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/dma.h>
  58 #include <linux/uaccess.h>
  59 #include <linux/errno.h>
  60 #include <linux/netdevice.h>
  61 #include <linux/inetdevice.h>
  62 #include <linux/igmp.h>
  63 #include <linux/etherdevice.h>
  64 #include <linux/skbuff.h>
  65 #include <net/sock.h>
  66 #include <linux/rtnetlink.h>
  67 #include <linux/smp.h>
  68 #include <linux/if_ether.h>
  69 #include <net/arp.h>
  70 #include <linux/mii.h>
  71 #include <linux/ethtool.h>
  72 #include <linux/if_vlan.h>
  73 #include <linux/if_bonding.h>
  74 #include <linux/jiffies.h>
  75 #include <linux/preempt.h>
  76 #include <net/route.h>
  77 #include <net/net_namespace.h>
  78 #include <net/netns/generic.h>
  79 #include <net/pkt_sched.h>
  80 #include <linux/rculist.h>
  81 #include <net/flow_keys.h>
  82 #include "bonding.h"
  83 #include "bond_3ad.h"
  84 #include "bond_alb.h"
  85
  86 /*---------------------------- Module parameters ----------------------------*/
  87
  88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  89
  90 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  91 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  92 static int num_peer_notif = 1;
  93 static int miimon;
  94 static int updelay;
  95 static int downdelay;
  96 static int use_carrier  = 1;
  97 static char *mode;
  98 static char *primary;
  99 static char *primary_reselect;
 100 static char *lacp_rate;
 101 static int min_links;
 102 static char *ad_select;
 103 static char *xmit_hash_policy;
 104 static int arp_interval;
 105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 106 static char *arp_validate;
 107 static char *arp_all_targets;
 108 static char *fail_over_mac;
 109 static int all_slaves_active;
 110 static struct bond_params bonding_defaults;
 111 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 112 static int packets_per_slave = 1;
 113 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
 114
 115 module_param(max_bonds, int, 0);
 116 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 117 module_param(tx_queues, int, 0);
 118 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 119 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 120 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 121                                "failover event (alias of num_unsol_na)");
 122 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 123 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 124                                "failover event (alias of num_grat_arp)");
 125 module_param(miimon, int, 0);
 126 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 127 module_param(updelay, int, 0);
 128 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 129 module_param(downdelay, int, 0);
 130 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 131                             "in milliseconds");
 132 module_param(use_carrier, int, 0);
 133 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 134                               "0 for off, 1 for on (default)");
 135 module_param(mode, charp, 0);
 136 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 137                        "1 for active-backup, 2 for balance-xor, "
 138                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 139                        "6 for balance-alb");
 140 module_param(primary, charp, 0);
 141 MODULE_PARM_DESC(primary, "Primary network device to use");
 142 module_param(primary_reselect, charp, 0);
 143 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 144                                    "once it comes up; "
 145                                    "0 for always (default), "
 146                                    "1 for only if speed of primary is "
 147                                    "better, "
 148                                    "2 for only on active slave "
 149                                    "failure");
 150 module_param(lacp_rate, charp, 0);
 151 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 152                             "0 for slow, 1 for fast");
 153 module_param(ad_select, charp, 0);
 154 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 155                             "0 for stable (default), 1 for bandwidth, "
 156                             "2 for count");
 157 module_param(min_links, int, 0);
 158 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 159
 160 module_param(xmit_hash_policy, charp, 0);
 161 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 162                                    "0 for layer 2 (default), 1 for layer 3+4, "
 163                                    "2 for layer 2+3, 3 for encap layer 2+3, "
 164                                    "4 for encap layer 3+4");
 165 module_param(arp_interval, int, 0);
 166 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 167 module_param_array(arp_ip_target, charp, NULL, 0);
 168 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 169 module_param(arp_validate, charp, 0);
 170 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 171                                "0 for none (default), 1 for active, "
 172                                "2 for backup, 3 for all");
 173 module_param(arp_all_targets, charp, 0);
 174 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 175 module_param(fail_over_mac, charp, 0);
 176 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 177                                 "the same MAC; 0 for none (default), "
 178                                 "1 for active, 2 for follow");
 179 module_param(all_slaves_active, int, 0);
 180 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 181                                      "by setting active flag for all slaves; "
 182                                      "0 for never (default), 1 for always.");
 183 module_param(resend_igmp, int, 0);
 184 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 185                               "link failure");
 186 module_param(packets_per_slave, int, 0);
 187 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
 188                                     "mode; 0 for a random slave, 1 packet per "
 189                                     "slave (default), >1 packets per slave.");
 190 module_param(lp_interval, uint, 0);
 191 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
 192                               "the bonding driver sends learning packets to "
 193                               "each slaves peer switch. The default is 1.");
 194
 195 /*----------------------------- Global variables ----------------------------*/
 196
 197 #ifdef CONFIG_NET_POLL_CONTROLLER
 198 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 199 #endif
 200
 201 int bond_net_id __read_mostly;
 202
 203 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 204 static int arp_ip_count;
 205 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 206 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 207 static int lacp_fast;
 208
 209 /*-------------------------- Forward declarations ---------------------------*/
 210
 211 static int bond_init(struct net_device *bond_dev);
 212 static void bond_uninit(struct net_device *bond_dev);
 213
 214 /*---------------------------- General routines -----------------------------*/
 215
 216 const char *bond_mode_name(int mode)
 217 {
 218         static const char *names[] = {
 219                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 220                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 221                 [BOND_MODE_XOR] = "load balancing (xor)",
 222                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 223                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 224                 [BOND_MODE_TLB] = "transmit load balancing",
 225                 [BOND_MODE_ALB] = "adaptive load balancing",
 226         };
 227
 228         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 229                 return "unknown";
 230
 231         return names[mode];
 232 }
 233
 234 /*---------------------------------- VLAN -----------------------------------*/
 235
 236 /**
 237  * bond_dev_queue_xmit - Prepare skb for xmit.
 238  *
 239  * @bond: bond device that got this skb for tx.
 240  * @skb: hw accel VLAN tagged skb to transmit
 241  * @slave_dev: slave that is supposed to xmit this skbuff
 242  */
 243 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 244                         struct net_device *slave_dev)
 245 {
 246         skb->dev = slave_dev;
 247
 248         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 249                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 250         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 251
 252         if (unlikely(netpoll_tx_running(bond->dev)))
 253                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 254         else
 255                 dev_queue_xmit(skb);
 256 }
 257
 258 /*
 259  * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 260  * We don't protect the slave list iteration with a lock because:
 261  * a. This operation is performed in IOCTL context,
 262  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 263  * c. Holding a lock with BH disabled while directly calling a base driver
 264  *    entry point is generally a BAD idea.
 265  *
 266  * The design of synchronization/protection for this operation in the 8021q
 267  * module is good for one or more VLAN devices over a single physical device
 268  * and cannot be extended for a teaming solution like bonding, so there is a
 269  * potential race condition here where a net device from the vlan group might
 270  * be referenced (either by a base driver or the 8021q code) while it is being
 271  * removed from the system. However, it turns out we're not making matters
 272  * worse, and if it works for regular VLAN usage it will work here too.
 273 */
 274
 275 /**
 276  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 277  * @bond_dev: bonding net device that got called
 278  * @vid: vlan id being added
 279  */
 280 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 281                                 __be16 proto, u16 vid)
 282 {
 283         struct bonding *bond = netdev_priv(bond_dev);
 284         struct slave *slave, *rollback_slave;
 285         struct list_head *iter;
 286         int res;
 287
 288         bond_for_each_slave(bond, slave, iter) {
 289                 res = vlan_vid_add(slave->dev, proto, vid);
 290                 if (res)
 291                         goto unwind;
 292         }
 293
 294         return 0;
 295
 296 unwind:
 297         /* unwind to the slave that failed */
 298         bond_for_each_slave(bond, rollback_slave, iter) {
 299                 if (rollback_slave == slave)
 300                         break;
 301
 302                 vlan_vid_del(rollback_slave->dev, proto, vid);
 303         }
 304
 305         return res;
 306 }
 307
 308 /**
 309  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 310  * @bond_dev: bonding net device that got called
 311  * @vid: vlan id being removed
 312  */
 313 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 314                                  __be16 proto, u16 vid)
 315 {
 316         struct bonding *bond = netdev_priv(bond_dev);
 317         struct list_head *iter;
 318         struct slave *slave;
 319
 320         bond_for_each_slave(bond, slave, iter)
 321                 vlan_vid_del(slave->dev, proto, vid);
 322
 323         if (bond_is_lb(bond))
 324                 bond_alb_clear_vlan(bond, vid);
 325
 326         return 0;
 327 }
 328
 329 /*------------------------------- Link status -------------------------------*/
 330
 331 /*
 332  * Set the carrier state for the master according to the state of its
 333  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 334  * do special 802.3ad magic.
 335  *
 336  * Returns zero if carrier state does not change, nonzero if it does.
 337  */
 338 static int bond_set_carrier(struct bonding *bond)
 339 {
 340         struct list_head *iter;
 341         struct slave *slave;
 342
 343         if (!bond_has_slaves(bond))
 344                 goto down;
 345
 346         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 347                 return bond_3ad_set_carrier(bond);
 348
 349         bond_for_each_slave(bond, slave, iter) {
 350                 if (slave->link == BOND_LINK_UP) {
 351                         if (!netif_carrier_ok(bond->dev)) {
 352                                 netif_carrier_on(bond->dev);
 353                                 return 1;
 354                         }
 355                         return 0;
 356                 }
 357         }
 358
 359 down:
 360         if (netif_carrier_ok(bond->dev)) {
 361                 netif_carrier_off(bond->dev);
 362                 return 1;
 363         }
 364         return 0;
 365 }
 366
 367 /*
 368  * Get link speed and duplex from the slave's base driver
 369  * using ethtool. If for some reason the call fails or the
 370  * values are invalid, set speed and duplex to -1,
 371  * and return.
 372  */
 373 static void bond_update_speed_duplex(struct slave *slave)
 374 {
 375         struct net_device *slave_dev = slave->dev;
 376         struct ethtool_cmd ecmd;
 377         u32 slave_speed;
 378         int res;
 379
 380         slave->speed = SPEED_UNKNOWN;
 381         slave->duplex = DUPLEX_UNKNOWN;
 382
 383         res = __ethtool_get_settings(slave_dev, &ecmd);
 384         if (res < 0)
 385                 return;
 386
 387         slave_speed = ethtool_cmd_speed(&ecmd);
 388         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 389                 return;
 390
 391         switch (ecmd.duplex) {
 392         case DUPLEX_FULL:
 393         case DUPLEX_HALF:
 394                 break;
 395         default:
 396                 return;
 397         }
 398
 399         slave->speed = slave_speed;
 400         slave->duplex = ecmd.duplex;
 401
 402         return;
 403 }
 404
 405 const char *bond_slave_link_status(s8 link)
 406 {
 407         switch (link) {
 408         case BOND_LINK_UP:
 409                 return "up";
 410         case BOND_LINK_FAIL:
 411                 return "going down";
 412         case BOND_LINK_DOWN:
 413                 return "down";
 414         case BOND_LINK_BACK:
 415                 return "going back";
 416         default:
 417                 return "unknown";
 418         }
 419 }
 420
 421 /*
 422  * if <dev> supports MII link status reporting, check its link status.
 423  *
 424  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 425  * depending upon the setting of the use_carrier parameter.
 426  *
 427  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 428  * can't tell and just pretend it is), or 0, meaning that the link is
 429  * down.
 430  *
 431  * If reporting is non-zero, instead of faking link up, return -1 if
 432  * both ETHTOOL and MII ioctls fail (meaning the device does not
 433  * support them).  If use_carrier is set, return whatever it says.
 434  * It'd be nice if there was a good way to tell if a driver supports
 435  * netif_carrier, but there really isn't.
 436  */
 437 static int bond_check_dev_link(struct bonding *bond,
 438                                struct net_device *slave_dev, int reporting)
 439 {
 440         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 441         int (*ioctl)(struct net_device *, struct ifreq *, int);
 442         struct ifreq ifr;
 443         struct mii_ioctl_data *mii;
 444
 445         if (!reporting && !netif_running(slave_dev))
 446                 return 0;
 447
 448         if (bond->params.use_carrier)
 449                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 450
 451         /* Try to get link status using Ethtool first. */
 452         if (slave_dev->ethtool_ops->get_link)
 453                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 454                         BMSR_LSTATUS : 0;
 455
 456         /* Ethtool can't be used, fallback to MII ioctls. */
 457         ioctl = slave_ops->ndo_do_ioctl;
 458         if (ioctl) {
 459                 /* TODO: set pointer to correct ioctl on a per team member */
 460                 /*       bases to make this more efficient. that is, once  */
 461                 /*       we determine the correct ioctl, we will always    */
 462                 /*       call it and not the others for that team          */
 463                 /*       member.                                           */
 464
 465                 /*
 466                  * We cannot assume that SIOCGMIIPHY will also read a
 467                  * register; not all network drivers (e.g., e100)
 468                  * support that.
 469                  */
 470
 471                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 472                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 473                 mii = if_mii(&ifr);
 474                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 475                         mii->reg_num = MII_BMSR;
 476                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 477                                 return mii->val_out & BMSR_LSTATUS;
 478                 }
 479         }
 480
 481         /*
 482          * If reporting, report that either there's no dev->do_ioctl,
 483          * or both SIOCGMIIREG and get_link failed (meaning that we
 484          * cannot report link status).  If not reporting, pretend
 485          * we're ok.
 486          */
 487         return reporting ? -1 : BMSR_LSTATUS;
 488 }
 489
 490 /*----------------------------- Multicast list ------------------------------*/
 491
 492 /*
 493  * Push the promiscuity flag down to appropriate slaves
 494  */
 495 static int bond_set_promiscuity(struct bonding *bond, int inc)
 496 {
 497         struct list_head *iter;
 498         int err = 0;
 499
 500         if (bond_uses_primary(bond)) {
 501                 struct slave *curr_active = bond_deref_active_protected(bond);
 502
 503                 /* write lock already acquired */
 504                 if (curr_active)
 505                         err = dev_set_promiscuity(curr_active->dev, inc);
 506         } else {
 507                 struct slave *slave;
 508
 509                 bond_for_each_slave(bond, slave, iter) {
 510                         err = dev_set_promiscuity(slave->dev, inc);
 511                         if (err)
 512                                 return err;
 513                 }
 514         }
 515         return err;
 516 }
 517
 518 /*
 519  * Push the allmulti flag down to all slaves
 520  */
 521 static int bond_set_allmulti(struct bonding *bond, int inc)
 522 {
 523         struct list_head *iter;
 524         int err = 0;
 525
 526         if (bond_uses_primary(bond)) {
 527                 struct slave *curr_active = bond_deref_active_protected(bond);
 528
 529                 /* write lock already acquired */
 530                 if (curr_active)
 531                         err = dev_set_allmulti(curr_active->dev, inc);
 532         } else {
 533                 struct slave *slave;
 534
 535                 bond_for_each_slave(bond, slave, iter) {
 536                         err = dev_set_allmulti(slave->dev, inc);
 537                         if (err)
 538                                 return err;
 539                 }
 540         }
 541         return err;
 542 }
 543
 544 /*
 545  * Retrieve the list of registered multicast addresses for the bonding
 546  * device and retransmit an IGMP JOIN request to the current active
 547  * slave.
 548  */
 549 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 550 {
 551         struct bonding *bond = container_of(work, struct bonding,
 552                                             mcast_work.work);
 553
 554         if (!rtnl_trylock()) {
 555                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 556                 return;
 557         }
 558         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 559
 560         if (bond->igmp_retrans > 1) {
 561                 bond->igmp_retrans--;
 562                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 563         }
 564         rtnl_unlock();
 565 }
 566
 567 /* Flush bond's hardware addresses from slave
 568  */
 569 static void bond_hw_addr_flush(struct net_device *bond_dev,
 570                                struct net_device *slave_dev)
 571 {
 572         struct bonding *bond = netdev_priv(bond_dev);
 573
 574         dev_uc_unsync(slave_dev, bond_dev);
 575         dev_mc_unsync(slave_dev, bond_dev);
 576
 577         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 578                 /* del lacpdu mc addr from mc list */
 579                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 580
 581                 dev_mc_del(slave_dev, lacpdu_multicast);
 582         }
 583 }
 584
 585 /*--------------------------- Active slave change ---------------------------*/
 586
 587 /* Update the hardware address list and promisc/allmulti for the new and
 588  * old active slaves (if any).  Modes that are not using primary keep all
 589  * slaves up date at all times; only the modes that use primary need to call
 590  * this function to swap these settings during a failover.
 591  */
 592 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 593                               struct slave *old_active)
 594 {
 595         ASSERT_RTNL();
 596
 597         if (old_active) {
 598                 if (bond->dev->flags & IFF_PROMISC)
 599                         dev_set_promiscuity(old_active->dev, -1);
 600
 601                 if (bond->dev->flags & IFF_ALLMULTI)
 602                         dev_set_allmulti(old_active->dev, -1);
 603
 604                 bond_hw_addr_flush(bond->dev, old_active->dev);
 605         }
 606
 607         if (new_active) {
 608                 /* FIXME: Signal errors upstream. */
 609                 if (bond->dev->flags & IFF_PROMISC)
 610                         dev_set_promiscuity(new_active->dev, 1);
 611
 612                 if (bond->dev->flags & IFF_ALLMULTI)
 613                         dev_set_allmulti(new_active->dev, 1);
 614
 615                 netif_addr_lock_bh(bond->dev);
 616                 dev_uc_sync(new_active->dev, bond->dev);
 617                 dev_mc_sync(new_active->dev, bond->dev);
 618                 netif_addr_unlock_bh(bond->dev);
 619         }
 620 }
 621
 622 /**
 623  * bond_set_dev_addr - clone slave's address to bond
 624  * @bond_dev: bond net device
 625  * @slave_dev: slave net device
 626  *
 627  * Should be called with RTNL held.
