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