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 #include <linux/kernel.h>
  35 #include <linux/module.h>
  36 #include <linux/types.h>
  37 #include <linux/fcntl.h>
  38 #include <linux/interrupt.h>
  39 #include <linux/ptrace.h>
  40 #include <linux/ioport.h>
  41 #include <linux/in.h>
  42 #include <net/ip.h>
  43 #include <linux/ip.h>
  44 #include <linux/tcp.h>
  45 #include <linux/udp.h>
  46 #include <linux/slab.h>
  47 #include <linux/string.h>
  48 #include <linux/init.h>
  49 #include <linux/timer.h>
  50 #include <linux/socket.h>
  51 #include <linux/ctype.h>
  52 #include <linux/inet.h>
  53 #include <linux/bitops.h>
  54 #include <linux/io.h>
  55 #include <asm/dma.h>
  56 #include <linux/uaccess.h>
  57 #include <linux/errno.h>
  58 #include <linux/netdevice.h>
  59 #include <linux/inetdevice.h>
  60 #include <linux/igmp.h>
  61 #include <linux/etherdevice.h>
  62 #include <linux/skbuff.h>
  63 #include <net/sock.h>
  64 #include <linux/rtnetlink.h>
  65 #include <linux/smp.h>
  66 #include <linux/if_ether.h>
  67 #include <net/arp.h>
  68 #include <linux/mii.h>
  69 #include <linux/ethtool.h>
  70 #include <linux/if_vlan.h>
  71 #include <linux/if_bonding.h>
  72 #include <linux/jiffies.h>
  73 #include <linux/preempt.h>
  74 #include <net/route.h>
  75 #include <net/net_namespace.h>
  76 #include <net/netns/generic.h>
  77 #include <net/pkt_sched.h>
  78 #include <linux/rculist.h>
  79 #include <net/flow_dissector.h>
  80 #include <net/switchdev.h>
  81 #include <net/bonding.h>
  82 #include <net/bond_3ad.h>
  83 #include <net/bond_alb.h>
  84
  85 #include "bonding_priv.h"
  86
  87 /*---------------------------- Module parameters ----------------------------*/
  88
  89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  90
  91 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  92 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  93 static int num_peer_notif = 1;
  94 static int miimon;
  95 static int updelay;
  96 static int downdelay;
  97 static int use_carrier  = 1;
  98 static char *mode;
  99 static char *primary;
 100 static char *primary_reselect;
 101 static char *lacp_rate;
 102 static int min_links;
 103 static char *ad_select;
 104 static char *xmit_hash_policy;
 105 static int arp_interval;
 106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 107 static char *arp_validate;
 108 static char *arp_all_targets;
 109 static char *fail_over_mac;
 110 static int all_slaves_active;
 111 static struct bond_params bonding_defaults;
 112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 113 static int packets_per_slave = 1;
 114 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
 115
 116 module_param(max_bonds, int, 0);
 117 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 118 module_param(tx_queues, int, 0);
 119 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 120 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 121 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 122                                "failover event (alias of num_unsol_na)");
 123 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 124 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 125                                "failover event (alias of num_grat_arp)");
 126 module_param(miimon, int, 0);
 127 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 128 module_param(updelay, int, 0);
 129 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 130 module_param(downdelay, int, 0);
 131 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 132                             "in milliseconds");
 133 module_param(use_carrier, int, 0);
 134 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 135                               "0 for off, 1 for on (default)");
 136 module_param(mode, charp, 0);
 137 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 138                        "1 for active-backup, 2 for balance-xor, "
 139                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 140                        "6 for balance-alb");
 141 module_param(primary, charp, 0);
 142 MODULE_PARM_DESC(primary, "Primary network device to use");
 143 module_param(primary_reselect, charp, 0);
 144 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 145                                    "once it comes up; "
 146                                    "0 for always (default), "
 147                                    "1 for only if speed of primary is "
 148                                    "better, "
 149                                    "2 for only on active slave "
 150                                    "failure");
 151 module_param(lacp_rate, charp, 0);
 152 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 153                             "0 for slow, 1 for fast");
 154 module_param(ad_select, charp, 0);
 155 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
 156                             "0 for stable (default), 1 for bandwidth, "
 157                             "2 for count");
 158 module_param(min_links, int, 0);