 628  */
 629 static void bond_set_dev_addr(struct net_device *bond_dev,
 630                               struct net_device *slave_dev)
 631 {
 632         pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 633                  bond_dev, slave_dev, slave_dev->addr_len);
 634         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 635         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 636         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 637 }
 638
 639 /*
 640  * bond_do_fail_over_mac
 641  *
 642  * Perform special MAC address swapping for fail_over_mac settings
 643  *
 644  * Called with RTNL, curr_slave_lock for write_bh.
 645  */
 646 static void bond_do_fail_over_mac(struct bonding *bond,
 647                                   struct slave *new_active,
 648                                   struct slave *old_active)
 649         __releases(&bond->curr_slave_lock)
 650         __acquires(&bond->curr_slave_lock)
 651 {
 652         u8 tmp_mac[ETH_ALEN];
 653         struct sockaddr saddr;
 654         int rv;
 655
 656         switch (bond->params.fail_over_mac) {
 657         case BOND_FOM_ACTIVE:
 658                 if (new_active) {
 659                         write_unlock_bh(&bond->curr_slave_lock);
 660                         bond_set_dev_addr(bond->dev, new_active->dev);
 661                         write_lock_bh(&bond->curr_slave_lock);
 662                 }
 663                 break;
 664         case BOND_FOM_FOLLOW:
 665                 /*
 666                  * if new_active && old_active, swap them
 667                  * if just old_active, do nothing (going to no active slave)
 668                  * if just new_active, set new_active to bond's MAC
 669                  */
 670                 if (!new_active)
 671                         return;
 672
 673                 write_unlock_bh(&bond->curr_slave_lock);
 674
 675                 if (old_active) {
 676                         ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
 677                         ether_addr_copy(saddr.sa_data,
 678                                         old_active->dev->dev_addr);
 679                         saddr.sa_family = new_active->dev->type;
 680                 } else {
 681                         ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
 682                         saddr.sa_family = bond->dev->type;
 683                 }
 684
 685                 rv = dev_set_mac_address(new_active->dev, &saddr);
 686                 if (rv) {
 687                         pr_err("%s: Error %d setting MAC of slave %s\n",
 688                                bond->dev->name, -rv, new_active->dev->name);
 689                         goto out;
 690                 }
 691
 692                 if (!old_active)
 693                         goto out;
 694
 695                 ether_addr_copy(saddr.sa_data, tmp_mac);
 696                 saddr.sa_family = old_active->dev->type;
 697
 698                 rv = dev_set_mac_address(old_active->dev, &saddr);
 699                 if (rv)
 700                         pr_err("%s: Error %d setting MAC of slave %s\n",
 701                                bond->dev->name, -rv, new_active->dev->name);
 702 out:
 703                 write_lock_bh(&bond->curr_slave_lock);
 704                 break;
 705         default:
 706                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 707                        bond->dev->name, bond->params.fail_over_mac);
 708                 break;
 709         }
 710
 711 }
 712
 713 static bool bond_should_change_active(struct bonding *bond)
 714 {
 715         struct slave *prim = bond->primary_slave;
 716         struct slave *curr = bond_deref_active_protected(bond);
 717
 718         if (!prim || !curr || curr->link != BOND_LINK_UP)
 719                 return true;
 720         if (bond->force_primary) {
 721                 bond->force_primary = false;
 722                 return true;
 723         }
 724         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 725             (prim->speed < curr->speed ||
 726              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 727                 return false;
 728         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 729                 return false;
 730         return true;
 731 }
 732
 733 /**
 734  * find_best_interface - select the best available slave to be the active one
 735  * @bond: our bonding struct
 736  */
 737 static struct slave *bond_find_best_slave(struct bonding *bond)
 738 {
 739         struct slave *slave, *bestslave = NULL;
 740         struct list_head *iter;
 741         int mintime = bond->params.updelay;
 742
 743         if (bond->primary_slave && bond->primary_slave->link == BOND_LINK_UP &&
 744             bond_should_change_active(bond))
 745                 return bond->primary_slave;
 746
 747         bond_for_each_slave(bond, slave, iter) {
 748                 if (slave->link == BOND_LINK_UP)
 749                         return slave;
 750                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
 751                     slave->delay < mintime) {
 752                         mintime = slave->delay;
 753                         bestslave = slave;
 754                 }
 755         }
 756
 757         return bestslave;
 758 }
 759
 760 static bool bond_should_notify_peers(struct bonding *bond)
 761 {
 762         struct slave *slave;
 763
 764         rcu_read_lock();
 765         slave = rcu_dereference(bond->curr_active_slave);
 766         rcu_read_unlock();
 767
 768         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
 769                  bond->dev->name, slave ? slave->dev->name : "NULL");
 770
 771         if (!slave || !bond->send_peer_notif ||
 772             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 773                 return false;
 774
 775         return true;
 776 }
 777
 778 /**
 779  * change_active_interface - change the active slave into the specified one
 780  * @bond: our bonding struct
 781  * @new: the new slave to make the active one
 782  *
 783  * Set the new slave to the bond's settings and unset them on the old
 784  * curr_active_slave.
 785  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 786  *
 787  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 788  * because it is apparently the best available slave we have, even though its
 789  * updelay hasn't timed out yet.
 790  *
 791  * If new_active is not NULL, caller must hold curr_slave_lock for write_bh.
 792  */
 793 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 794 {
 795         struct slave *old_active;
 796
 797         old_active = rcu_dereference_protected(bond->curr_active_slave,
 798                                                !new_active ||
 799                                                lockdep_is_held(&bond->curr_slave_lock));
 800
 801         if (old_active == new_active)
 802                 return;
 803
 804         if (new_active) {
 805                 new_active->last_link_up = jiffies;
 806
 807                 if (new_active->link == BOND_LINK_BACK) {
 808                         if (bond_uses_primary(bond)) {
 809                                 pr_info("%s: making interface %s the new active one %d ms earlier\n",
 810                                         bond->dev->name, new_active->dev->name,
 811                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
 812                         }
 813
 814                         new_active->delay = 0;
 815                         new_active->link = BOND_LINK_UP;
 816
 817                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 818                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 819
 820                         if (bond_is_lb(bond))
 821                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 822                 } else {
 823                         if (bond_uses_primary(bond)) {
 824                                 pr_info("%s: making interface %s the new active one\n",
 825                                         bond->dev->name, new_active->dev->name);
 826                         }
 827                 }
 828         }
 829
 830         if (bond_uses_primary(bond))
 831                 bond_hw_addr_swap(bond, new_active, old_active);
 832
 833         if (bond_is_lb(bond)) {
 834                 bond_alb_handle_active_change(bond, new_active);
 835                 if (old_active)
 836                         bond_set_slave_inactive_flags(old_active,
 837                                                       BOND_SLAVE_NOTIFY_NOW);
 838                 if (new_active)
 839                         bond_set_slave_active_flags(new_active,
 840                                                     BOND_SLAVE_NOTIFY_NOW);
 841         } else {
 842                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 843         }
 844
 845         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
 846                 if (old_active)
 847                         bond_set_slave_inactive_flags(old_active,
 848                                                       BOND_SLAVE_NOTIFY_NOW);
 849
 850                 if (new_active) {
 851                         bool should_notify_peers = false;
 852
 853                         bond_set_slave_active_flags(new_active,
 854                                                     BOND_SLAVE_NOTIFY_NOW);
 855
 856                         if (bond->params.fail_over_mac)
 857                                 bond_do_fail_over_mac(bond, new_active,
 858                                                       old_active);
 859
 860                         if (netif_running(bond->dev)) {
 861                                 bond->send_peer_notif =
 862                                         bond->params.num_peer_notif;
 863                                 should_notify_peers =
 864                                         bond_should_notify_peers(bond);
 865                         }
 866
 867                         write_unlock_bh(&bond->curr_slave_lock);
 868
 869                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 870                         if (should_notify_peers)
 871                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 872                                                          bond->dev);
 873
 874                         write_lock_bh(&bond->curr_slave_lock);
 875                 }
 876         }
 877
 878         /* resend IGMP joins since active slave has changed or
 879          * all were sent on curr_active_slave.
 880          * resend only if bond is brought up with the affected
 881          * bonding modes and the retransmission is enabled */
 882         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 883             ((bond_uses_primary(bond) && new_active) ||
 884              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
 885                 bond->igmp_retrans = bond->params.resend_igmp;
 886                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 887         }
 888 }
 889
 890 /**
 891  * bond_select_active_slave - select a new active slave, if needed
 892  * @bond: our bonding struct
 893  *
 894  * This functions should be called when one of the following occurs:
 895  * - The old curr_active_slave has been released or lost its link.
 896  * - The primary_slave has got its link back.
 897  * - A slave has got its link back and there's no old curr_active_slave.
 898  *
 899  * Caller must hold curr_slave_lock for write_bh.
 900  */
 901 void bond_select_active_slave(struct bonding *bond)
 902 {
 903         struct slave *best_slave;
 904         int rv;
 905
 906         best_slave = bond_find_best_slave(bond);
 907         if (best_slave != bond_deref_active_protected(bond)) {
 908                 bond_change_active_slave(bond, best_slave);
 909                 rv = bond_set_carrier(bond);
 910                 if (!rv)
 911                         return;
 912
 913                 if (netif_carrier_ok(bond->dev)) {
 914                         pr_info("%s: first active interface up!\n",
 915                                 bond->dev->name);
 916                 } else {
 917                         pr_info("%s: now running without any active interface!\n",
 918                                 bond->dev->name);
 919                 }
 920         }
 921 }
 922
 923 #ifdef CONFIG_NET_POLL_CONTROLLER
 924 static inline int slave_enable_netpoll(struct slave *slave)
 925 {
 926         struct netpoll *np;
 927         int err = 0;
 928
 929         np = kzalloc(sizeof(*np), GFP_KERNEL);
 930         err = -ENOMEM;
 931         if (!np)
 932                 goto out;
 933
 934         err = __netpoll_setup(np, slave->dev);
 935         if (err) {
 936                 kfree(np);
 937                 goto out;
 938         }
 939         slave->np = np;
 940 out:
 941         return err;
 942 }
 943 static inline void slave_disable_netpoll(struct slave *slave)
 944 {
 945         struct netpoll *np = slave->np;
 946
 947         if (!np)
 948                 return;
 949
 950         slave->np = NULL;
 951         __netpoll_free_async(np);
 952 }
 953
 954 static void bond_poll_controller(struct net_device *bond_dev)
 955 {
 956 }
 957
 958 static void bond_netpoll_cleanup(struct net_device *bond_dev)
 959 {
 960         struct bonding *bond = netdev_priv(bond_dev);
 961         struct list_head *iter;
 962         struct slave *slave;
 963
 964         bond_for_each_slave(bond, slave, iter)
 965                 if (bond_slave_is_up(slave))
 966                         slave_disable_netpoll(slave);
 967 }
 968
 969 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
 970 {
 971         struct bonding *bond = netdev_priv(dev);
 972         struct list_head *iter;
 973         struct slave *slave;
 974         int err = 0;
 975
 976         bond_for_each_slave(bond, slave, iter) {
 977                 err = slave_enable_netpoll(slave);
 978                 if (err) {
 979                         bond_netpoll_cleanup(dev);
 980                         break;
 981                 }
 982         }
 983         return err;
 984 }
 985 #else
 986 static inline int slave_enable_netpoll(struct slave *slave)
 987 {
 988         return 0;
 989 }
 990 static inline void slave_disable_netpoll(struct slave *slave)
 991 {
 992 }
 993 static void bond_netpoll_cleanup(struct net_device *bond_dev)
 994 {
 995 }
 996 #endif
 997
 998 /*---------------------------------- IOCTL ----------------------------------*/
 999
1000 static netdev_features_t bond_fix_features(struct net_device *dev,
1001                                            netdev_features_t features)
1002 {
1003         struct bonding *bond = netdev_priv(dev);
1004         struct list_head *iter;
1005         netdev_features_t mask;
1006         struct slave *slave;
1007
1008         mask = features;
1009         features &= ~NETIF_F_ONE_FOR_ALL;
1010         features |= NETIF_F_ALL_FOR_ALL;
1011
1012         bond_for_each_slave(bond, slave, iter) {
1013                 features = netdev_increment_features(features,
1014                                                      slave->dev->features,
1015                                                      mask);
1016         }
1017         features = netdev_add_tso_features(features, mask);
1018
1019         return features;
1020 }
1021
1022 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1023                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1024                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1025
1026 #define BOND_ENC_FEATURES       (NETIF_F_ALL_CSUM | NETIF_F_SG | NETIF_F_RXCSUM |\
1027                                  NETIF_F_TSO | NETIF_F_GSO_UDP_TUNNEL)
1028
1029 static void bond_compute_features(struct bonding *bond)
1030 {
1031         unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1032         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1033         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1034         struct net_device *bond_dev = bond->dev;
1035         struct list_head *iter;
1036         struct slave *slave;
1037         unsigned short max_hard_header_len = ETH_HLEN;
1038         unsigned int gso_max_size = GSO_MAX_SIZE;
1039         u16 gso_max_segs = GSO_MAX_SEGS;
1040
1041         if (!bond_has_slaves(bond))
1042                 goto done;
1043         vlan_features &= NETIF_F_ALL_FOR_ALL;
1044
1045         bond_for_each_slave(bond, slave, iter) {
1046                 vlan_features = netdev_increment_features(vlan_features,
1047                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1048
1049                 enc_features = netdev_increment_features(enc_features,
1050                                                          slave->dev->hw_enc_features,
1051                                                          BOND_ENC_FEATURES);
1052                 dst_release_flag &= slave->dev->priv_flags;
1053                 if (slave->dev->hard_header_len > max_hard_header_len)
1054                         max_hard_header_len = slave->dev->hard_header_len;
1055
1056                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1057                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1058         }
1059
1060 done:
1061         bond_dev->vlan_features = vlan_features;
1062         bond_dev->hw_enc_features = enc_features;
1063         bond_dev->hard_header_len = max_hard_header_len;
1064         bond_dev->gso_max_segs = gso_max_segs;
1065         netif_set_gso_max_size(bond_dev, gso_max_size);
1066
1067         flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1068         bond_dev->priv_flags = flags | dst_release_flag;
1069
1070         netdev_change_features(bond_dev);
1071 }
1072
1073 static void bond_setup_by_slave(struct net_device *bond_dev,
1074                                 struct net_device *slave_dev)
1075 {
1076         bond_dev->header_ops        = slave_dev->header_ops;
1077
1078         bond_dev->type              = slave_dev->type;
1079         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1080         bond_dev->addr_len          = slave_dev->addr_len;
1081
1082         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1083                 slave_dev->addr_len);
1084 }
1085
1086 /* On bonding slaves other than the currently active slave, suppress
1087  * duplicates except for alb non-mcast/bcast.
1088  */
1089 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1090                                             struct slave *slave,
1091                                             struct bonding *bond)
1092 {
1093         if (bond_is_slave_inactive(slave)) {
1094                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1095                     skb->pkt_type != PACKET_BROADCAST &&
1096                     skb->pkt_type != PACKET_MULTICAST)
1097                         return false;
1098                 return true;
1099         }
1100         return false;
1101 }
1102
1103 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1104 {
1105         struct sk_buff *skb = *pskb;
1106         struct slave *slave;
1107         struct bonding *bond;
1108         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1109                           struct slave *);
1110         int ret = RX_HANDLER_ANOTHER;
1111
1112         skb = skb_share_check(skb, GFP_ATOMIC);
1113         if (unlikely(!skb))
1114                 return RX_HANDLER_CONSUMED;
1115
1116         *pskb = skb;
1117
1118         slave = bond_slave_get_rcu(skb->dev);
1119         bond = slave->bond;
1120
1121         recv_probe = ACCESS_ONCE(bond->recv_probe);
1122         if (recv_probe) {
1123                 ret = recv_probe(skb, bond, slave);
1124                 if (ret == RX_HANDLER_CONSUMED) {
1125                         consume_skb(skb);
1126                         return ret;
1127                 }
1128         }
1129
1130         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1131                 return RX_HANDLER_EXACT;
1132         }
1133
1134         skb->dev = bond->dev;
1135
1136         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1137             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1138             skb->pkt_type == PACKET_HOST) {
1139
1140                 if (unlikely(skb_cow_head(skb,
1141                                           skb->data - skb_mac_header(skb)))) {
1142                         kfree_skb(skb);
1143                         return RX_HANDLER_CONSUMED;
1144                 }
1145                 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1146         }
1147
1148         return ret;
1149 }
1150
1151 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1152                                       struct net_device *slave_dev,
1153                                       struct slave *slave)
1154 {
1155         int err;
1156
1157         err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1158         if (err)
1159                 return err;
1160         slave_dev->flags |= IFF_SLAVE;
1161         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1162         return 0;
1163 }
1164
1165 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1166                                   struct net_device *slave_dev)
1167 {
1168         netdev_upper_dev_unlink(slave_dev, bond_dev);
1169         slave_dev->flags &= ~IFF_SLAVE;
1170         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1171 }
1172
1173 static struct slave *bond_alloc_slave(struct bonding *bond)
1174 {
1175         struct slave *slave = NULL;
1176
1177         slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1178         if (!slave)
1179                 return NULL;
1180
1181         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1182                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1183                                                GFP_KERNEL);
1184                 if (!SLAVE_AD_INFO(slave)) {
1185                         kfree(slave);
1186                         return NULL;
1187                 }
1188         }
1189         return slave;
1190 }
1191
1192 static void bond_free_slave(struct slave *slave)
1193 {
1194         struct bonding *bond = bond_get_bond_by_slave(slave);
1195
1196         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1197                 kfree(SLAVE_AD_INFO(slave));
1198
1199         kfree(slave);
1200 }
1201
1202 /* enslave device <slave> to bond device <master> */
1203 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1204 {
1205         struct bonding *bond = netdev_priv(bond_dev);
1206         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1207         struct slave *new_slave = NULL, *prev_slave;
1208         struct sockaddr addr;
1209         int link_reporting;
1210         int res = 0, i;
1211
1212         if (!bond->params.use_carrier &&
1213             slave_dev->ethtool_ops->get_link == NULL &&
1214             slave_ops->ndo_do_ioctl == NULL) {
1215                 pr_warn("%s: Warning: no link monitoring support for %s\n",
1216                         bond_dev->name, slave_dev->name);
1217         }
1218
1219         /* already enslaved */
1220         if (slave_dev->flags & IFF_SLAVE) {
1221                 pr_debug("Error: Device was already enslaved\n");
1222                 return -EBUSY;
1223         }
1224
1225         if (bond_dev == slave_dev) {
1226                 pr_err("%s: cannot enslave bond to itself.\n", bond_dev->name);
1227                 return -EPERM;
1228         }
1229
1230         /* vlan challenged mutual exclusion */
1231         /* no need to lock since we're protected by rtnl_lock */
1232         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1233                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1234                 if (vlan_uses_dev(bond_dev)) {
1235                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1236                                bond_dev->name, slave_dev->name, bond_dev->name);
1237                         return -EPERM;
1238                 } else {
1239                         pr_warn("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1240                                 bond_dev->name, slave_dev->name,
1241                                 slave_dev->name, bond_dev->name);
1242                 }
1243         } else {
1244                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1245         }
1246
1247         /*
1248          * Old ifenslave binaries are no longer supported.  These can
1249          * be identified with moderate accuracy by the state of the slave:
1250          * the current ifenslave will set the interface down prior to
1251          * enslaving it; the old ifenslave will not.
