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