 159 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 160
 161 module_param(xmit_hash_policy, charp, 0);
 162 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
 163                                    "0 for layer 2 (default), 1 for layer 3+4, "
 164                                    "2 for layer 2+3, 3 for encap layer 2+3, "
 165                                    "4 for encap layer 3+4");
 166 module_param(arp_interval, int, 0);
 167 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 168 module_param_array(arp_ip_target, charp, NULL, 0);
 169 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 170 module_param(arp_validate, charp, 0);
 171 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 172                                "0 for none (default), 1 for active, "
 173                                "2 for backup, 3 for all");
 174 module_param(arp_all_targets, charp, 0);
 175 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 176 module_param(fail_over_mac, charp, 0);
 177 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 178                                 "the same MAC; 0 for none (default), "
 179                                 "1 for active, 2 for follow");
 180 module_param(all_slaves_active, int, 0);
 181 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
 182                                      "by setting active flag for all slaves; "
 183                                      "0 for never (default), 1 for always.");
 184 module_param(resend_igmp, int, 0);
 185 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 186                               "link failure");
 187 module_param(packets_per_slave, int, 0);
 188 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
 189                                     "mode; 0 for a random slave, 1 packet per "
 190                                     "slave (default), >1 packets per slave.");
 191 module_param(lp_interval, uint, 0);
 192 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
 193                               "the bonding driver sends learning packets to "
 194                               "each slaves peer switch. The default is 1.");
 195
 196 /*----------------------------- Global variables ----------------------------*/
 197
 198 #ifdef CONFIG_NET_POLL_CONTROLLER
 199 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 200 #endif
 201
 202 unsigned int bond_net_id __read_mostly;
 203
 204 /*-------------------------- Forward declarations ---------------------------*/
 205
 206 static int bond_init(struct net_device *bond_dev);
 207 static void bond_uninit(struct net_device *bond_dev);
 208 static void bond_get_stats(struct net_device *bond_dev,
 209                            struct rtnl_link_stats64 *stats);
 210 static void bond_slave_arr_handler(struct work_struct *work);
 211 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
 212                                   int mod);
 213 static void bond_netdev_notify_work(struct work_struct *work);
 214
 215 /*---------------------------- General routines -----------------------------*/
 216
 217 const char *bond_mode_name(int mode)
 218 {
 219         static const char *names[] = {
 220                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 221                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 222                 [BOND_MODE_XOR] = "load balancing (xor)",
 223                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 224                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 225                 [BOND_MODE_TLB] = "transmit load balancing",
 226                 [BOND_MODE_ALB] = "adaptive load balancing",
 227         };
 228
 229         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 230                 return "unknown";
 231
 232         return names[mode];
 233 }
 234
 235 /*---------------------------------- VLAN -----------------------------------*/
 236
 237 /**
 238  * bond_dev_queue_xmit - Prepare skb for xmit.
 239  *
 240  * @bond: bond device that got this skb for tx.
 241  * @skb: hw accel VLAN tagged skb to transmit
 242  * @slave_dev: slave that is supposed to xmit this skbuff
 243  */
 244 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 245                         struct net_device *slave_dev)
 246 {
 247         skb->dev = slave_dev;
 248
 249         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 250                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 251         skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
 252
 253         if (unlikely(netpoll_tx_running(bond->dev)))
 254                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 255         else
 256                 dev_queue_xmit(skb);
 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 /* Set the carrier state for the master according to the state of its
 332  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 333  * do special 802.3ad magic.
 334  *
 335  * Returns zero if carrier state does not change, nonzero if it does.