1252          */
1253         if ((slave_dev->flags & IFF_UP)) {
1254                 pr_err("%s is up - this may be due to an out of date ifenslave\n",
1255                        slave_dev->name);
1256                 res = -EPERM;
1257                 goto err_undo_flags;
1258         }
1259
1260         /* set bonding device ether type by slave - bonding netdevices are
1261          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1262          * there is a need to override some of the type dependent attribs/funcs.
1263          *
1264          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1265          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1266          */
1267         if (!bond_has_slaves(bond)) {
1268                 if (bond_dev->type != slave_dev->type) {
1269                         pr_debug("%s: change device type from %d to %d\n",
1270                                  bond_dev->name,
1271                                  bond_dev->type, slave_dev->type);
1272
1273                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1274                                                        bond_dev);
1275                         res = notifier_to_errno(res);
1276                         if (res) {
1277                                 pr_err("%s: refused to change device type\n",
1278                                        bond_dev->name);
1279                                 res = -EBUSY;
1280                                 goto err_undo_flags;
1281                         }
1282
1283                         /* Flush unicast and multicast addresses */
1284                         dev_uc_flush(bond_dev);
1285                         dev_mc_flush(bond_dev);
1286
1287                         if (slave_dev->type != ARPHRD_ETHER)
1288                                 bond_setup_by_slave(bond_dev, slave_dev);
1289                         else {
1290                                 ether_setup(bond_dev);
1291                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1292                         }
1293
1294                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1295                                                  bond_dev);
1296                 }
1297         } else if (bond_dev->type != slave_dev->type) {
1298                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1299                        slave_dev->name, slave_dev->type, bond_dev->type);
1300                 res = -EINVAL;
1301                 goto err_undo_flags;
1302         }
1303
1304         if (slave_ops->ndo_set_mac_address == NULL) {
1305                 if (!bond_has_slaves(bond)) {
1306                         pr_warn("%s: Warning: The first slave device specified does not support setting the MAC address\n",
1307                                 bond_dev->name);
1308                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
1309                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1310                                 pr_warn("%s: Setting fail_over_mac to active for active-backup mode\n",
1311                                         bond_dev->name);
1312                         }
1313                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1314                         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",
1315                                bond_dev->name);
1316                         res = -EOPNOTSUPP;
1317                         goto err_undo_flags;
1318                 }
1319         }
1320
1321         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1322
1323         /* If this is the first slave, then we need to set the master's hardware
1324          * address to be the same as the slave's. */
1325         if (!bond_has_slaves(bond) &&
1326             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1327                 bond_set_dev_addr(bond->dev, slave_dev);
1328
1329         new_slave = bond_alloc_slave(bond);
1330         if (!new_slave) {
1331                 res = -ENOMEM;
1332                 goto err_undo_flags;
1333         }
1334
1335         new_slave->bond = bond;
1336         new_slave->dev = slave_dev;
1337         /*
1338          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1339          * is set via sysfs or module option if desired.
1340          */
1341         new_slave->queue_id = 0;
1342
1343         /* Save slave's original mtu and then set it to match the bond */
1344         new_slave->original_mtu = slave_dev->mtu;
1345         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1346         if (res) {
1347                 pr_debug("Error %d calling dev_set_mtu\n", res);
1348                 goto err_free;
1349         }
1350
1351         /*
1352          * Save slave's original ("permanent") mac address for modes
1353          * that need it, and for restoring it upon release, and then
1354          * set it to the master's address
1355          */
1356         ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1357
1358         if (!bond->params.fail_over_mac ||
1359             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1360                 /*
1361                  * Set slave to master's mac address.  The application already
1362                  * set the master's mac address to that of the first slave
1363                  */
1364                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1365                 addr.sa_family = slave_dev->type;
1366                 res = dev_set_mac_address(slave_dev, &addr);
1367                 if (res) {
1368                         pr_debug("Error %d calling set_mac_address\n", res);
1369                         goto err_restore_mtu;
1370                 }
1371         }
1372
1373         /* open the slave since the application closed it */
1374         res = dev_open(slave_dev);
1375         if (res) {
1376                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1377                 goto err_restore_mac;
1378         }
1379
1380         slave_dev->priv_flags |= IFF_BONDING;
1381
1382         if (bond_is_lb(bond)) {
1383                 /* bond_alb_init_slave() must be called before all other stages since
1384                  * it might fail and we do not want to have to undo everything
1385                  */
1386                 res = bond_alb_init_slave(bond, new_slave);
1387                 if (res)
1388                         goto err_close;
1389         }
1390
1391         /* If the mode uses primary, then the following is handled by
1392          * bond_change_active_slave().
1393          */
1394         if (!bond_uses_primary(bond)) {
1395                 /* set promiscuity level to new slave */
1396                 if (bond_dev->flags & IFF_PROMISC) {
1397                         res = dev_set_promiscuity(slave_dev, 1);
1398                         if (res)
1399                                 goto err_close;
1400                 }
1401
1402                 /* set allmulti level to new slave */
1403                 if (bond_dev->flags & IFF_ALLMULTI) {
1404                         res = dev_set_allmulti(slave_dev, 1);
1405                         if (res)
1406                                 goto err_close;
1407                 }
1408
1409                 netif_addr_lock_bh(bond_dev);
1410
1411                 dev_mc_sync_multiple(slave_dev, bond_dev);
1412                 dev_uc_sync_multiple(slave_dev, bond_dev);
1413
1414                 netif_addr_unlock_bh(bond_dev);
1415         }
1416
1417         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1418                 /* add lacpdu mc addr to mc list */
1419                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1420
1421                 dev_mc_add(slave_dev, lacpdu_multicast);
1422         }
1423
1424         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1425         if (res) {
1426                 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1427                        bond_dev->name, slave_dev->name);
1428                 goto err_close;
1429         }
1430
1431         prev_slave = bond_last_slave(bond);
1432
1433         new_slave->delay = 0;
1434         new_slave->link_failure_count = 0;
1435
1436         bond_update_speed_duplex(new_slave);
1437
1438         new_slave->last_rx = jiffies -
1439                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1440         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1441                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1442
1443         if (bond->params.miimon && !bond->params.use_carrier) {
1444                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1445
1446                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1447                         /*
1448                          * miimon is set but a bonded network driver
1449                          * does not support ETHTOOL/MII and
1450                          * arp_interval is not set.  Note: if
1451                          * use_carrier is enabled, we will never go
1452                          * here (because netif_carrier is always
1453                          * supported); thus, we don't need to change
1454                          * the messages for netif_carrier.
1455                          */
1456                         pr_warn("%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",
1457                                 bond_dev->name, slave_dev->name);
1458                 } else if (link_reporting == -1) {
1459                         /* unable get link status using mii/ethtool */
1460                         pr_warn("%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",
1461                                 bond_dev->name, slave_dev->name);
1462                 }
1463         }
1464
1465         /* check for initial state */
1466         if (bond->params.miimon) {
1467                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1468                         if (bond->params.updelay) {
1469                                 new_slave->link = BOND_LINK_BACK;
1470                                 new_slave->delay = bond->params.updelay;
1471                         } else {
1472                                 new_slave->link = BOND_LINK_UP;
1473                         }
1474                 } else {
1475                         new_slave->link = BOND_LINK_DOWN;
1476                 }
1477         } else if (bond->params.arp_interval) {
1478                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1479                         BOND_LINK_UP : BOND_LINK_DOWN);
1480         } else {
1481                 new_slave->link = BOND_LINK_UP;
1482         }
1483
1484         if (new_slave->link != BOND_LINK_DOWN)
1485                 new_slave->last_link_up = jiffies;
1486         pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1487                  new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1488                  (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1489
1490         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1491                 /* if there is a primary slave, remember it */
1492                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1493                         bond->primary_slave = new_slave;
1494                         bond->force_primary = true;
1495                 }
1496         }
1497
1498         switch (BOND_MODE(bond)) {
1499         case BOND_MODE_ACTIVEBACKUP:
1500                 bond_set_slave_inactive_flags(new_slave,
1501                                               BOND_SLAVE_NOTIFY_NOW);
1502                 break;
1503         case BOND_MODE_8023AD:
1504                 /* in 802.3ad mode, the internal mechanism
1505                  * will activate the slaves in the selected
1506                  * aggregator
1507                  */
1508                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1509                 /* if this is the first slave */
1510                 if (!prev_slave) {
1511                         SLAVE_AD_INFO(new_slave)->id = 1;
1512                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1513                          * can be called only after the mac address of the bond is set
1514                          */
1515                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1516                 } else {
1517                         SLAVE_AD_INFO(new_slave)->id =
1518                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1519                 }
1520
1521                 bond_3ad_bind_slave(new_slave);
1522                 break;
1523         case BOND_MODE_TLB:
1524         case BOND_MODE_ALB:
1525                 bond_set_active_slave(new_slave);
1526                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1527                 break;
1528         default:
1529                 pr_debug("This slave is always active in trunk mode\n");
1530
1531                 /* always active in trunk mode */
1532                 bond_set_active_slave(new_slave);
1533
1534                 /* In trunking mode there is little meaning to curr_active_slave
1535                  * anyway (it holds no special properties of the bond device),
1536                  * so we can change it without calling change_active_interface()
1537                  */
1538                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1539                     new_slave->link == BOND_LINK_UP)
1540                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1541
1542                 break;
1543         } /* switch(bond_mode) */
1544
1545 #ifdef CONFIG_NET_POLL_CONTROLLER
1546         slave_dev->npinfo = bond->dev->npinfo;
1547         if (slave_dev->npinfo) {
1548                 if (slave_enable_netpoll(new_slave)) {
1549                         pr_info("Error, %s: master_dev is using netpoll, but new slave device does not support netpoll\n",
1550                                 bond_dev->name);
1551                         res = -EBUSY;
1552                         goto err_detach;
1553                 }
1554         }
1555 #endif
1556
1557         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1558                                          new_slave);
1559         if (res) {
1560                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1561                 goto err_detach;
1562         }
1563
1564         res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1565         if (res) {
1566                 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1567                 goto err_unregister;
1568         }
1569
1570         res = bond_sysfs_slave_add(new_slave);
1571         if (res) {
1572                 pr_debug("Error %d calling bond_sysfs_slave_add\n", res);
1573                 goto err_upper_unlink;
1574         }
1575
1576         bond->slave_cnt++;
1577         bond_compute_features(bond);
1578         bond_set_carrier(bond);
1579
1580         if (bond_uses_primary(bond)) {
1581                 block_netpoll_tx();
1582                 write_lock_bh(&bond->curr_slave_lock);
1583                 bond_select_active_slave(bond);
1584                 write_unlock_bh(&bond->curr_slave_lock);
1585                 unblock_netpoll_tx();
1586         }
1587
1588         pr_info("%s: Enslaving %s as %s interface with %s link\n",
1589                 bond_dev->name, slave_dev->name,
1590                 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1591                 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1592
1593         /* enslave is successful */
1594         return 0;
1595
1596 /* Undo stages on error */
1597 err_upper_unlink:
1598         bond_upper_dev_unlink(bond_dev, slave_dev);
1599
1600 err_unregister:
1601         netdev_rx_handler_unregister(slave_dev);
1602
1603 err_detach:
1604         if (!bond_uses_primary(bond))
1605                 bond_hw_addr_flush(bond_dev, slave_dev);
1606
1607         vlan_vids_del_by_dev(slave_dev, bond_dev);
1608         if (bond->primary_slave == new_slave)
1609                 bond->primary_slave = NULL;
1610         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1611                 block_netpoll_tx();
1612                 write_lock_bh(&bond->curr_slave_lock);
1613                 bond_change_active_slave(bond, NULL);
1614                 bond_select_active_slave(bond);
1615                 write_unlock_bh(&bond->curr_slave_lock);
1616                 unblock_netpoll_tx();
1617         }
1618         slave_disable_netpoll(new_slave);
1619
1620 err_close:
1621         slave_dev->priv_flags &= ~IFF_BONDING;
1622         dev_close(slave_dev);
1623
1624 err_restore_mac:
1625         if (!bond->params.fail_over_mac ||
1626             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1627                 /* XXX TODO - fom follow mode needs to change master's
1628                  * MAC if this slave's MAC is in use by the bond, or at
1629                  * least print a warning.
1630                  */
1631                 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1632                 addr.sa_family = slave_dev->type;
1633                 dev_set_mac_address(slave_dev, &addr);
1634         }
1635
1636 err_restore_mtu:
1637         dev_set_mtu(slave_dev, new_slave->original_mtu);
1638
1639 err_free:
1640         bond_free_slave(new_slave);
1641
1642 err_undo_flags:
1643         /* Enslave of first slave has failed and we need to fix master's mac */
1644         if (!bond_has_slaves(bond) &&
1645             ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
1646                 eth_hw_addr_random(bond_dev);
1647
1648         return res;
1649 }
1650
1651 /*
1652  * Try to release the slave device <slave> from the bond device <master>
1653  * It is legal to access curr_active_slave without a lock because all the function
1654  * is write-locked. If "all" is true it means that the function is being called
1655  * while destroying a bond interface and all slaves are being released.
1656  *
1657  * The rules for slave state should be:
1658  *   for Active/Backup:
1659  *     Active stays on all backups go down
1660  *   for Bonded connections:
1661  *     The first up interface should be left on and all others downed.
1662  */
1663 static int __bond_release_one(struct net_device *bond_dev,
1664                               struct net_device *slave_dev,
1665                               bool all)
1666 {
1667         struct bonding *bond = netdev_priv(bond_dev);
1668         struct slave *slave, *oldcurrent;
1669         struct sockaddr addr;
1670         int old_flags = bond_dev->flags;
1671         netdev_features_t old_features = bond_dev->features;
1672
1673         /* slave is not a slave or master is not master of this slave */
1674         if (!(slave_dev->flags & IFF_SLAVE) ||
1675             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1676                 pr_err("%s: Error: cannot release %s\n",
1677                        bond_dev->name, slave_dev->name);
1678                 return -EINVAL;
1679         }
1680
1681         block_netpoll_tx();
1682
1683         slave = bond_get_slave_by_dev(bond, slave_dev);
1684         if (!slave) {
1685                 /* not a slave of this bond */
1686                 pr_info("%s: %s not enslaved\n",
1687                         bond_dev->name, slave_dev->name);
1688                 unblock_netpoll_tx();
1689                 return -EINVAL;
1690         }
1691
1692         bond_sysfs_slave_del(slave);
1693
1694         bond_upper_dev_unlink(bond_dev, slave_dev);
1695         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1696          * for this slave anymore.
1697          */
1698         netdev_rx_handler_unregister(slave_dev);
1699         write_lock_bh(&bond->lock);
1700
1701         /* Inform AD package of unbinding of slave. */
1702         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1703                 bond_3ad_unbind_slave(slave);
1704
1705         write_unlock_bh(&bond->lock);
1706
1707         pr_info("%s: Releasing %s interface %s\n",
1708                 bond_dev->name,
1709                 bond_is_active_slave(slave) ? "active" : "backup",
1710                 slave_dev->name);
1711
1712         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1713
1714         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1715
1716         if (!all && (!bond->params.fail_over_mac ||
1717                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1718                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1719                     bond_has_slaves(bond))
1720                         pr_warn("%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",
1721                                 bond_dev->name, slave_dev->name,
1722                                 slave->perm_hwaddr,
1723                                 bond_dev->name, slave_dev->name);
1724         }
1725
1726         if (bond->primary_slave == slave)
1727                 bond->primary_slave = NULL;
1728
1729         if (oldcurrent == slave) {
1730                 write_lock_bh(&bond->curr_slave_lock);
1731                 bond_change_active_slave(bond, NULL);
1732                 write_unlock_bh(&bond->curr_slave_lock);
1733         }
1734
1735         if (bond_is_lb(bond)) {
1736                 /* Must be called only after the slave has been
1737                  * detached from the list and the curr_active_slave
1738                  * has been cleared (if our_slave == old_current),
1739                  * but before a new active slave is selected.
1740                  */
1741                 bond_alb_deinit_slave(bond, slave);
1742         }
1743
1744         if (all) {
1745                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1746         } else if (oldcurrent == slave) {
1747                 /*
1748                  * Note that we hold RTNL over this sequence, so there
1749                  * is no concern that another slave add/remove event
1750                  * will interfere.
