, "%d", &modeint);
4468
4469         if (!rv)
4470                 return -1;
4471
4472         for (i = 0; tbl[i].modename; i++) {
4473                 if (modeint == tbl[i].mode)
4474                         return tbl[i].mode;
4475                 if (strcmp(modestr, tbl[i].modename) == 0)
4476                         return tbl[i].mode;
4477         }
4478
4479         return -1;
4480 }
4481
4482 static int bond_check_params(struct bond_params *params)
4483 {
4484         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4485
4486         /*
4487          * Convert string parameters.
4488          */
4489         if (mode) {
4490                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4491                 if (bond_mode == -1) {
4492                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4493                                mode == NULL ? "NULL" : mode);
4494                         return -EINVAL;
4495                 }
4496         }
4497
4498         if (xmit_hash_policy) {
4499                 if ((bond_mode != BOND_MODE_XOR) &&
4500                     (bond_mode != BOND_MODE_8023AD)) {
4501                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4502                                bond_mode_name(bond_mode));
4503                 } else {
4504                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4505                                                         xmit_hashtype_tbl);
4506                         if (xmit_hashtype == -1) {
4507                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4508                                        xmit_hash_policy == NULL ? "NULL" :
4509                                        xmit_hash_policy);
4510                                 return -EINVAL;
4511                         }
4512                 }
4513         }
4514
4515         if (lacp_rate) {
4516                 if (bond_mode != BOND_MODE_8023AD) {
4517                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4518                                 bond_mode_name(bond_mode));
4519                 } else {
4520                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4521                         if (lacp_fast == -1) {
4522                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4523                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4524                                 return -EINVAL;
4525                         }
4526                 }
4527         }
4528
4529         if (ad_select) {
4530                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4531                 if (params->ad_select == -1) {
4532                         pr_err("Error: Invalid ad_select \"%s\"\n",
4533                                ad_select == NULL ? "NULL" : ad_select);
4534                         return -EINVAL;
4535                 }
4536
4537                 if (bond_mode != BOND_MODE_8023AD) {
4538                         pr_warning("ad_select param only affects 802.3ad mode\n");
4539                 }
4540         } else {
4541                 params->ad_select = BOND_AD_STABLE;
4542         }
4543
4544         if (max_bonds < 0) {
4545                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4546                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4547                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4548         }
4549
4550         if (miimon < 0) {
4551                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4552                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4553                 miimon = BOND_LINK_MON_INTERV;
4554         }
4555
4556         if (updelay < 0) {
4557                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4558                            updelay, INT_MAX);
4559                 updelay = 0;
4560         }
4561
4562         if (downdelay < 0) {
4563                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4564                            downdelay, INT_MAX);
4565                 downdelay = 0;
4566         }
4567
4568         if ((use_carrier != 0) && (use_carrier != 1)) {
4569                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4570                            use_carrier);
4571                 use_carrier = 1;
4572         }
4573
4574         if (num_peer_notif < 0 || num_peer_notif > 255) {
4575                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4576                            num_peer_notif);
4577                 num_peer_notif = 1;
4578         }
4579
4580         /* reset values for 802.3ad */
4581         if (bond_mode == BOND_MODE_8023AD) {
4582                 if (!miimon) {
4583                         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");
4584                         pr_warning("Forcing miimon to 100msec\n");
4585                         miimon = 100;
4586                 }
4587         }
4588
4589         if (tx_queues < 1 || tx_queues > 255) {
4590                 pr_warning("Warning: tx_queues (%d) should be between "
4591                            "1 and 255, resetting to %d\n",
4592                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4593                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4594         }
4595
4596         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4597                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4598                            "not of valid value (0/1), so it was set to "
4599                            "0\n", all_slaves_active);
4600                 all_slaves_active = 0;
4601         }
4602
4603         if (resend_igmp < 0 || resend_igmp > 255) {
4604                 pr_warning("Warning: resend_igmp (%d) should be between "
4605                            "0 and 255, resetting to %d\n",
4606                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4607                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4608         }
4609
4610         /* reset values for TLB/ALB */
4611         if ((bond_mode == BOND_MODE_TLB) ||
4612             (bond_mode == BOND_MODE_ALB)) {
4613                 if (!miimon) {
4614                         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");
4615                         pr_warning("Forcing miimon to 100msec\n");
4616                         miimon = 100;
4617                 }
4618         }
4619
4620         if (bond_mode == BOND_MODE_ALB) {
4621                 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",
4622                           updelay);
4623         }
4624
4625         if (!miimon) {
4626                 if (updelay || downdelay) {
4627                         /* just warn the user the up/down delay will have
4628                          * no effect since miimon is zero...