 336  */
 337 int bond_set_carrier(struct bonding *bond)
 338 {
 339         struct list_head *iter;
 340         struct slave *slave;
 341
 342         if (!bond_has_slaves(bond))
 343                 goto down;
 344
 345         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 346                 return bond_3ad_set_carrier(bond);
 347
 348         bond_for_each_slave(bond, slave, iter) {
 349                 if (slave->link == BOND_LINK_UP) {
 350                         if (!netif_carrier_ok(bond->dev)) {
 351                                 netif_carrier_on(bond->dev);
 352                                 return 1;
 353                         }
 354                         return 0;
 355                 }
 356         }
 357
 358 down:
 359         if (netif_carrier_ok(bond->dev)) {
 360                 netif_carrier_off(bond->dev);
 361                 return 1;
 362         }
 363         return 0;
 364 }
 365
 366 /* Get link speed and duplex from the slave's base driver
 367  * using ethtool. If for some reason the call fails or the
 368  * values are invalid, set speed and duplex to -1,
 369  * and return. Return 1 if speed or duplex settings are
 370  * UNKNOWN; 0 otherwise.
 371  */
 372 static int bond_update_speed_duplex(struct slave *slave)
 373 {
 374         struct net_device *slave_dev = slave->dev;
 375         struct ethtool_link_ksettings ecmd;
 376         int res;
 377
 378         slave->speed = SPEED_UNKNOWN;
 379         slave->duplex = DUPLEX_UNKNOWN;
 380
 381         res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
 382         if (res < 0)
 383                 return 1;
 384         if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
 385                 return 1;
 386         switch (ecmd.base.duplex) {
 387         case DUPLEX_FULL:
 388         case DUPLEX_HALF:
 389                 break;
 390         default:
 391                 return 1;
 392         }
 393
 394         slave->speed = ecmd.base.speed;
 395         slave->duplex = ecmd.base.duplex;
 396
 397         return 0;
 398 }
 399
 400 const char *bond_slave_link_status(s8 link)
 401 {
 402         switch (link) {
 403         case BOND_LINK_UP:
 404                 return "up";
 405         case BOND_LINK_FAIL:
 406                 return "going down";
 407         case BOND_LINK_DOWN:
 408                 return "down";
 409         case BOND_LINK_BACK:
 410                 return "going back";
 411         default:
 412                 return "unknown";
 413         }
 414 }
 415
 416 /* if <dev> supports MII link status reporting, check its link status.
 417  *
 418  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 419  * depending upon the setting of the use_carrier parameter.
 420  *
 421  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 422  * can't tell and just pretend it is), or 0, meaning that the link is
 423  * down.
 424  *
 425  * If reporting is non-zero, instead of faking link up, return -1 if
 426  * both ETHTOOL and MII ioctls fail (meaning the device does not
 427  * support them).  If use_carrier is set, return whatever it says.
 428  * It'd be nice if there was a good way to tell if a driver supports
 429  * netif_carrier, but there really isn't.
 430  */
 431 static int bond_check_dev_link(struct bonding *bond,
 432                                struct net_device *slave_dev, int reporting)
 433 {
 434         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 435         int (*ioctl)(struct net_device *, struct ifreq *, int);
 436         struct ifreq ifr;
 437         struct mii_ioctl_data *mii;
 438
 439         if (!reporting && !netif_running(slave_dev))
 440                 return 0;
 441
 442         if (bond->params.use_carrier)
 443                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 444
 445         /* Try to get link status using Ethtool first. */
 446         if (slave_dev->ethtool_ops->get_link)
 447                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 448                         BMSR_LSTATUS : 0;
 449
 450         /* Ethtool can't be used, fallback to MII ioctls. */
 451         ioctl = slave_ops->ndo_do_ioctl;
 452         if (ioctl) {
 453                 /* TODO: set pointer to correct ioctl on a per team member
 454                  *       bases to make this more efficient. that is, once
 455                  *       we determine the correct ioctl, we will always
 456                  *       call it and not the others for that team
 457                  *       member.