1751                  */
1752                 write_lock_bh(&bond->curr_slave_lock);
1753
1754                 bond_select_active_slave(bond);
1755
1756                 write_unlock_bh(&bond->curr_slave_lock);
1757         }
1758
1759         if (!bond_has_slaves(bond)) {
1760                 bond_set_carrier(bond);
1761                 eth_hw_addr_random(bond_dev);
1762         }
1763
1764         unblock_netpoll_tx();
1765         synchronize_rcu();
1766         bond->slave_cnt--;
1767
1768         if (!bond_has_slaves(bond)) {
1769                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1770                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1771         }
1772
1773         bond_compute_features(bond);
1774         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1775             (old_features & NETIF_F_VLAN_CHALLENGED))
1776                 pr_info("%s: last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1777                         bond_dev->name, slave_dev->name, bond_dev->name);
1778
1779         /* must do this from outside any spinlocks */
1780         vlan_vids_del_by_dev(slave_dev, bond_dev);
1781
1782         /* If the mode uses primary, then this cases was handled above by
1783          * bond_change_active_slave(..., NULL)
1784          */
1785         if (!bond_uses_primary(bond)) {
1786                 /* unset promiscuity level from slave
1787                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1788                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1789                  * value of that flag before that change, as that was the value
1790                  * when this slave was attached, so we cache at the start of the
1791                  * function and use it here. Same goes for ALLMULTI below
1792                  */
1793                 if (old_flags & IFF_PROMISC)
1794                         dev_set_promiscuity(slave_dev, -1);
1795
1796                 /* unset allmulti level from slave */
1797                 if (old_flags & IFF_ALLMULTI)
1798                         dev_set_allmulti(slave_dev, -1);
1799
1800                 bond_hw_addr_flush(bond_dev, slave_dev);
1801         }
1802
1803         slave_disable_netpoll(slave);
1804
1805         /* close slave before restoring its mac address */
1806         dev_close(slave_dev);
1807
1808         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1809             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1810                 /* restore original ("permanent") mac address */
1811                 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1812                 addr.sa_family = slave_dev->type;
1813                 dev_set_mac_address(slave_dev, &addr);
1814         }
1815
1816         dev_set_mtu(slave_dev, slave->original_mtu);
1817
1818         slave_dev->priv_flags &= ~IFF_BONDING;
1819
1820         bond_free_slave(slave);
1821
1822         return 0;  /* deletion OK */
1823 }
1824
1825 /* A wrapper used because of ndo_del_link */
1826 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1827 {
1828         return __bond_release_one(bond_dev, slave_dev, false);
1829 }
1830
1831 /*
1832 * First release a slave and then destroy the bond if no more slaves are left.
1833 * Must be under rtnl_lock when this function is called.
1834 */
1835 static int  bond_release_and_destroy(struct net_device *bond_dev,
1836                                      struct net_device *slave_dev)
1837 {
1838         struct bonding *bond = netdev_priv(bond_dev);
1839         int ret;
1840
1841         ret = bond_release(bond_dev, slave_dev);
1842         if (ret == 0 && !bond_has_slaves(bond)) {
1843                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1844                 pr_info("%s: Destroying bond %s\n",
1845                         bond_dev->name, bond_dev->name);
1846                 unregister_netdevice(bond_dev);
1847         }
1848         return ret;
1849 }
1850
1851 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1852 {
1853         struct bonding *bond = netdev_priv(bond_dev);
1854
1855         info->bond_mode = BOND_MODE(bond);
1856         info->miimon = bond->params.miimon;
1857
1858         info->num_slaves = bond->slave_cnt;
1859
1860         return 0;
1861 }
1862
1863 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1864 {
1865         struct bonding *bond = netdev_priv(bond_dev);
1866         struct list_head *iter;
1867         int i = 0, res = -ENODEV;
1868         struct slave *slave;
1869
1870         bond_for_each_slave(bond, slave, iter) {
1871                 if (i++ == (int)info->slave_id) {
1872                         res = 0;
1873                         strcpy(info->slave_name, slave->dev->name);
1874                         info->link = slave->link;
1875                         info->state = bond_slave_state(slave);
1876                         info->link_failure_count = slave->link_failure_count;
1877                         break;
1878                 }
1879         }
1880
1881         return res;
1882 }
1883
1884 /*-------------------------------- Monitoring -------------------------------*/
1885
1886 /* called with rcu_read_lock() */
1887 static int bond_miimon_inspect(struct bonding *bond)
1888 {
1889         int link_state, commit = 0;
1890         struct list_head *iter;
1891         struct slave *slave;
1892         bool ignore_updelay;
1893
1894         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
1895
1896         bond_for_each_slave_rcu(bond, slave, iter) {
1897                 slave->new_link = BOND_LINK_NOCHANGE;
1898
1899                 link_state = bond_check_dev_link(bond, slave->dev, 0);
1900
1901                 switch (slave->link) {
1902                 case BOND_LINK_UP:
1903                         if (link_state)
1904                                 continue;
1905
1906                         slave->link = BOND_LINK_FAIL;
1907                         slave->delay = bond->params.downdelay;
1908                         if (slave->delay) {
1909                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms\n",
1910                                         bond->dev->name,
1911                                         (BOND_MODE(bond) ==
1912                                          BOND_MODE_ACTIVEBACKUP) ?
1913                                         (bond_is_active_slave(slave) ?
1914                                          "active " : "backup ") : "",
1915                                         slave->dev->name,
1916                                         bond->params.downdelay * bond->params.miimon);
1917                         }
1918                         /*FALLTHRU*/
1919                 case BOND_LINK_FAIL:
1920                         if (link_state) {
1921                                 /*
1922                                  * recovered before downdelay expired
1923                                  */
1924                                 slave->link = BOND_LINK_UP;
1925                                 slave->last_link_up = jiffies;
1926                                 pr_info("%s: link status up again after %d ms for interface %s\n",
1927                                         bond->dev->name,
1928                                         (bond->params.downdelay - slave->delay) *
1929                                         bond->params.miimon,
1930                                         slave->dev->name);
1931                                 continue;
1932                         }
1933
1934                         if (slave->delay <= 0) {
1935                                 slave->new_link = BOND_LINK_DOWN;
1936                                 commit++;
1937                                 continue;
1938                         }
1939
1940                         slave->delay--;
1941                         break;
1942
1943                 case BOND_LINK_DOWN:
1944                         if (!link_state)
1945                                 continue;
1946
1947                         slave->link = BOND_LINK_BACK;
1948                         slave->delay = bond->params.updelay;
1949
1950                         if (slave->delay) {
1951                                 pr_info("%s: link status up for interface %s, enabling it in %d ms\n",
1952                                         bond->dev->name, slave->dev->name,
1953                                         ignore_updelay ? 0 :
1954                                         bond->params.updelay *
1955                                         bond->params.miimon);
1956                         }
1957                         /*FALLTHRU*/
1958                 case BOND_LINK_BACK:
1959                         if (!link_state) {
1960                                 slave->link = BOND_LINK_DOWN;
1961                                 pr_info("%s: link status down again after %d ms for interface %s\n",
1962                                         bond->dev->name,
1963                                         (bond->params.updelay - slave->delay) *
1964                                         bond->params.miimon,
1965                                         slave->dev->name);
1966
1967                                 continue;
1968                         }
1969
1970                         if (ignore_updelay)
1971                                 slave->delay = 0;
1972
1973                         if (slave->delay <= 0) {
1974                                 slave->new_link = BOND_LINK_UP;
1975                                 commit++;
1976                                 ignore_updelay = false;
1977                                 continue;
1978                         }
1979
1980                         slave->delay--;
1981                         break;
1982                 }
1983         }
1984
1985         return commit;
1986 }
1987
1988 static void bond_miimon_commit(struct bonding *bond)
1989 {
1990         struct list_head *iter;
1991         struct slave *slave;
1992
1993         bond_for_each_slave(bond, slave, iter) {
1994                 switch (slave->new_link) {
1995                 case BOND_LINK_NOCHANGE:
1996                         continue;
1997
1998                 case BOND_LINK_UP:
1999                         slave->link = BOND_LINK_UP;
2000                         slave->last_link_up = jiffies;
2001
2002                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2003                                 /* prevent it from being the active one */
2004                                 bond_set_backup_slave(slave);
2005                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2006                                 /* make it immediately active */
2007                                 bond_set_active_slave(slave);
2008                         } else if (slave != bond->primary_slave) {
2009                                 /* prevent it from being the active one */
2010                                 bond_set_backup_slave(slave);
2011                         }
2012
2013                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex\n",
2014                                 bond->dev->name, slave->dev->name,
2015                                 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2016                                 slave->duplex ? "full" : "half");
2017
2018                         /* notify ad that the link status has changed */
2019                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2020                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2021
2022                         if (bond_is_lb(bond))
2023                                 bond_alb_handle_link_change(bond, slave,
2024                                                             BOND_LINK_UP);
2025
2026                         if (!bond->curr_active_slave ||
2027                             (slave == bond->primary_slave))
2028                                 goto do_failover;
2029
2030                         continue;
2031
2032                 case BOND_LINK_DOWN:
2033                         if (slave->link_failure_count < UINT_MAX)
2034                                 slave->link_failure_count++;
2035
2036                         slave->link = BOND_LINK_DOWN;
2037
2038                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2039                             BOND_MODE(bond) == BOND_MODE_8023AD)
2040                                 bond_set_slave_inactive_flags(slave,
2041                                                               BOND_SLAVE_NOTIFY_NOW);
2042
2043                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2044                                 bond->dev->name, slave->dev->name);
2045
2046                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2047                                 bond_3ad_handle_link_change(slave,
2048                                                             BOND_LINK_DOWN);
2049
2050                         if (bond_is_lb(bond))
2051                                 bond_alb_handle_link_change(bond, slave,
2052                                                             BOND_LINK_DOWN);
2053
2054                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2055                                 goto do_failover;
2056
2057                         continue;
2058
2059                 default:
2060                         pr_err("%s: invalid new link %d on slave %s\n",
2061                                bond->dev->name, slave->new_link,
2062                                slave->dev->name);
2063                         slave->new_link = BOND_LINK_NOCHANGE;
2064
2065                         continue;
2066                 }
2067
2068 do_failover:
2069                 ASSERT_RTNL();
2070                 block_netpoll_tx();
2071                 write_lock_bh(&bond->curr_slave_lock);
2072                 bond_select_active_slave(bond);
2073                 write_unlock_bh(&bond->curr_slave_lock);
2074                 unblock_netpoll_tx();
2075         }
2076
2077         bond_set_carrier(bond);
2078 }
2079
2080 /*
2081  * bond_mii_monitor
2082  *
2083  * Really a wrapper that splits the mii monitor into two phases: an
2084  * inspection, then (if inspection indicates something needs to be done)
2085  * an acquisition of appropriate locks followed by a commit phase to
2086  * implement whatever link state changes are indicated.
2087  */
2088 static void bond_mii_monitor(struct work_struct *work)
2089 {
2090         struct bonding *bond = container_of(work, struct bonding,
2091                                             mii_work.work);
2092         bool should_notify_peers = false;
2093         unsigned long delay;
2094
2095         delay = msecs_to_jiffies(bond->params.miimon);
2096
2097         if (!bond_has_slaves(bond))
2098                 goto re_arm;
2099
2100         rcu_read_lock();
2101
2102         should_notify_peers = bond_should_notify_peers(bond);
2103
2104         if (bond_miimon_inspect(bond)) {
2105                 rcu_read_unlock();
2106
2107                 /* Race avoidance with bond_close cancel of workqueue */
2108                 if (!rtnl_trylock()) {
2109                         delay = 1;
2110                         should_notify_peers = false;
2111                         goto re_arm;
2112                 }
2113
2114                 bond_miimon_commit(bond);
2115
2116                 rtnl_unlock();  /* might sleep, hold no other locks */
2117         } else
2118                 rcu_read_unlock();
2119
2120 re_arm:
2121         if (bond->params.miimon)
2122                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2123
2124         if (should_notify_peers) {
2125                 if (!rtnl_trylock())
2126                         return;
2127                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2128                 rtnl_unlock();
2129         }
2130 }
2131
2132 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2133 {
2134         struct net_device *upper;
2135         struct list_head *iter;
2136         bool ret = false;
2137
2138         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2139                 return true;
2140
2141         rcu_read_lock();
2142         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2143                 if (ip == bond_confirm_addr(upper, 0, ip)) {
2144                         ret = true;
2145                         break;
2146                 }
2147         }
2148         rcu_read_unlock();
2149
2150         return ret;
2151 }
2152
2153 /*
2154  * We go to the (large) trouble of VLAN tagging ARP frames because
2155  * switches in VLAN mode (especially if ports are configured as
2156  * "native" to a VLAN) might not pass non-tagged frames.
2157  */
2158 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2159                           __be32 dest_ip, __be32 src_ip,
2160                           struct bond_vlan_tag *tags)
2161 {
2162         struct sk_buff *skb;
2163         int i;
2164
2165         pr_debug("arp %d on slave %s: dst %pI4 src %pI4\n",
2166                  arp_op, slave_dev->name, &dest_ip, &src_ip);
2167
2168         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2169                          NULL, slave_dev->dev_addr, NULL);
2170
2171         if (!skb) {
2172                 net_err_ratelimited("ARP packet allocation failed\n");
2173                 return;
2174         }
2175
2176         /* Go through all the tags backwards and add them to the packet */
2177         for (i = BOND_MAX_VLAN_ENCAP - 1; i > 0; i--) {
2178                 if (!tags[i].vlan_id)
2179                         continue;
2180
2181                 pr_debug("inner tag: proto %X vid %X\n",
2182                          ntohs(tags[i].vlan_proto), tags[i].vlan_id);
2183                 skb = __vlan_put_tag(skb, tags[i].vlan_proto,
2184                                      tags[i].vlan_id);
2185                 if (!skb) {
2186                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2187                         return;
2188                 }
2189         }
2190         /* Set the outer tag */
2191         if (tags[0].vlan_id) {
2192                 pr_debug("outer tag: proto %X vid %X\n",
2193                          ntohs(tags[0].vlan_proto), tags[0].vlan_id);
2194                 skb = vlan_put_tag(skb, tags[0].vlan_proto, tags[0].vlan_id);
2195                 if (!skb) {
2196                         net_err_ratelimited("failed to insert outer VLAN tag\n");
2197                         return;
2198                 }
2199         }
2200         arp_xmit(skb);
2201 }
2202
2203 /* Validate the device path between the @start_dev and the @end_dev.
2204  * The path is valid if the @end_dev is reachable through device
2205  * stacking.
2206  * When the path is validated, collect any vlan information in the
2207  * path.
2208  */
2209 bool bond_verify_device_path(struct net_device *start_dev,
2210                              struct net_device *end_dev,
2211                              struct bond_vlan_tag *tags)
2212 {
2213         struct net_device *upper;
2214         struct list_head  *iter;
2215         int  idx;
2216
2217         if (start_dev == end_dev)
2218                 return true;
2219
2220         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2221                 if (bond_verify_device_path(upper, end_dev, tags)) {
2222                         if (is_vlan_dev(upper)) {
2223                                 idx = vlan_get_encap_level(upper);
2224                                 if (idx >= BOND_MAX_VLAN_ENCAP)
2225                                         return false;
2226
2227                                 tags[idx].vlan_proto =
2228                                                     vlan_dev_vlan_proto(upper);
2229                                 tags[idx].vlan_id = vlan_dev_vlan_id(upper);
2230                         }
2231                         return true;
2232                 }
2233         }
2234
2235         return false;
2236 }
2237
2238 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2239 {
2240         struct rtable *rt;
2241         struct bond_vlan_tag tags[BOND_MAX_VLAN_ENCAP];
2242         __be32 *targets = bond->params.arp_targets, addr;
2243         int i;
2244         bool ret;
2245
2246         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2247                 pr_debug("basa: target %pI4\n", &targets[i]);
2248                 memset(tags, 0, sizeof(tags));
2249
2250                 /* Find out through which dev should the packet go */
2251                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2252                                      RTO_ONLINK, 0);
2253                 if (IS_ERR(rt)) {
2254                         /* there's no route to target - try to send arp
2255                          * probe to generate any traffic (arp_validate=0)
2256                          */
2257                         if (bond->params.arp_validate)
2258                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2259                                                      bond->dev->name,
2260                                                      &targets[i]);
2261                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2262                                       0, tags);
2263                         continue;
2264                 }
2265
2266                 /* bond device itself */
2267                 if (rt->dst.dev == bond->dev)
2268                         goto found;
2269
2270                 rcu_read_lock();
2271                 ret = bond_verify_device_path(bond->dev, rt->dst.dev, tags);
2272                 rcu_read_unlock();
2273
2274                 if (ret)
2275                         goto found;
2276
2277                 /* Not our device - skip */
2278                 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2279                          bond->dev->name, &targets[i],
2280                          rt->dst.dev ? rt->dst.dev->name : "NULL");
2281
2282                 ip_rt_put(rt);
2283                 continue;
2284
2285 found:
2286                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2287                 ip_rt_put(rt);
2288                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2289                               addr, tags);
2290         }
2291 }
2292
2293 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2294 {
2295         int i;
2296
2297         if (!sip || !bond_has_this_ip(bond, tip)) {
2298                 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2299                 return;
2300         }
2301
2302         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2303         if (i == -1) {
2304                 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2305                 return;
2306         }
2307         slave->last_rx = jiffies;
2308         slave->target_last_arp_rx[i] = jiffies;
2309 }
2310
2311 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2312                  struct slave *slave)
2313 {
2314         struct arphdr *arp = (struct arphdr *)skb->data;
2315         struct slave *curr_active_slave;
2316         unsigned char *arp_ptr;
2317         __be32 sip, tip;
2318         int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2319
2320         if (!slave_do_arp_validate(bond, slave)) {
2321                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2322                     !slave_do_arp_validate_only(bond))
2323                         slave->last_rx = jiffies;
2324                 return RX_HANDLER_ANOTHER;
2325         } else if (!is_arp) {
2326                 return RX_HANDLER_ANOTHER;
2327         }
2328
2329         alen = arp_hdr_len(bond->dev);
2330
2331         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2332                  bond->dev->name, skb->dev->name);
2333
2334         if (alen > skb_headlen(skb)) {
2335                 arp = kmalloc(alen, GFP_ATOMIC);
2336                 if (!arp)
2337                         goto out_unlock;
2338                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2339                         goto out_unlock;
2340         }
2341
2342         if (arp->ar_hln != bond->dev->addr_len ||
2343             skb->pkt_type == PACKET_OTHERHOST ||
2344             skb->pkt_type == PACKET_LOOPBACK ||
2345             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2346             arp->ar_pro != htons(ETH_P_IP) ||
2347             arp->ar_pln != 4)
2348                 goto out_unlock;
2349
2350         arp_ptr = (unsigned char *)(arp + 1);
2351         arp_ptr += bond->dev->addr_len;
2352         memcpy(&sip, arp_ptr, 4);
2353         arp_ptr += 4 + bond->dev->addr_len;
2354         memcpy(&tip, arp_ptr, 4);
2355
2356         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2357                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2358                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2359                  &sip, &tip);
2360
2361         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2362
2363         /*
2364          * Backup slaves won't see the ARP reply, but do come through
2365          * here for each ARP probe (so we swap the sip/tip to validate
2366          * the probe).  In a "redundant switch, common router" type of
2367          * configuration, the ARP probe will (hopefully) travel from
2368          * the active, through one switch, the router, then the other
2369          * switch before reaching the backup.