4629                          */
4630                         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",
4631                                    updelay, downdelay);
4632                 }
4633         } else {
4634                 /* don't allow arp monitoring */
4635                 if (arp_interval) {
4636                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4637                                    miimon, arp_interval);
4638                         arp_interval = 0;
4639                 }
4640
4641                 if ((updelay % miimon) != 0) {
4642                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4643                                    updelay, miimon,
4644                                    (updelay / miimon) * miimon);
4645                 }
4646
4647                 updelay /= miimon;
4648
4649                 if ((downdelay % miimon) != 0) {
4650                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4651                                    downdelay, miimon,
4652                                    (downdelay / miimon) * miimon);
4653                 }
4654
4655                 downdelay /= miimon;
4656         }
4657
4658         if (arp_interval < 0) {
4659                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4660                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4661                 arp_interval = BOND_LINK_ARP_INTERV;
4662         }
4663
4664         for (arp_ip_count = 0;
4665              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4666              arp_ip_count++) {
4667                 /* not complete check, but should be good enough to
4668                    catch mistakes */
4669                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4670                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4671                                    arp_ip_target[arp_ip_count]);
4672                         arp_interval = 0;
4673                 } else {
4674                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4675                         arp_target[arp_ip_count] = ip;
4676                 }
4677         }
4678
4679         if (arp_interval && !arp_ip_count) {
4680                 /* don't allow arping if no arp_ip_target given... */
4681                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4682                            arp_interval);
4683                 arp_interval = 0;
4684         }
4685
4686         if (arp_validate) {
4687                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4688                         pr_err("arp_validate only supported in active-backup mode\n");
4689                         return -EINVAL;
4690                 }
4691                 if (!arp_interval) {
4692                         pr_err("arp_validate requires arp_interval\n");
4693                         return -EINVAL;
4694                 }
4695
4696                 arp_validate_value = bond_parse_parm(arp_validate,
4697                                                      arp_validate_tbl);
4698                 if (arp_validate_value == -1) {
4699                         pr_err("Error: invalid arp_validate \"%s\"\n",
4700                                arp_validate == NULL ? "NULL" : arp_validate);
4701                         return -EINVAL;
4702                 }
4703         } else
4704                 arp_validate_value = 0;
4705
4706         if (miimon) {
4707                 pr_info("MII link monitoring set to %d ms\n", miimon);
4708         } else if (arp_interval) {
4709                 int i;
4710
4711                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4712                         arp_interval,
4713                         arp_validate_tbl[arp_validate_value].modename,
4714                         arp_ip_count);
4715
4716                 for (i = 0; i < arp_ip_count; i++)
4717                         pr_info(" %s", arp_ip_target[i]);
4718
4719                 pr_info("\n");
4720
4721         } else if (max_bonds) {
4722                 /* miimon and arp_interval not set, we need one so things
4723                  * work as expected, see bonding.txt for details
4724                  */
4725                 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");
4726         }
4727
4728         if (primary && !USES_PRIMARY(bond_mode)) {
4729                 /* currently, using a primary only makes sense
4730                  * in active backup, TLB or ALB modes
4731                  */
4732                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4733                            primary, bond_mode_name(bond_mode));
4734                 primary = NULL;
4735         }
4736
4737         if (primary && primary_reselect) {
4738                 primary_reselect_value = bond_parse_parm(primary_reselect,
4739                                                          pri_reselect_tbl);
4740                 if (primary_reselect_value == -1) {