2370          *
2371          * We 'trust' the arp requests if there is an active slave and
2372          * it received valid arp reply(s) after it became active. This
2373          * is done to avoid endless looping when we can't reach the
2374          * arp_ip_target and fool ourselves with our own arp requests.
2375          */
2376
2377         if (bond_is_active_slave(slave))
2378                 bond_validate_arp(bond, slave, sip, tip);
2379         else if (curr_active_slave &&
2380                  time_after(slave_last_rx(bond, curr_active_slave),
2381                             curr_active_slave->last_link_up))
2382                 bond_validate_arp(bond, slave, tip, sip);
2383
2384 out_unlock:
2385         if (arp != (struct arphdr *)skb->data)
2386                 kfree(arp);
2387         return RX_HANDLER_ANOTHER;
2388 }
2389
2390 /* function to verify if we're in the arp_interval timeslice, returns true if
2391  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2392  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2393  */
2394 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2395                                   int mod)
2396 {
2397         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2398
2399         return time_in_range(jiffies,
2400                              last_act - delta_in_ticks,
2401                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2402 }
2403
2404 /*
2405  * this function is called regularly to monitor each slave's link
2406  * ensuring that traffic is being sent and received when arp monitoring
2407  * is used in load-balancing mode. if the adapter has been dormant, then an
2408  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2409  * arp monitoring in active backup mode.
2410  */
2411 static void bond_loadbalance_arp_mon(struct work_struct *work)
2412 {
2413         struct bonding *bond = container_of(work, struct bonding,
2414                                             arp_work.work);
2415         struct slave *slave, *oldcurrent;
2416         struct list_head *iter;
2417         int do_failover = 0, slave_state_changed = 0;
2418
2419         if (!bond_has_slaves(bond))
2420                 goto re_arm;
2421
2422         rcu_read_lock();
2423
2424         oldcurrent = rcu_dereference(bond->curr_active_slave);
2425         /* see if any of the previous devices are up now (i.e. they have
2426          * xmt and rcv traffic). the curr_active_slave does not come into
2427          * the picture unless it is null. also, slave->last_link_up is not
2428          * needed here because we send an arp on each slave and give a slave
2429          * as long as it needs to get the tx/rx within the delta.
2430          * TODO: what about up/down delay in arp mode? it wasn't here before
2431          *       so it can wait
2432          */
2433         bond_for_each_slave_rcu(bond, slave, iter) {
2434                 unsigned long trans_start = dev_trans_start(slave->dev);
2435
2436                 if (slave->link != BOND_LINK_UP) {
2437                         if (bond_time_in_interval(bond, trans_start, 1) &&
2438                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2439
2440                                 slave->link  = BOND_LINK_UP;
2441                                 slave_state_changed = 1;
2442
2443                                 /* primary_slave has no meaning in round-robin
2444                                  * mode. the window of a slave being up and
2445                                  * curr_active_slave being null after enslaving
2446                                  * is closed.
2447                                  */
2448                                 if (!oldcurrent) {
2449                                         pr_info("%s: link status definitely up for interface %s\n",
2450                                                 bond->dev->name,
2451                                                 slave->dev->name);
2452                                         do_failover = 1;
2453                                 } else {
2454                                         pr_info("%s: interface %s is now up\n",
2455                                                 bond->dev->name,
2456                                                 slave->dev->name);
2457                                 }
2458                         }
2459                 } else {
2460                         /* slave->link == BOND_LINK_UP */
2461
2462                         /* not all switches will respond to an arp request
2463                          * when the source ip is 0, so don't take the link down
2464                          * if we don't know our ip yet
2465                          */
2466                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2467                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2468
2469                                 slave->link  = BOND_LINK_DOWN;
2470                                 slave_state_changed = 1;
2471
2472                                 if (slave->link_failure_count < UINT_MAX)
2473                                         slave->link_failure_count++;
2474
2475                                 pr_info("%s: interface %s is now down\n",
2476                                         bond->dev->name, slave->dev->name);
2477
2478                                 if (slave == oldcurrent)
2479                                         do_failover = 1;
2480                         }
2481                 }
2482
2483                 /* note: if switch is in round-robin mode, all links
2484                  * must tx arp to ensure all links rx an arp - otherwise
2485                  * links may oscillate or not come up at all; if switch is
2486                  * in something like xor mode, there is nothing we can
2487                  * do - all replies will be rx'ed on same link causing slaves
2488                  * to be unstable during low/no traffic periods
2489                  */
2490                 if (bond_slave_is_up(slave))
2491                         bond_arp_send_all(bond, slave);
2492         }
2493
2494         rcu_read_unlock();
2495
2496         if (do_failover || slave_state_changed) {
2497                 if (!rtnl_trylock())
2498                         goto re_arm;
2499
2500                 if (slave_state_changed) {
2501                         bond_slave_state_change(bond);
2502                 } else if (do_failover) {
2503                         /* the bond_select_active_slave must hold RTNL
2504                          * and curr_slave_lock for write.
2505                          */
2506                         block_netpoll_tx();
2507                         write_lock_bh(&bond->curr_slave_lock);
2508
2509                         bond_select_active_slave(bond);
2510
2511                         write_unlock_bh(&bond->curr_slave_lock);
2512                         unblock_netpoll_tx();
2513                 }
2514                 rtnl_unlock();
2515         }
2516
2517 re_arm:
2518         if (bond->params.arp_interval)
2519                 queue_delayed_work(bond->wq, &bond->arp_work,
2520                                    msecs_to_jiffies(bond->params.arp_interval));
2521 }
2522
2523 /*
2524  * Called to inspect slaves for active-backup mode ARP monitor link state
2525  * changes.  Sets new_link in slaves to specify what action should take
2526  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2527  * to link states must be committed.
2528  *
2529  * Called with rcu_read_lock hold.
2530  */
2531 static int bond_ab_arp_inspect(struct bonding *bond)
2532 {
2533         unsigned long trans_start, last_rx;
2534         struct list_head *iter;
2535         struct slave *slave;
2536         int commit = 0;
2537
2538         bond_for_each_slave_rcu(bond, slave, iter) {
2539                 slave->new_link = BOND_LINK_NOCHANGE;
2540                 last_rx = slave_last_rx(bond, slave);
2541
2542                 if (slave->link != BOND_LINK_UP) {
2543                         if (bond_time_in_interval(bond, last_rx, 1)) {
2544                                 slave->new_link = BOND_LINK_UP;
2545                                 commit++;
2546                         }
2547                         continue;
2548                 }
2549
2550                 /*
2551                  * Give slaves 2*delta after being enslaved or made
2552                  * active.  This avoids bouncing, as the last receive
2553                  * times need a full ARP monitor cycle to be updated.
2554                  */
2555                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2556                         continue;
2557
2558                 /*
2559                  * Backup slave is down if:
2560                  * - No current_arp_slave AND
2561                  * - more than 3*delta since last receive AND
2562                  * - the bond has an IP address
2563                  *
2564                  * Note: a non-null current_arp_slave indicates
2565                  * the curr_active_slave went down and we are
2566                  * searching for a new one; under this condition
2567                  * we only take the curr_active_slave down - this
2568                  * gives each slave a chance to tx/rx traffic
2569                  * before being taken out
2570                  */
2571                 if (!bond_is_active_slave(slave) &&
2572                     !rcu_access_pointer(bond->current_arp_slave) &&
2573                     !bond_time_in_interval(bond, last_rx, 3)) {
2574                         slave->new_link = BOND_LINK_DOWN;
2575                         commit++;
2576                 }
2577
2578                 /*
2579                  * Active slave is down if:
2580                  * - more than 2*delta since transmitting OR
2581                  * - (more than 2*delta since receive AND
2582                  *    the bond has an IP address)
2583                  */
2584                 trans_start = dev_trans_start(slave->dev);
2585                 if (bond_is_active_slave(slave) &&
2586                     (!bond_time_in_interval(bond, trans_start, 2) ||
2587                      !bond_time_in_interval(bond, last_rx, 2))) {
2588                         slave->new_link = BOND_LINK_DOWN;
2589                         commit++;
2590                 }
2591         }
2592
2593         return commit;
2594 }
2595
2596 /*
2597  * Called to commit link state changes noted by inspection step of
2598  * active-backup mode ARP monitor.
2599  *
2600  * Called with RTNL hold.
2601  */
2602 static void bond_ab_arp_commit(struct bonding *bond)
2603 {
2604         unsigned long trans_start;
2605         struct list_head *iter;
2606         struct slave *slave;
2607
2608         bond_for_each_slave(bond, slave, iter) {
2609                 switch (slave->new_link) {
2610                 case BOND_LINK_NOCHANGE:
2611                         continue;
2612
2613                 case BOND_LINK_UP:
2614                         trans_start = dev_trans_start(slave->dev);
2615                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
2616                             (!rtnl_dereference(bond->curr_active_slave) &&
2617                              bond_time_in_interval(bond, trans_start, 1))) {
2618                                 struct slave *current_arp_slave;
2619
2620                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2621                                 slave->link = BOND_LINK_UP;
2622                                 if (current_arp_slave) {
2623                                         bond_set_slave_inactive_flags(
2624                                                 current_arp_slave,
2625                                                 BOND_SLAVE_NOTIFY_NOW);
2626                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2627                                 }
2628
2629                                 pr_info("%s: link status definitely up for interface %s\n",
2630                                         bond->dev->name, slave->dev->name);
2631
2632                                 if (!rtnl_dereference(bond->curr_active_slave) ||
2633                                     (slave == bond->primary_slave))
2634                                         goto do_failover;
2635
2636                         }
2637
2638                         continue;
2639
2640                 case BOND_LINK_DOWN:
2641                         if (slave->link_failure_count < UINT_MAX)
2642                                 slave->link_failure_count++;
2643
2644                         slave->link = BOND_LINK_DOWN;
2645                         bond_set_slave_inactive_flags(slave,
2646                                                       BOND_SLAVE_NOTIFY_NOW);
2647
2648                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2649                                 bond->dev->name, slave->dev->name);
2650
2651                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
2652                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2653                                 goto do_failover;
2654                         }
2655
2656                         continue;
2657
2658                 default:
2659                         pr_err("%s: impossible: new_link %d on slave %s\n",
2660                                bond->dev->name, slave->new_link,
2661                                slave->dev->name);
2662                         continue;
2663                 }
2664
2665 do_failover:
2666                 ASSERT_RTNL();
2667                 block_netpoll_tx();
2668                 write_lock_bh(&bond->curr_slave_lock);
2669                 bond_select_active_slave(bond);
2670                 write_unlock_bh(&bond->curr_slave_lock);
2671                 unblock_netpoll_tx();
2672         }
2673
2674         bond_set_carrier(bond);
2675 }
2676
2677 /*
2678  * Send ARP probes for active-backup mode ARP monitor.
2679  *
2680  * Called with rcu_read_lock hold.
2681  */
2682 static bool bond_ab_arp_probe(struct bonding *bond)
2683 {
2684         struct slave *slave, *before = NULL, *new_slave = NULL,
2685                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2686                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2687         struct list_head *iter;
2688         bool found = false;
2689         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2690
2691         if (curr_arp_slave && curr_active_slave)
2692                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2693                         curr_arp_slave->dev->name,
2694                         curr_active_slave->dev->name);
2695
2696         if (curr_active_slave) {
2697                 bond_arp_send_all(bond, curr_active_slave);
2698                 return should_notify_rtnl;
2699         }
2700
2701         /* if we don't have a curr_active_slave, search for the next available
2702          * backup slave from the current_arp_slave and make it the candidate
2703          * for becoming the curr_active_slave
2704          */
2705
2706         if (!curr_arp_slave) {
2707                 curr_arp_slave = bond_first_slave_rcu(bond);
2708                 if (!curr_arp_slave)
2709                         return should_notify_rtnl;
2710         }
2711
2712         bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2713
2714         bond_for_each_slave_rcu(bond, slave, iter) {
2715                 if (!found && !before && bond_slave_is_up(slave))
2716                         before = slave;
2717
2718                 if (found && !new_slave && bond_slave_is_up(slave))
2719                         new_slave = slave;
2720                 /* if the link state is up at this point, we
2721                  * mark it down - this can happen if we have
2722                  * simultaneous link failures and
2723                  * reselect_active_interface doesn't make this
2724                  * one the current slave so it is still marked
2725                  * up when it is actually down
2726                  */
2727                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2728                         slave->link = BOND_LINK_DOWN;
2729                         if (slave->link_failure_count < UINT_MAX)
2730                                 slave->link_failure_count++;
2731
2732                         bond_set_slave_inactive_flags(slave,
2733                                                       BOND_SLAVE_NOTIFY_LATER);
2734
2735                         pr_info("%s: backup interface %s is now down\n",
2736                                 bond->dev->name, slave->dev->name);
2737                 }
2738                 if (slave == curr_arp_slave)
2739                         found = true;
2740         }
2741
2742         if (!new_slave && before)
2743                 new_slave = before;
2744
2745         if (!new_slave)
2746                 goto check_state;
2747
2748         new_slave->link = BOND_LINK_BACK;
2749         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2750         bond_arp_send_all(bond, new_slave);
2751         new_slave->last_link_up = jiffies;
2752         rcu_assign_pointer(bond->current_arp_slave, new_slave);
2753
2754 check_state:
2755         bond_for_each_slave_rcu(bond, slave, iter) {
2756                 if (slave->should_notify) {
2757                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2758                         break;
2759                 }
2760         }
2761         return should_notify_rtnl;
2762 }
2763
2764 static void bond_activebackup_arp_mon(struct work_struct *work)
2765 {
2766         struct bonding *bond = container_of(work, struct bonding,
2767                                             arp_work.work);
2768         bool should_notify_peers = false;
2769         bool should_notify_rtnl = false;
2770         int delta_in_ticks;
2771
2772         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2773
2774         if (!bond_has_slaves(bond))
2775                 goto re_arm;
2776
2777         rcu_read_lock();
2778
2779         should_notify_peers = bond_should_notify_peers(bond);
2780
2781         if (bond_ab_arp_inspect(bond)) {
2782                 rcu_read_unlock();
2783
2784                 /* Race avoidance with bond_close flush of workqueue */
2785                 if (!rtnl_trylock()) {
2786                         delta_in_ticks = 1;
2787                         should_notify_peers = false;
2788                         goto re_arm;
2789                 }
2790
2791                 bond_ab_arp_commit(bond);
2792
2793                 rtnl_unlock();
2794                 rcu_read_lock();
2795         }
2796
2797         should_notify_rtnl = bond_ab_arp_probe(bond);
2798         rcu_read_unlock();
2799
2800 re_arm:
2801         if (bond->params.arp_interval)
2802                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2803
2804         if (should_notify_peers || should_notify_rtnl) {
2805                 if (!rtnl_trylock())
2806                         return;
2807
2808                 if (should_notify_peers)
2809                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2810                                                  bond->dev);
2811                 if (should_notify_rtnl)
2812                         bond_slave_state_notify(bond);
2813
2814                 rtnl_unlock();
2815         }
2816 }
2817
2818 /*-------------------------- netdev event handling --------------------------*/
2819
2820 /*
2821  * Change device name
2822  */
2823 static int bond_event_changename(struct bonding *bond)
2824 {
2825         bond_remove_proc_entry(bond);
2826         bond_create_proc_entry(bond);
2827
2828         bond_debug_reregister(bond);
2829
2830         return NOTIFY_DONE;
2831 }
2832
2833 static int bond_master_netdev_event(unsigned long event,
2834                                     struct net_device *bond_dev)
2835 {
2836         struct bonding *event_bond = netdev_priv(bond_dev);
2837
2838         switch (event) {
2839         case NETDEV_CHANGENAME:
2840                 return bond_event_changename(event_bond);
2841         case NETDEV_UNREGISTER:
2842                 bond_remove_proc_entry(event_bond);
2843                 break;
2844         case NETDEV_REGISTER:
2845                 bond_create_proc_entry(event_bond);
2846                 break;
2847         case NETDEV_NOTIFY_PEERS:
2848                 if (event_bond->send_peer_notif)
2849                         event_bond->send_peer_notif--;
2850                 break;
2851         default:
2852                 break;
2853         }
2854
2855         return NOTIFY_DONE;
2856 }
2857
2858 static int bond_slave_netdev_event(unsigned long event,
2859                                    struct net_device *slave_dev)
2860 {
2861         struct slave *slave = bond_slave_get_rtnl(slave_dev);
2862         struct bonding *bond;
2863         struct net_device *bond_dev;
2864         u32 old_speed;
2865         u8 old_duplex;
2866
2867         /* A netdev event can be generated while enslaving a device
2868          * before netdev_rx_handler_register is called in which case
2869          * slave will be NULL
2870          */
2871         if (!slave)
2872                 return NOTIFY_DONE;
2873         bond_dev = slave->bond->dev;
2874         bond = slave->bond;
2875
2876         switch (event) {
2877         case NETDEV_UNREGISTER:
2878                 if (bond_dev->type != ARPHRD_ETHER)
2879                         bond_release_and_destroy(bond_dev, slave_dev);
2880                 else
2881                         bond_release(bond_dev, slave_dev);
2882                 break;
2883         case NETDEV_UP:
2884         case NETDEV_CHANGE:
2885                 old_speed = slave->speed;
2886                 old_duplex = slave->duplex;
2887
2888                 bond_update_speed_duplex(slave);
2889
2890                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2891                         if (old_speed != slave->speed)
2892                                 bond_3ad_adapter_speed_changed(slave);
2893                         if (old_duplex != slave->duplex)
2894                                 bond_3ad_adapter_duplex_changed(slave);
2895                 }
2896                 break;
2897         case NETDEV_DOWN:
2898                 /*
2899                  * ... Or is it this?