4741                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4742                                primary_reselect ==
4743                                         NULL ? "NULL" : primary_reselect);
4744                         return -EINVAL;
4745                 }
4746         } else {
4747                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4748         }
4749
4750         if (fail_over_mac) {
4751                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4752                                                       fail_over_mac_tbl);
4753                 if (fail_over_mac_value == -1) {
4754                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4755                                arp_validate == NULL ? "NULL" : arp_validate);
4756                         return -EINVAL;
4757                 }
4758
4759                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4760                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4761         } else {
4762                 fail_over_mac_value = BOND_FOM_NONE;
4763         }
4764
4765         /* fill params struct with the proper values */
4766         params->mode = bond_mode;
4767         params->xmit_policy = xmit_hashtype;
4768         params->miimon = miimon;
4769         params->num_peer_notif = num_peer_notif;
4770         params->arp_interval = arp_interval;
4771         params->arp_validate = arp_validate_value;
4772         params->updelay = updelay;
4773         params->downdelay = downdelay;
4774         params->use_carrier = use_carrier;
4775         params->lacp_fast = lacp_fast;
4776         params->primary[0] = 0;
4777         params->primary_reselect = primary_reselect_value;
4778         params->fail_over_mac = fail_over_mac_value;
4779         params->tx_queues = tx_queues;
4780         params->all_slaves_active = all_slaves_active;
4781         params->resend_igmp = resend_igmp;
4782         params->min_links = min_links;
4783
4784         if (primary) {
4785                 strncpy(params->primary, primary, IFNAMSIZ);
4786                 params->primary[IFNAMSIZ - 1] = 0;
4787         }
4788
4789         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4790
4791         return 0;
4792 }
4793
4794 static struct lock_class_key bonding_netdev_xmit_lock_key;
4795 static struct lock_class_key bonding_netdev_addr_lock_key;
4796
4797 static void bond_set_lockdep_class_one(struct net_device *dev,
4798                                        struct netdev_queue *txq,
4799                                        void *_unused)
4800 {
4801         lockdep_set_class(&txq->_xmit_lock,
4802                           &bonding_netdev_xmit_lock_key);
4803 }
4804
4805 static void bond_set_lockdep_class(struct net_device *dev)
4806 {
4807         lockdep_set_class(&dev->addr_list_lock,
4808                           &bonding_netdev_addr_lock_key);
4809         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4810 }
4811
4812 /*
4813  * Called from registration process
4814  */
4815 static int bond_init(struct net_device *bond_dev)
4816 {
4817         struct bonding *bond = netdev_priv(bond_dev);
4818         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4819         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4820
4821         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4822
4823         /*
4824          * Initialize locks that may be required during
4825          * en/deslave operations.  All of the bond_open work
4826          * (of which this is part) should really be moved to
4827          * a phase prior to dev_open
4828          */
4829         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4830         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4831
4832         bond->wq = create_singlethread_workqueue(bond_dev->name);
4833         if (!bond->wq)
4834                 return -ENOMEM;
4835
4836         bond_set_lockdep_class(bond_dev);
4837
4838         bond_create_proc_entry(bond);
4839         list_add_tail(&bond->bond_list, &bn->dev_list);
4840
4841         bond_prepare_sysfs_group(bond);
4842
4843         bond_debug_register(bond);
4844
4845         __hw_addr_init(&bond->mc_list);
4846         return 0;
4847 }
4848
4849 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4850 {
4851         if (tb[IFLA_ADDRESS]) {
4852                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4853                         return -EINVAL;
4854                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4855                         return -EADDRNOTAVAIL;
4856         }
4857         return 0;
4858 }
4859
4860 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4861         .kind           = "bond",
4862         .priv_size      = sizeof(struct bonding),
4863         .setup          = bond_setup,
4864         .validate       = bond_validate,
4865 };
4866
4867 /* Create a new bond based on the specified name and bonding parameters.