2900                  */
2901                 break;
2902         case NETDEV_CHANGEMTU:
2903                 /*
2904                  * TODO: Should slaves be allowed to
2905                  * independently alter their MTU?  For
2906                  * an active-backup bond, slaves need
2907                  * not be the same type of device, so
2908                  * MTUs may vary.  For other modes,
2909                  * slaves arguably should have the
2910                  * same MTUs. To do this, we'd need to
2911                  * take over the slave's change_mtu
2912                  * function for the duration of their
2913                  * servitude.
2914                  */
2915                 break;
2916         case NETDEV_CHANGENAME:
2917                 /* we don't care if we don't have primary set */
2918                 if (!bond_uses_primary(bond) ||
2919                     !bond->params.primary[0])
2920                         break;
2921
2922                 if (slave == bond->primary_slave) {
2923                         /* slave's name changed - he's no longer primary */
2924                         bond->primary_slave = NULL;
2925                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2926                         /* we have a new primary slave */
2927                         bond->primary_slave = slave;
2928                 } else { /* we didn't change primary - exit */
2929                         break;
2930                 }
2931
2932                 pr_info("%s: Primary slave changed to %s, reselecting active slave\n",
2933                         bond->dev->name,
2934                         bond->primary_slave ? slave_dev->name : "none");
2935
2936                 block_netpoll_tx();
2937                 write_lock_bh(&bond->curr_slave_lock);
2938                 bond_select_active_slave(bond);
2939                 write_unlock_bh(&bond->curr_slave_lock);
2940                 unblock_netpoll_tx();
2941                 break;
2942         case NETDEV_FEAT_CHANGE:
2943                 bond_compute_features(bond);
2944                 break;
2945         case NETDEV_RESEND_IGMP:
2946                 /* Propagate to master device */
2947                 call_netdevice_notifiers(event, slave->bond->dev);
2948                 break;
2949         default:
2950                 break;
2951         }
2952
2953         return NOTIFY_DONE;
2954 }
2955
2956 /*
2957  * bond_netdev_event: handle netdev notifier chain events.
2958  *
2959  * This function receives events for the netdev chain.  The caller (an
2960  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2961  * locks for us to safely manipulate the slave devices (RTNL lock,
2962  * dev_probe_lock).
2963  */
2964 static int bond_netdev_event(struct notifier_block *this,
2965                              unsigned long event, void *ptr)
2966 {
2967         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2968
2969         pr_debug("event_dev: %s, event: %lx\n",
2970                  event_dev ? event_dev->name : "None", event);
2971
2972         if (!(event_dev->priv_flags & IFF_BONDING))
2973                 return NOTIFY_DONE;
2974
2975         if (event_dev->flags & IFF_MASTER) {
2976                 pr_debug("IFF_MASTER\n");
2977                 return bond_master_netdev_event(event, event_dev);
2978         }
2979
2980         if (event_dev->flags & IFF_SLAVE) {
2981                 pr_debug("IFF_SLAVE\n");
2982                 return bond_slave_netdev_event(event, event_dev);
2983         }
2984
2985         return NOTIFY_DONE;
2986 }
2987
2988 static struct notifier_block bond_netdev_notifier = {
2989         .notifier_call = bond_netdev_event,
2990 };
2991
2992 /*---------------------------- Hashing Policies -----------------------------*/
2993
2994 /* L2 hash helper */
2995 static inline u32 bond_eth_hash(struct sk_buff *skb)
2996 {
2997         struct ethhdr *data = (struct ethhdr *)skb->data;
2998
2999         if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3000                 return data->h_dest[5] ^ data->h_source[5];
3001
3002         return 0;
3003 }
3004
3005 /* Extract the appropriate headers based on bond's xmit policy */
3006 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3007                               struct flow_keys *fk)
3008 {
3009         const struct ipv6hdr *iph6;
3010         const struct iphdr *iph;
3011         int noff, proto = -1;
3012
3013         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3014                 return skb_flow_dissect(skb, fk);
3015
3016         fk->ports = 0;
3017         noff = skb_network_offset(skb);
3018         if (skb->protocol == htons(ETH_P_IP)) {
3019                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3020                         return false;
3021                 iph = ip_hdr(skb);
3022                 fk->src = iph->saddr;
3023                 fk->dst = iph->daddr;
3024                 noff += iph->ihl << 2;
3025                 if (!ip_is_fragment(iph))
3026                         proto = iph->protocol;
3027         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3028                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3029                         return false;
3030                 iph6 = ipv6_hdr(skb);
3031                 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
3032                 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
3033                 noff += sizeof(*iph6);
3034                 proto = iph6->nexthdr;
3035         } else {
3036                 return false;
3037         }
3038         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3039                 fk->ports = skb_flow_get_ports(skb, noff, proto);
3040
3041         return true;
3042 }
3043
3044 /**
3045  * bond_xmit_hash - generate a hash value based on the xmit policy
3046  * @bond: bonding device
3047  * @skb: buffer to use for headers
3048  *
3049  * This function will extract the necessary headers from the skb buffer and use
3050  * them to generate a hash based on the xmit_policy set in the bonding device
3051  */
3052 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3053 {
3054         struct flow_keys flow;
3055         u32 hash;
3056
3057         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3058             !bond_flow_dissect(bond, skb, &flow))
3059                 return bond_eth_hash(skb);
3060
3061         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3062             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3063                 hash = bond_eth_hash(skb);
3064         else
3065                 hash = (__force u32)flow.ports;
3066         hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
3067         hash ^= (hash >> 16);
3068         hash ^= (hash >> 8);
3069
3070         return hash;
3071 }
3072
3073 /*-------------------------- Device entry points ----------------------------*/
3074
3075 static void bond_work_init_all(struct bonding *bond)
3076 {
3077         INIT_DELAYED_WORK(&bond->mcast_work,
3078                           bond_resend_igmp_join_requests_delayed);
3079         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3080         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3081         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3082                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3083         else
3084                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3085         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3086 }
3087
3088 static void bond_work_cancel_all(struct bonding *bond)
3089 {
3090         cancel_delayed_work_sync(&bond->mii_work);
3091         cancel_delayed_work_sync(&bond->arp_work);
3092         cancel_delayed_work_sync(&bond->alb_work);
3093         cancel_delayed_work_sync(&bond->ad_work);
3094         cancel_delayed_work_sync(&bond->mcast_work);
3095 }
3096
3097 static int bond_open(struct net_device *bond_dev)
3098 {
3099         struct bonding *bond = netdev_priv(bond_dev);
3100         struct list_head *iter;
3101         struct slave *slave;
3102
3103         /* reset slave->backup and slave->inactive */
3104         read_lock(&bond->lock);
3105         if (bond_has_slaves(bond)) {
3106                 read_lock(&bond->curr_slave_lock);
3107                 bond_for_each_slave(bond, slave, iter) {
3108                         if (bond_uses_primary(bond) &&
3109                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3110                                 bond_set_slave_inactive_flags(slave,
3111                                                               BOND_SLAVE_NOTIFY_NOW);
3112                         } else {
3113                                 bond_set_slave_active_flags(slave,
3114                                                             BOND_SLAVE_NOTIFY_NOW);
3115                         }
3116                 }
3117                 read_unlock(&bond->curr_slave_lock);
3118         }
3119         read_unlock(&bond->lock);
3120
3121         bond_work_init_all(bond);
3122
3123         if (bond_is_lb(bond)) {
3124                 /* bond_alb_initialize must be called before the timer
3125                  * is started.
3126                  */
3127                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3128                         return -ENOMEM;
3129                 if (bond->params.tlb_dynamic_lb)
3130                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3131         }
3132
3133         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3134                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3135
3136         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3137                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3138                 bond->recv_probe = bond_arp_rcv;
3139         }
3140
3141         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3142                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3143                 /* register to receive LACPDUs */
3144                 bond->recv_probe = bond_3ad_lacpdu_recv;
3145                 bond_3ad_initiate_agg_selection(bond, 1);
3146         }
3147
3148         return 0;
3149 }
3150
3151 static int bond_close(struct net_device *bond_dev)
3152 {
3153         struct bonding *bond = netdev_priv(bond_dev);
3154
3155         bond_work_cancel_all(bond);
3156         bond->send_peer_notif = 0;
3157         if (bond_is_lb(bond))
3158                 bond_alb_deinitialize(bond);
3159         bond->recv_probe = NULL;
3160
3161         return 0;
3162 }
3163
3164 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3165                                                 struct rtnl_link_stats64 *stats)
3166 {
3167         struct bonding *bond = netdev_priv(bond_dev);
3168         struct rtnl_link_stats64 temp;
3169         struct list_head *iter;
3170         struct slave *slave;
3171
3172         memset(stats, 0, sizeof(*stats));
3173
3174         read_lock_bh(&bond->lock);
3175         bond_for_each_slave(bond, slave, iter) {
3176                 const struct rtnl_link_stats64 *sstats =
3177                         dev_get_stats(slave->dev, &temp);
3178
3179                 stats->rx_packets += sstats->rx_packets;
3180                 stats->rx_bytes += sstats->rx_bytes;
3181                 stats->rx_errors += sstats->rx_errors;
3182                 stats->rx_dropped += sstats->rx_dropped;
3183
3184                 stats->tx_packets += sstats->tx_packets;
3185                 stats->tx_bytes += sstats->tx_bytes;
3186                 stats->tx_errors += sstats->tx_errors;
3187                 stats->tx_dropped += sstats->tx_dropped;
3188
3189                 stats->multicast += sstats->multicast;
3190                 stats->collisions += sstats->collisions;
3191
3192                 stats->rx_length_errors += sstats->rx_length_errors;
3193                 stats->rx_over_errors += sstats->rx_over_errors;
3194                 stats->rx_crc_errors += sstats->rx_crc_errors;
3195                 stats->rx_frame_errors += sstats->rx_frame_errors;
3196                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3197                 stats->rx_missed_errors += sstats->rx_missed_errors;
3198
3199                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3200                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3201                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3202                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3203                 stats->tx_window_errors += sstats->tx_window_errors;
3204         }
3205         read_unlock_bh(&bond->lock);
3206
3207         return stats;
3208 }
3209
3210 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3211 {
3212         struct bonding *bond = netdev_priv(bond_dev);
3213         struct net_device *slave_dev = NULL;
3214         struct ifbond k_binfo;
3215         struct ifbond __user *u_binfo = NULL;
3216         struct ifslave k_sinfo;
3217         struct ifslave __user *u_sinfo = NULL;
3218         struct mii_ioctl_data *mii = NULL;
3219         struct bond_opt_value newval;
3220         struct net *net;
3221         int res = 0;
3222
3223         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3224
3225         switch (cmd) {
3226         case SIOCGMIIPHY:
3227                 mii = if_mii(ifr);
3228                 if (!mii)
3229                         return -EINVAL;
3230
3231                 mii->phy_id = 0;
3232                 /* Fall Through */
3233         case SIOCGMIIREG:
3234                 /*
3235                  * We do this again just in case we were called by SIOCGMIIREG
3236                  * instead of SIOCGMIIPHY.
3237                  */
3238                 mii = if_mii(ifr);
3239                 if (!mii)
3240                         return -EINVAL;
3241
3242
3243                 if (mii->reg_num == 1) {
3244                         mii->val_out = 0;
3245                         read_lock(&bond->lock);
3246                         read_lock(&bond->curr_slave_lock);
3247                         if (netif_carrier_ok(bond->dev))
3248                                 mii->val_out = BMSR_LSTATUS;
3249
3250                         read_unlock(&bond->curr_slave_lock);
3251                         read_unlock(&bond->lock);
3252                 }
3253
3254                 return 0;
3255         case BOND_INFO_QUERY_OLD:
3256         case SIOCBONDINFOQUERY:
3257                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3258
3259                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3260                         return -EFAULT;
3261
3262                 res = bond_info_query(bond_dev, &k_binfo);
3263                 if (res == 0 &&
3264                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3265                         return -EFAULT;
3266
3267                 return res;
3268         case BOND_SLAVE_INFO_QUERY_OLD:
3269         case SIOCBONDSLAVEINFOQUERY:
3270                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3271
3272                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3273                         return -EFAULT;
3274
3275                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3276                 if (res == 0 &&
3277                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3278                         return -EFAULT;
3279
3280                 return res;
3281         default:
3282                 /* Go on */
3283                 break;
3284         }
3285
3286         net = dev_net(bond_dev);
3287
3288         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3289                 return -EPERM;
3290
3291         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3292
3293         pr_debug("slave_dev=%p:\n", slave_dev);
3294
3295         if (!slave_dev)
3296                 return -ENODEV;
3297
3298         pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3299         switch (cmd) {
3300         case BOND_ENSLAVE_OLD:
3301         case SIOCBONDENSLAVE:
3302                 res = bond_enslave(bond_dev, slave_dev);
3303                 break;
3304         case BOND_RELEASE_OLD:
3305         case SIOCBONDRELEASE:
3306                 res = bond_release(bond_dev, slave_dev);
3307                 break;
3308         case BOND_SETHWADDR_OLD:
3309         case SIOCBONDSETHWADDR:
3310                 bond_set_dev_addr(bond_dev, slave_dev);
3311                 res = 0;
3312                 break;
3313         case BOND_CHANGE_ACTIVE_OLD:
3314         case SIOCBONDCHANGEACTIVE:
3315                 bond_opt_initstr(&newval, slave_dev->name);
3316                 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3317                 break;
3318         default:
3319                 res = -EOPNOTSUPP;
3320         }
3321
3322         return res;
3323 }
3324
3325 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3326 {
3327         struct bonding *bond = netdev_priv(bond_dev);
3328
3329         if (change & IFF_PROMISC)
3330                 bond_set_promiscuity(bond,
3331                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3332
3333         if (change & IFF_ALLMULTI)
3334                 bond_set_allmulti(bond,
3335                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3336 }
3337
3338 static void bond_set_rx_mode(struct net_device *bond_dev)
3339 {
3340         struct bonding *bond = netdev_priv(bond_dev);
3341         struct list_head *iter;
3342         struct slave *slave;
3343
3344
3345         rcu_read_lock();
3346         if (bond_uses_primary(bond)) {
3347                 slave = rcu_dereference(bond->curr_active_slave);
3348                 if (slave) {
3349                         dev_uc_sync(slave->dev, bond_dev);
3350                         dev_mc_sync(slave->dev, bond_dev);
3351                 }
3352         } else {
3353                 bond_for_each_slave_rcu(bond, slave, iter) {
3354                         dev_uc_sync_multiple(slave->dev, bond_dev);
3355                         dev_mc_sync_multiple(slave->dev, bond_dev);
3356                 }
3357         }
3358         rcu_read_unlock();
3359 }
3360
3361 static int bond_neigh_init(struct neighbour *n)
3362 {
3363         struct bonding *bond = netdev_priv(n->dev);
3364         const struct net_device_ops *slave_ops;
3365         struct neigh_parms parms;
3366         struct slave *slave;
3367         int ret;
3368
3369         slave = bond_first_slave(bond);
3370         if (!slave)
3371                 return 0;
3372         slave_ops = slave->dev->netdev_ops;
3373         if (!slave_ops->ndo_neigh_setup)
3374                 return 0;
3375
3376         parms.neigh_setup = NULL;
3377         parms.neigh_cleanup = NULL;
3378         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3379         if (ret)
3380                 return ret;
3381
3382         /*
3383          * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3384          * after the last slave has been detached.  Assumes that all slaves
3385          * utilize the same neigh_cleanup (true at this writing as only user
3386          * is ipoib).
3387          */
3388         n->parms->neigh_cleanup = parms.neigh_cleanup;
3389
3390         if (!parms.neigh_setup)
3391                 return 0;
3392
3393         return parms.neigh_setup(n);
3394 }
3395
3396 /*
3397  * The bonding ndo_neigh_setup is called at init time beofre any
3398  * slave exists. So we must declare proxy setup function which will
3399  * be used at run time to resolve the actual slave neigh param setup.
3400  *
3401  * It's also called by master devices (such as vlans) to setup their
3402  * underlying devices. In that case - do nothing, we're already set up from
3403  * our init.