4868  * If name is NULL, obtain a suitable "bond%d" name for us.
4869  * Caller must NOT hold rtnl_lock; we need to release it here before we
4870  * set up our sysfs entries.
4871  */
4872 int bond_create(struct net *net, const char *name)
4873 {
4874         struct net_device *bond_dev;
4875         int res;
4876
4877         rtnl_lock();
4878
4879         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4880                                    name ? name : "bond%d",
4881                                    bond_setup, tx_queues);
4882         if (!bond_dev) {
4883                 pr_err("%s: eek! can't alloc netdev!\n", name);
4884                 rtnl_unlock();
4885                 return -ENOMEM;
4886         }
4887
4888         dev_net_set(bond_dev, net);
4889         bond_dev->rtnl_link_ops = &bond_link_ops;
4890
4891         res = register_netdevice(bond_dev);
4892
4893         netif_carrier_off(bond_dev);
4894
4895         rtnl_unlock();
4896         if (res < 0)
4897                 bond_destructor(bond_dev);
4898         return res;
4899 }
4900
4901 static int __net_init bond_net_init(struct net *net)
4902 {
4903         struct bond_net *bn = net_generic(net, bond_net_id);
4904
4905         bn->net = net;
4906         INIT_LIST_HEAD(&bn->dev_list);
4907
4908         bond_create_proc_dir(bn);
4909
4910         return 0;
4911 }
4912
4913 static void __net_exit bond_net_exit(struct net *net)
4914 {
4915         struct bond_net *bn = net_generic(net, bond_net_id);
4916
4917         bond_destroy_proc_dir(bn);
4918 }
4919
4920 static struct pernet_operations bond_net_ops = {
4921         .init = bond_net_init,
4922         .exit = bond_net_exit,
4923         .id   = &bond_net_id,
4924         .size = sizeof(struct bond_net),
4925 };
4926
4927 static int __init bonding_init(void)
4928 {
4929         int i;
4930         int res;
4931
4932         pr_info("%s", bond_version);
4933
4934         res = bond_check_params(&bonding_defaults);
4935         if (res)
4936                 goto out;
4937
4938         res = register_pernet_subsys(&bond_net_ops);
4939         if (res)
4940                 goto out;
4941
4942         res = rtnl_link_register(&bond_link_ops);
4943         if (res)
4944                 goto err_link;
4945
4946         bond_create_debugfs();
4947
4948         for (i = 0; i < max_bonds; i++) {
4949                 res = bond_create(&init_net, NULL);
4950                 if (res)
4951                         goto err;
4952         }
4953
4954         res = bond_create_sysfs();
4955         if (res)
4956                 goto err;
4957
4958         register_netdevice_notifier(&bond_netdev_notifier);
4959         register_inetaddr_notifier(&bond_inetaddr_notifier);
4960 out:
4961         return res;
4962 err:
4963         rtnl_link_unregister(&bond_link_ops);
4964 err_link:
4965         unregister_pernet_subsys(&bond_net_ops);
4966         goto out;
4967
4968 }
4969
4970 static void __exit bonding_exit(void)
4971 {
4972         unregister_netdevice_notifier(&bond_netdev_notifier);
4973         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4974
4975         bond_destroy_sysfs();
4976         bond_destroy_debugfs();
4977
4978         rtnl_link_unregister(&bond_link_ops);
4979         unregister_pernet_subsys(&bond_net_ops);
4980
4981 #ifdef CONFIG_NET_POLL_CONTROLLER
4982         /*
4983          * Make sure we don't have an imbalance on our netpoll blocking
4984          */
4985         WARN_ON(atomic_read(&netpoll_block_tx));
4986 #endif
4987 }
4988
4989 module_init(bonding_init);
4990 module_exit(bonding_exit);
4991 MODULE_LICENSE("GPL");
4992 MODULE_VERSION(DRV_VERSION);
4993 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4994 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4995 MODULE_ALIAS_RTNL_LINK("bond");