3404  */
3405 static int bond_neigh_setup(struct net_device *dev,
3406                             struct neigh_parms *parms)
3407 {
3408         /* modify only our neigh_parms */
3409         if (parms->dev == dev)
3410                 parms->neigh_setup = bond_neigh_init;
3411
3412         return 0;
3413 }
3414
3415 /*
3416  * Change the MTU of all of a master's slaves to match the master
3417  */
3418 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3419 {
3420         struct bonding *bond = netdev_priv(bond_dev);
3421         struct slave *slave, *rollback_slave;
3422         struct list_head *iter;
3423         int res = 0;
3424
3425         pr_debug("bond=%p, name=%s, new_mtu=%d\n",
3426                  bond, bond_dev ? bond_dev->name : "None", new_mtu);
3427
3428         /* Can't hold bond->lock with bh disabled here since
3429          * some base drivers panic. On the other hand we can't
3430          * hold bond->lock without bh disabled because we'll
3431          * deadlock. The only solution is to rely on the fact
3432          * that we're under rtnl_lock here, and the slaves
3433          * list won't change. This doesn't solve the problem
3434          * of setting the slave's MTU while it is
3435          * transmitting, but the assumption is that the base
3436          * driver can handle that.
3437          *
3438          * TODO: figure out a way to safely iterate the slaves
3439          * list, but without holding a lock around the actual
3440          * call to the base driver.
3441          */
3442
3443         bond_for_each_slave(bond, slave, iter) {
3444                 pr_debug("s %p c_m %p\n",
3445                          slave, slave->dev->netdev_ops->ndo_change_mtu);
3446
3447                 res = dev_set_mtu(slave->dev, new_mtu);
3448
3449                 if (res) {
3450                         /* If we failed to set the slave's mtu to the new value
3451                          * we must abort the operation even in ACTIVE_BACKUP
3452                          * mode, because if we allow the backup slaves to have
3453                          * different mtu values than the active slave we'll
3454                          * need to change their mtu when doing a failover. That
3455                          * means changing their mtu from timer context, which
3456                          * is probably not a good idea.
3457                          */
3458                         pr_debug("err %d %s\n", res, slave->dev->name);
3459                         goto unwind;
3460                 }
3461         }
3462
3463         bond_dev->mtu = new_mtu;
3464
3465         return 0;
3466
3467 unwind:
3468         /* unwind from head to the slave that failed */
3469         bond_for_each_slave(bond, rollback_slave, iter) {
3470                 int tmp_res;
3471
3472                 if (rollback_slave == slave)
3473                         break;
3474
3475                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3476                 if (tmp_res) {
3477                         pr_debug("unwind err %d dev %s\n",
3478                                  tmp_res, rollback_slave->dev->name);
3479                 }
3480         }
3481
3482         return res;
3483 }
3484
3485 /*
3486  * Change HW address
3487  *
3488  * Note that many devices must be down to change the HW address, and
3489  * downing the master releases all slaves.  We can make bonds full of
3490  * bonding devices to test this, however.
3491  */
3492 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3493 {
3494         struct bonding *bond = netdev_priv(bond_dev);
3495         struct slave *slave, *rollback_slave;
3496         struct sockaddr *sa = addr, tmp_sa;
3497         struct list_head *iter;
3498         int res = 0;
3499
3500         if (BOND_MODE(bond) == BOND_MODE_ALB)
3501                 return bond_alb_set_mac_address(bond_dev, addr);
3502
3503
3504         pr_debug("bond=%p, name=%s\n",
3505                  bond, bond_dev ? bond_dev->name : "None");
3506
3507         /* If fail_over_mac is enabled, do nothing and return success.
3508          * Returning an error causes ifenslave to fail.
3509          */
3510         if (bond->params.fail_over_mac &&
3511             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3512                 return 0;
3513
3514         if (!is_valid_ether_addr(sa->sa_data))
3515                 return -EADDRNOTAVAIL;
3516
3517         /* Can't hold bond->lock with bh disabled here since
3518          * some base drivers panic. On the other hand we can't
3519          * hold bond->lock without bh disabled because we'll
3520          * deadlock. The only solution is to rely on the fact
3521          * that we're under rtnl_lock here, and the slaves
3522          * list won't change. This doesn't solve the problem
3523          * of setting the slave's hw address while it is
3524          * transmitting, but the assumption is that the base
3525          * driver can handle that.
3526          *
3527          * TODO: figure out a way to safely iterate the slaves
3528          * list, but without holding a lock around the actual
3529          * call to the base driver.
3530          */
3531
3532         bond_for_each_slave(bond, slave, iter) {
3533                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3534                 res = dev_set_mac_address(slave->dev, addr);
3535                 if (res) {
3536                         /* TODO: consider downing the slave
3537                          * and retry ?
3538                          * User should expect communications
3539                          * breakage anyway until ARP finish
3540                          * updating, so...
3541                          */
3542                         pr_debug("err %d %s\n", res, slave->dev->name);
3543                         goto unwind;
3544                 }
3545         }
3546
3547         /* success */
3548         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3549         return 0;
3550
3551 unwind:
3552         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3553         tmp_sa.sa_family = bond_dev->type;
3554
3555         /* unwind from head to the slave that failed */
3556         bond_for_each_slave(bond, rollback_slave, iter) {
3557                 int tmp_res;
3558
3559                 if (rollback_slave == slave)
3560                         break;
3561
3562                 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3563                 if (tmp_res) {
3564                         pr_debug("unwind err %d dev %s\n",
3565                                  tmp_res, rollback_slave->dev->name);
3566                 }
3567         }
3568
3569         return res;
3570 }
3571
3572 /**
3573  * bond_xmit_slave_id - transmit skb through slave with slave_id
3574  * @bond: bonding device that is transmitting
3575  * @skb: buffer to transmit
3576  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3577  *
3578  * This function tries to transmit through slave with slave_id but in case
3579  * it fails, it tries to find the first available slave for transmission.
3580  * The skb is consumed in all cases, thus the function is void.
3581  */
3582 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3583 {
3584         struct list_head *iter;
3585         struct slave *slave;
3586         int i = slave_id;
3587
3588         /* Here we start from the slave with slave_id */
3589         bond_for_each_slave_rcu(bond, slave, iter) {
3590                 if (--i < 0) {
3591                         if (bond_slave_can_tx(slave)) {
3592                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3593                                 return;
3594                         }
3595                 }
3596         }
3597
3598         /* Here we start from the first slave up to slave_id */
3599         i = slave_id;
3600         bond_for_each_slave_rcu(bond, slave, iter) {
3601                 if (--i < 0)
3602                         break;
3603                 if (bond_slave_can_tx(slave)) {
3604                         bond_dev_queue_xmit(bond, skb, slave->dev);
3605                         return;
3606                 }
3607         }
3608         /* no slave that can tx has been found */
3609         dev_kfree_skb_any(skb);
3610 }
3611
3612 /**
3613  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3614  * @bond: bonding device to use
3615  *
3616  * Based on the value of the bonding device's packets_per_slave parameter
3617  * this function generates a slave id, which is usually used as the next
3618  * slave to transmit through.
3619  */
3620 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3621 {
3622         u32 slave_id;
3623         struct reciprocal_value reciprocal_packets_per_slave;
3624         int packets_per_slave = bond->params.packets_per_slave;
3625
3626         switch (packets_per_slave) {
3627         case 0:
3628                 slave_id = prandom_u32();
3629                 break;
3630         case 1:
3631                 slave_id = bond->rr_tx_counter;
3632                 break;
3633         default:
3634                 reciprocal_packets_per_slave =
3635                         bond->params.reciprocal_packets_per_slave;
3636                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3637                                              reciprocal_packets_per_slave);
3638                 break;
3639         }
3640         bond->rr_tx_counter++;
3641
3642         return slave_id;
3643 }
3644
3645 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3646 {
3647         struct bonding *bond = netdev_priv(bond_dev);
3648         struct iphdr *iph = ip_hdr(skb);
3649         struct slave *slave;
3650         u32 slave_id;
3651
3652         /* Start with the curr_active_slave that joined the bond as the
3653          * default for sending IGMP traffic.  For failover purposes one
3654          * needs to maintain some consistency for the interface that will
3655          * send the join/membership reports.  The curr_active_slave found
3656          * will send all of this type of traffic.
3657          */
3658         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3659                 slave = rcu_dereference(bond->curr_active_slave);
3660                 if (slave && bond_slave_can_tx(slave))
3661                         bond_dev_queue_xmit(bond, skb, slave->dev);
3662                 else
3663                         bond_xmit_slave_id(bond, skb, 0);
3664         } else {
3665                 slave_id = bond_rr_gen_slave_id(bond);
3666                 bond_xmit_slave_id(bond, skb, slave_id % bond->slave_cnt);
3667         }
3668
3669         return NETDEV_TX_OK;
3670 }
3671
3672 /*
3673  * in active-backup mode, we know that bond->curr_active_slave is always valid if
3674  * the bond has a usable interface.
3675  */
3676 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3677 {
3678         struct bonding *bond = netdev_priv(bond_dev);
3679         struct slave *slave;
3680
3681         slave = rcu_dereference(bond->curr_active_slave);
3682         if (slave)
3683                 bond_dev_queue_xmit(bond, skb, slave->dev);
3684         else
3685                 dev_kfree_skb_any(skb);
3686
3687         return NETDEV_TX_OK;
3688 }
3689
3690 /* In bond_xmit_xor() , we determine the output device by using a pre-
3691  * determined xmit_hash_policy(), If the selected device is not enabled,
3692  * find the next active slave.
3693  */
3694 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3695 {
3696         struct bonding *bond = netdev_priv(bond_dev);
3697
3698         bond_xmit_slave_id(bond, skb, bond_xmit_hash(bond, skb) % bond->slave_cnt);
3699
3700         return NETDEV_TX_OK;
3701 }
3702
3703 /* in broadcast mode, we send everything to all usable interfaces. */
3704 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3705 {
3706         struct bonding *bond = netdev_priv(bond_dev);
3707         struct slave *slave = NULL;
3708         struct list_head *iter;
3709
3710         bond_for_each_slave_rcu(bond, slave, iter) {
3711                 if (bond_is_last_slave(bond, slave))
3712                         break;
3713                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3714                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3715
3716                         if (!skb2) {
3717                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3718                                                     bond_dev->name, __func__);
3719                                 continue;
3720                         }
3721                         /* bond_dev_queue_xmit always returns 0 */
3722                         bond_dev_queue_xmit(bond, skb2, slave->dev);
3723                 }
3724         }
3725         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3726                 bond_dev_queue_xmit(bond, skb, slave->dev);
3727         else
3728                 dev_kfree_skb_any(skb);
3729
3730         return NETDEV_TX_OK;
3731 }
3732
3733 /*------------------------- Device initialization ---------------------------*/
3734
3735 /*
3736  * Lookup the slave that corresponds to a qid
3737  */
3738 static inline int bond_slave_override(struct bonding *bond,
3739                                       struct sk_buff *skb)
3740 {
3741         struct slave *slave = NULL;
3742         struct list_head *iter;
3743
3744         if (!skb->queue_mapping)
3745                 return 1;
3746
3747         /* Find out if any slaves have the same mapping as this skb. */
3748         bond_for_each_slave_rcu(bond, slave, iter) {
3749                 if (slave->queue_id == skb->queue_mapping) {
3750                         if (bond_slave_can_tx(slave)) {
3751                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3752                                 return 0;
3753                         }
3754                         /* If the slave isn't UP, use default transmit policy. */
3755                         break;
3756                 }
3757         }
3758
3759         return 1;
3760 }
3761
3762
3763 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3764                              void *accel_priv, select_queue_fallback_t fallback)
3765 {
3766         /*
3767          * This helper function exists to help dev_pick_tx get the correct
3768          * destination queue.  Using a helper function skips a call to
3769          * skb_tx_hash and will put the skbs in the queue we expect on their
3770          * way down to the bonding driver.
3771          */
3772         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3773
3774         /*
3775          * Save the original txq to restore before passing to the driver
3776          */
3777         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3778
3779         if (unlikely(txq >= dev->real_num_tx_queues)) {
3780                 do {
3781                         txq -= dev->real_num_tx_queues;
3782                 } while (txq >= dev->real_num_tx_queues);
3783         }
3784         return txq;
3785 }
3786
3787 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3788 {
3789         struct bonding *bond = netdev_priv(dev);
3790
3791         if (bond_should_override_tx_queue(bond) &&
3792             !bond_slave_override(bond, skb))
3793                 return NETDEV_TX_OK;
3794
3795         switch (BOND_MODE(bond)) {
3796         case BOND_MODE_ROUNDROBIN:
3797                 return bond_xmit_roundrobin(skb, dev);
3798         case BOND_MODE_ACTIVEBACKUP:
3799                 return bond_xmit_activebackup(skb, dev);
3800         case BOND_MODE_XOR:
3801                 return bond_xmit_xor(skb, dev);
3802         case BOND_MODE_BROADCAST:
3803                 return bond_xmit_broadcast(skb, dev);
3804         case BOND_MODE_8023AD:
3805                 return bond_3ad_xmit_xor(skb, dev);
3806         case BOND_MODE_ALB:
3807                 return bond_alb_xmit(skb, dev);
3808         case BOND_MODE_TLB:
3809                 return bond_tlb_xmit(skb, dev);
3810         default:
3811                 /* Should never happen, mode already checked */
3812                 pr_err("%s: Error: Unknown bonding mode %d\n",
3813                        dev->name, BOND_MODE(bond));
3814                 WARN_ON_ONCE(1);
3815                 dev_kfree_skb_any(skb);
3816                 return NETDEV_TX_OK;
3817         }
3818 }
3819
3820 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3821 {
3822         struct bonding *bond = netdev_priv(dev);
3823         netdev_tx_t ret = NETDEV_TX_OK;
3824
3825         /*
3826          * If we risk deadlock from transmitting this in the
3827          * netpoll path, tell netpoll to queue the frame for later tx
3828          */
3829         if (unlikely(is_netpoll_tx_blocked(dev)))
3830                 return NETDEV_TX_BUSY;
3831
3832         rcu_read_lock();
3833         if (bond_has_slaves(bond))
3834                 ret = __bond_start_xmit(skb, dev);
3835         else
3836                 dev_kfree_skb_any(skb);
3837         rcu_read_unlock();
3838
3839         return ret;
3840 }
3841
3842 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3843                                      struct ethtool_cmd *ecmd)
3844 {
3845         struct bonding *bond = netdev_priv(bond_dev);
3846         unsigned long speed = 0;
3847         struct list_head *iter;
3848         struct slave *slave;
3849
3850         ecmd->duplex = DUPLEX_UNKNOWN;
3851         ecmd->port = PORT_OTHER;
3852
3853         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
3854          * do not need to check mode.  Though link speed might not represent
3855          * the true receive or transmit bandwidth (not all modes are symmetric)
3856          * this is an accurate maximum.
3857          */
3858         read_lock(&bond->lock);
3859         bond_for_each_slave(bond, slave, iter) {
3860                 if (bond_slave_can_tx(slave)) {
3861                         if (slave->speed != SPEED_UNKNOWN)
3862                                 speed += slave->speed;
3863                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
3864                             slave->duplex != DUPLEX_UNKNOWN)
3865                                 ecmd->duplex = slave->duplex;
3866                 }
3867         }
3868         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3869         read_unlock(&bond->lock);
3870
3871         return 0;
3872 }
3873
3874 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3875                                      struct ethtool_drvinfo *drvinfo)
3876 {
3877         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3878         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3879         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3880                  BOND_ABI_VERSION);
3881 }
3882
3883 static const struct ethtool_ops bond_ethtool_ops = {
3884         .get_drvinfo            = bond_ethtool_get_drvinfo,
3885         .get_settings           = bond_ethtool_get_settings,
3886         .get_link               = ethtool_op_get_link,
3887 };
3888
3889 static const struct net_device_ops bond_netdev_ops = {
3890         .ndo_init               = bond_init,
3891         .ndo_uninit             = bond_uninit,
3892         .ndo_open               = bond_open,
3893         .ndo_stop               = bond_close,
3894         .ndo_start_xmit         = bond_start_xmit,
3895         .ndo_select_queue       = bond_select_queue,
3896         .ndo_get_stats64        = bond_get_stats,
3897         .ndo_do_ioctl           = bond_do_ioctl,
3898         .ndo_change_rx_flags    = bond_change_rx_flags,
3899         .ndo_set_rx_mode        = bond_set_rx_mode,
3900         .ndo_change_mtu         = bond_change_mtu,
3901         .ndo_set_mac_address    = bond_set_mac_address,
3902         .ndo_neigh_setup        = bond_neigh_setup,
3903         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
3904         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
3905 #ifdef CONFIG_NET_POLL_CONTROLLER
3906         .ndo_netpoll_setup      = bond_netpoll_setup,
3907         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
3908         .ndo_poll_controller    = bond_poll_controller,
3909 #endif
3910         .ndo_add_slave          = bond_enslave,
3911         .ndo_del_slave          = bond_release,
3912         .ndo_fix_features       = bond_fix_features,
3913 };
3914
3915 static const struct device_type bond_type = {
3916         .name = "bond",
3917 };
3918
3919 static void bond_destructor(struct net_device *bond_dev)
3920 {
3921         struct bonding *bond = netdev_priv(bond_dev);
3922         if (bond->wq)
3923                 destroy_workqueue(bond->wq);
3924         free_netdev(bond_dev);
3925 }
3926
3927 void bond_setup(struct net_device *bond_dev)
3928 {
3929         struct bonding *bond = netdev_priv(bond_dev);
3930
3931         /* initialize rwlocks */
3932         rwlock_init(&bond->lock);
3933         rwlock_init(&bond->curr_slave_lock);
3934         bond->params = bonding_defaults;
3935
3936         /* Initialize pointers */
3937         bond->dev = bond_dev;
3938
3939         /* Initialize the device entry points */
3940         ether_setup(bond_dev);
3941         bond_dev->netdev_ops = &bond_netdev_ops;
3942         bond_dev->ethtool_ops = &bond_ethtool_ops;
3943
3944         bond_dev->destructor = bond_destructor;
3945
3946         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
3947
3948         /* Initialize the device options */
3949         bond_dev->tx_queue_len = 0;
3950         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
3951         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT;
3952         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
3953
3954         /* don't acquire bond device's netif_tx_lock when
3955          * transmitting */
3956         bond_dev->features |= NETIF_F_LLTX;
3957
3958         /* By default, we declare the bond to be fully
3959          * VLAN hardware accelerated capable. Special
3960          * care is taken in the various xmit functions
3961          * when there are slaves that are not hw accel
3962          * capable
3963          */
3964
3965         /* Don't allow bond devices to change network namespaces. */
3966         bond_dev->features |= NETIF_F_NETNS_LOCAL;
3967
3968         bond_dev->hw_features = BOND_VLAN_FEATURES |
3969                                 NETIF_F_HW_VLAN_CTAG_TX |
3970                                 NETIF_F_HW_VLAN_CTAG_RX |
3971                                 NETIF_F_HW_VLAN_CTAG_FILTER;
3972
3973         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
3974         bond_dev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
3975         bond_dev->features |= bond_dev->hw_features;
3976 }
3977
3978 /*
3979 * Destroy a bonding device.
3980 * Must be under rtnl_lock when this function is called.
3981 */
3982 static void bond_uninit(struct net_device *bond_dev)
3983 {
3984         struct bonding *bond = netdev_priv(bond_dev);
3985         struct list_head *iter;
3986         struct slave *slave;
3987
3988         bond_netpoll_cleanup(bond_dev);
3989
3990         /* Release the bonded slaves */
3991         bond_for_each_slave(bond, slave, iter)
3992                 __bond_release_one(bond_dev, slave->dev, true);
3993         pr_info("%s: Released all slaves\n", bond_dev->name);
3994
3995         list_del(&bond->bond_list);
3996
3997         bond_debug_unregister(bond);
3998 }
3999
4000 /*------------------------- Module initialization ---------------------------*/
4001
4002 static int bond_check_params(struct bond_params *params)
4003 {
4004         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4005         struct bond_opt_value newval;
4006         const struct bond_opt_value *valptr;
4007         int arp_all_targets_value;
4008
4009         /*
4010          * Convert string parameters.
4011          */
4012         if (mode) {
4013                 bond_opt_initstr(&newval, mode);
4014                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4015                 if (!valptr) {
4016                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4017                         return -EINVAL;
4018                 }
4019                 bond_mode = valptr->value;
4020         }
4021
4022         if (xmit_hash_policy) {
4023                 if ((bond_mode != BOND_MODE_XOR) &&
4024                     (bond_mode != BOND_MODE_8023AD) &&
4025                     (bond_mode != BOND_MODE_TLB)) {
4026                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4027                                 bond_mode_name(bond_mode));
4028                 } else {
4029                         bond_opt_initstr(&newval, xmit_hash_policy);
4030                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4031                                                 &newval);
4032                         if (!valptr) {
4033                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4034                                        xmit_hash_policy);
4035                                 return -EINVAL;
4036                         }
4037                         xmit_hashtype = valptr->value;
4038                 }
4039         }
4040
4041         if (lacp_rate) {
4042                 if (bond_mode != BOND_MODE_8023AD) {
4043                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4044                                 bond_mode_name(bond_mode));
4045                 } else {
4046                         bond_opt_initstr(&newval, lacp_rate);
4047                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4048                                                 &newval);
4049                         if (!valptr) {
4050                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4051                                        lacp_rate);
4052                                 return -EINVAL;
4053                         }
4054                         lacp_fast = valptr->value;
4055                 }
4056         }
4057
4058         if (ad_select) {
4059                 bond_opt_initstr(&newval, lacp_rate);
4060                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4061                                         &newval);
4062                 if (!valptr) {
4063                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4064                         return -EINVAL;
4065                 }
4066                 params->ad_select = valptr->value;
4067                 if (bond_mode != BOND_MODE_8023AD)
4068                         pr_warn("ad_select param only affects 802.3ad mode\n");
4069         } else {
4070                 params->ad_select = BOND_AD_STABLE;
4071         }
4072
4073         if (max_bonds < 0) {
4074                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4075                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4076                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4077         }
4078
4079         if (miimon < 0) {
4080                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4081                         miimon, INT_MAX);
4082                 miimon = 0;
4083         }
4084
4085         if (updelay < 0) {
4086                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4087                         updelay, INT_MAX);
4088                 updelay = 0;
4089         }
4090
4091         if (downdelay < 0) {
4092                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4093                         downdelay, INT_MAX);
4094                 downdelay = 0;
4095         }
4096
4097         if ((use_carrier != 0) && (use_carrier != 1)) {
4098                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4099                         use_carrier);
4100                 use_carrier = 1;
4101         }
4102
4103         if (num_peer_notif < 0 || num_peer_notif > 255) {
4104                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4105                         num_peer_notif);
4106                 num_peer_notif = 1;
4107         }
4108
4109         /* reset values for 802.3ad/TLB/ALB */
4110         if (!bond_mode_uses_arp(bond_mode)) {
4111                 if (!miimon) {
4112                         pr_warn("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4113                         pr_warn("Forcing miimon to 100msec\n");
4114                         miimon = BOND_DEFAULT_MIIMON;
4115                 }
4116         }
4117
4118         if (tx_queues < 1 || tx_queues > 255) {
4119                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4120                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4121                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4122         }
4123
4124         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4125                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4126                         all_slaves_active);
4127                 all_slaves_active = 0;
4128         }
4129
4130         if (resend_igmp < 0 || resend_igmp > 255) {
4131                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4132                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4133                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4134         }
4135
4136         bond_opt_initval(&newval, packets_per_slave);
4137         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4138                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4139                         packets_per_slave, USHRT_MAX);
4140                 packets_per_slave = 1;
4141         }
4142
4143         if (bond_mode == BOND_MODE_ALB) {
4144                 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",
4145                           updelay);
4146         }
4147
4148         if (!miimon) {
4149                 if (updelay || downdelay) {
4150                         /* just warn the user the up/down delay will have
4151                          * no effect since miimon is zero...
4152                          */
4153                         pr_warn("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",
4154                                 updelay, downdelay);
4155                 }
4156         } else {
4157                 /* don't allow arp monitoring */
4158                 if (arp_interval) {
4159                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4160                                 miimon, arp_interval);
4161                         arp_interval = 0;
4162                 }
4163
4164                 if ((updelay % miimon) != 0) {
4165                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4166                                 updelay, miimon, (updelay / miimon) * miimon);
4167                 }
4168
4169                 updelay /= miimon;
4170
4171                 if ((downdelay % miimon) != 0) {
4172                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4173                                 downdelay, miimon,
4174                                 (downdelay / miimon) * miimon);
4175                 }
4176
4177                 downdelay /= miimon;
4178         }
4179
4180         if (arp_interval < 0) {
4181                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4182                         arp_interval, INT_MAX);
4183                 arp_interval = 0;
4184         }
4185
4186         for (arp_ip_count = 0, i = 0;
4187              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4188                 /* not complete check, but should be good enough to
4189                    catch mistakes */
4190                 __be32 ip;
4191                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4192                     !bond_is_ip_target_ok(ip)) {
4193                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4194                                 arp_ip_target[i]);
4195                         arp_interval = 0;
4196                 } else {
4197                         if (bond_get_targets_ip(arp_target, ip) == -1)
4198                                 arp_target[arp_ip_count++] = ip;
4199                         else
4200                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4201                                         &ip);
4202                 }
4203         }
4204
4205         if (arp_interval && !arp_ip_count) {
4206                 /* don't allow arping if no arp_ip_target given... */
4207                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4208                         arp_interval);
4209                 arp_interval = 0;
4210         }
4211
4212         if (arp_validate) {
4213                 if (!arp_interval) {
4214                         pr_err("arp_validate requires arp_interval\n");
4215                         return -EINVAL;
4216                 }
4217
4218                 bond_opt_initstr(&newval, arp_validate);
4219                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4220                                         &newval);
4221                 if (!valptr) {
4222                         pr_err("Error: invalid arp_validate \"%s\"\n",
4223                                arp_validate);
4224                         return -EINVAL;
4225                 }
4226                 arp_validate_value = valptr->value;
4227         } else {
4228                 arp_validate_value = 0;
4229         }
4230
4231         arp_all_targets_value = 0;
4232         if (arp_all_targets) {
4233                 bond_opt_initstr(&newval, arp_all_targets);
4234                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4235                                         &newval);
4236                 if (!valptr) {
4237                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4238                                arp_all_targets);
4239                         arp_all_targets_value = 0;
4240                 } else {
4241                         arp_all_targets_value = valptr->value;
4242                 }
4243         }
4244
4245         if (miimon) {
4246                 pr_info("MII link monitoring set to %d ms\n", miimon);
4247         } else if (arp_interval) {
4248                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4249                                           arp_validate_value);
4250                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4251                         arp_interval, valptr->string, arp_ip_count);
4252
4253                 for (i = 0; i < arp_ip_count; i++)
4254                         pr_cont(" %s", arp_ip_target[i]);
4255
4256                 pr_cont("\n");
4257
4258         } else if (max_bonds) {
4259                 /* miimon and arp_interval not set, we need one so things
4260                  * work as expected, see bonding.txt for details
4261                  */
4262                 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");
4263         }
4264
4265         if (primary && !bond_mode_uses_primary(bond_mode)) {
4266                 /* currently, using a primary only makes sense
4267                  * in active backup, TLB or ALB modes
4268                  */
4269                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4270                         primary, bond_mode_name(bond_mode));
4271                 primary = NULL;
4272         }
4273
4274         if (primary && primary_reselect) {
4275                 bond_opt_initstr(&newval, primary_reselect);
4276                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4277                                         &newval);
4278                 if (!valptr) {
4279                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4280                                primary_reselect);
4281                         return -EINVAL;
4282                 }
4283                 primary_reselect_value = valptr->value;
4284         } else {
4285                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4286         }
4287
4288         if (fail_over_mac) {
4289                 bond_opt_initstr(&newval, fail_over_mac);
4290                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4291                                         &newval);
4292                 if (!valptr) {
4293                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4294                                fail_over_mac);
4295                         return -EINVAL;
4296                 }
4297                 fail_over_mac_value = valptr->value;
4298                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4299                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4300         } else {
4301                 fail_over_mac_value = BOND_FOM_NONE;
4302         }
4303
4304         if (lp_interval == 0) {
4305                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4306                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4307                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4308         }
4309
4310         /* fill params struct with the proper values */
4311         params->mode = bond_mode;
4312         params->xmit_policy = xmit_hashtype;
4313         params->miimon = miimon;
4314         params->num_peer_notif = num_peer_notif;
4315         params->arp_interval = arp_interval;
4316         params->arp_validate = arp_validate_value;
4317         params->arp_all_targets = arp_all_targets_value;
4318         params->updelay = updelay;
4319         params->downdelay = downdelay;
4320         params->use_carrier = use_carrier;
4321         params->lacp_fast = lacp_fast;
4322         params->primary[0] = 0;
4323         params->primary_reselect = primary_reselect_value;
4324         params->fail_over_mac = fail_over_mac_value;
4325         params->tx_queues = tx_queues;
4326         params->all_slaves_active = all_slaves_active;
4327         params->resend_igmp = resend_igmp;
4328         params->min_links = min_links;
4329         params->lp_interval = lp_interval;
4330         params->packets_per_slave = packets_per_slave;
4331         params->tlb_dynamic_lb = 1; /* Default value */
4332         if (packets_per_slave > 0) {
4333                 params->reciprocal_packets_per_slave =
4334                         reciprocal_value(packets_per_slave);
4335         } else {
4336                 /* reciprocal_packets_per_slave is unused if
4337                  * packets_per_slave is 0 or 1, just initialize it
4338                  */
4339                 params->reciprocal_packets_per_slave =
4340                         (struct reciprocal_value) { 0 };
4341         }
4342
4343         if (primary) {
4344                 strncpy(params->primary, primary, IFNAMSIZ);
4345                 params->primary[IFNAMSIZ - 1] = 0;
4346         }
4347
4348         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4349
4350         return 0;
4351 }
4352
4353 static struct lock_class_key bonding_netdev_xmit_lock_key;
4354 static struct lock_class_key bonding_netdev_addr_lock_key;
4355 static struct lock_class_key bonding_tx_busylock_key;
4356
4357 static void bond_set_lockdep_class_one(struct net_device *dev,
4358                                        struct netdev_queue *txq,
4359                                        void *_unused)
4360 {
4361         lockdep_set_class(&txq->_xmit_lock,
4362                           &bonding_netdev_xmit_lock_key);
4363 }
4364
4365 static void bond_set_lockdep_class(struct net_device *dev)
4366 {
4367         lockdep_set_class(&dev->addr_list_lock,
4368                           &bonding_netdev_addr_lock_key);
4369         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4370         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4371 }
4372
4373 /*
4374  * Called from registration process
4375  */
4376 static int bond_init(struct net_device *bond_dev)
4377 {
4378         struct bonding *bond = netdev_priv(bond_dev);
4379         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4380         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4381
4382         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4383
4384         /*
4385          * Initialize locks that may be required during
4386          * en/deslave operations.  All of the bond_open work
4387          * (of which this is part) should really be moved to
4388          * a phase prior to dev_open
4389          */
4390         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4391         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4392
4393         bond->wq = create_singlethread_workqueue(bond_dev->name);
4394         if (!bond->wq)
4395                 return -ENOMEM;
4396
4397         bond_set_lockdep_class(bond_dev);
4398
4399         list_add_tail(&bond->bond_list, &bn->dev_list);
4400
4401         bond_prepare_sysfs_group(bond);
4402
4403         bond_debug_register(bond);
4404
4405         /* Ensure valid dev_addr */
4406         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4407             bond_dev->addr_assign_type == NET_ADDR_PERM)
4408                 eth_hw_addr_random(bond_dev);
4409
4410         return 0;
4411 }
4412
4413 unsigned int bond_get_num_tx_queues(void)
4414 {
4415         return tx_queues;
4416 }
4417
4418 /* Create a new bond based on the specified name and bonding parameters.
4419  * If name is NULL, obtain a suitable "bond%d" name for us.
4420  * Caller must NOT hold rtnl_lock; we need to release it here before we
4421  * set up our sysfs entries.
4422  */
4423 int bond_create(struct net *net, const char *name)
4424 {
4425         struct net_device *bond_dev;
4426         int res;
4427
4428         rtnl_lock();
4429
4430         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4431                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
4432                                    bond_setup, tx_queues);
4433         if (!bond_dev) {
4434                 pr_err("%s: eek! can't alloc netdev!\n", name);
4435                 rtnl_unlock();
4436                 return -ENOMEM;
4437         }
4438
4439         dev_net_set(bond_dev, net);
4440         bond_dev->rtnl_link_ops = &bond_link_ops;
4441
4442         res = register_netdevice(bond_dev);
4443
4444         netif_carrier_off(bond_dev);
4445
4446         rtnl_unlock();
4447         if (res < 0)
4448                 bond_destructor(bond_dev);
4449         return res;
4450 }
4451
4452 static int __net_init bond_net_init(struct net *net)
4453 {
4454         struct bond_net *bn = net_generic(net, bond_net_id);
4455
4456         bn->net = net;
4457         INIT_LIST_HEAD(&bn->dev_list);
4458
4459         bond_create_proc_dir(bn);
4460         bond_create_sysfs(bn);
4461
4462         return 0;
4463 }
4464
4465 static void __net_exit bond_net_exit(struct net *net)
4466 {
4467         struct bond_net *bn = net_generic(net, bond_net_id);
4468         struct bonding *bond, *tmp_bond;
4469         LIST_HEAD(list);
4470
4471         bond_destroy_sysfs(bn);
4472         bond_destroy_proc_dir(bn);
4473
4474         /* Kill off any bonds created after unregistering bond rtnl ops */
4475         rtnl_lock();
4476         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4477                 unregister_netdevice_queue(bond->dev, &list);
4478         unregister_netdevice_many(&list);
4479         rtnl_unlock();
4480 }
4481
4482 static struct pernet_operations bond_net_ops = {
4483         .init = bond_net_init,
4484         .exit = bond_net_exit,
4485         .id   = &bond_net_id,
4486         .size = sizeof(struct bond_net),
4487 };
4488
4489 static int __init bonding_init(void)
4490 {
4491         int i;
4492         int res;
4493
4494         pr_info("%s", bond_version);
4495
4496         res = bond_check_params(&bonding_defaults);
4497         if (res)
4498                 goto out;
4499
4500         res = register_pernet_subsys(&bond_net_ops);
4501         if (res)
4502                 goto out;
4503
4504         res = bond_netlink_init();
4505         if (res)
4506                 goto err_link;
4507
4508         bond_create_debugfs();
4509
4510         for (i = 0; i < max_bonds; i++) {
4511                 res = bond_create(&init_net, NULL);
4512                 if (res)
4513                         goto err;
4514         }
4515
4516         register_netdevice_notifier(&bond_netdev_notifier);
4517 out:
4518         return res;
4519 err:
4520         bond_destroy_debugfs();
4521         bond_netlink_fini();
4522 err_link:
4523         unregister_pernet_subsys(&bond_net_ops);
4524         goto out;
4525
4526 }
4527
4528 static void __exit bonding_exit(void)
4529 {
4530         unregister_netdevice_notifier(&bond_netdev_notifier);
4531
4532         bond_destroy_debugfs();
4533
4534         bond_netlink_fini();
4535         unregister_pernet_subsys(&bond_net_ops);
4536
4537 #ifdef CONFIG_NET_POLL_CONTROLLER
4538         /*
4539          * Make sure we don't have an imbalance on our netpoll blocking
4540          */
4541         WARN_ON(atomic_read(&netpoll_block_tx));
4542 #endif
4543 }
4544
4545 module_init(bonding_init);
4546 module_exit(bonding_exit);
4547 MODULE_LICENSE("GPL");
4548 MODULE_VERSION(DRV_VERSION);
4549 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4550 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");