 458                  */
 459
 460                 /* We cannot assume that SIOCGMIIPHY will also read a
 461                  * register; not all network drivers (e.g., e100)
 462                  * support that.
 463                  */
 464
 465                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 466                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 467                 mii = if_mii(&ifr);
 468                 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 469                         mii->reg_num = MII_BMSR;
 470                         if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
 471                                 return mii->val_out & BMSR_LSTATUS;
 472                 }
 473         }
 474
 475         /* If reporting, report that either there's no dev->do_ioctl,
 476          * or both SIOCGMIIREG and get_link failed (meaning that we
 477          * cannot report link status).  If not reporting, pretend
 478          * we're ok.
 479          */
 480         return reporting ? -1 : BMSR_LSTATUS;
 481 }
 482
 483 /*----------------------------- Multicast list ------------------------------*/
 484
 485 /* Push the promiscuity flag down to appropriate slaves */
 486 static int bond_set_promiscuity(struct bonding *bond, int inc)
 487 {
 488         struct list_head *iter;
 489         int err = 0;
 490
 491         if (bond_uses_primary(bond)) {
 492                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 493
 494                 if (curr_active)
 495                         err = dev_set_promiscuity(curr_active->dev, inc);
 496         } else {
 497                 struct slave *slave;
 498
 499                 bond_for_each_slave(bond, slave, iter) {
 500                         err = dev_set_promiscuity(slave->dev, inc);
 501                         if (err)
 502                                 return err;
 503                 }
 504         }
 505         return err;
 506 }
 507
 508 /* Push the allmulti flag down to all slaves */
 509 static int bond_set_allmulti(struct bonding *bond, int inc)
 510 {
 511         struct list_head *iter;
 512         int err = 0;
 513
 514         if (bond_uses_primary(bond)) {
 515                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 516
 517                 if (curr_active)
 518                         err = dev_set_allmulti(curr_active->dev, inc);
 519         } else {
 520                 struct slave *slave;
 521
 522                 bond_for_each_slave(bond, slave, iter) {
 523                         err = dev_set_allmulti(slave->dev, inc);
 524                         if (err)
 525                                 return err;
 526                 }
 527         }
 528         return err;
 529 }
 530
 531 /* Retrieve the list of registered multicast addresses for the bonding
 532  * device and retransmit an IGMP JOIN request to the current active
 533  * slave.
 534  */
 535 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 536 {
 537         struct bonding *bond = container_of(work, struct bonding,
 538                                             mcast_work.work);
 539
 540         if (!rtnl_trylock()) {
 541                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 542                 return;
 543         }
 544         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 545
 546         if (bond->igmp_retrans > 1) {
 547                 bond->igmp_retrans--;
 548                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 549         }
 550         rtnl_unlock();
 551 }
 552
 553 /* Flush bond's hardware addresses from slave */
 554 static void bond_hw_addr_flush(struct net_device *bond_dev,
 555                                struct net_device *slave_dev)
 556 {
 557         struct bonding *bond = netdev_priv(bond_dev);
 558
 559         dev_uc_unsync(slave_dev, bond_dev);
 560         dev_mc_unsync(slave_dev, bond_dev);
 561
 562         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 563                 /* del lacpdu mc addr from mc list */
 564                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 565
 566                 dev_mc_del(slave_dev, lacpdu_multicast);
 567         }
 568 }
 569
 570 /*--------------------------- Active slave change ---------------------------*/
 571
 572 /* Update the hardware address list and promisc/allmulti for the new and
 573  * old active slaves (if any).  Modes that are not using primary keep all
 574  * slaves up date at all times; only the modes that use primary need to call
 575  * this function to swap these settings during a failover.
 576  */
 577 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 578                               struct slave *old_active)
 579 {
 580         if (old_active) {
 581                 if (bond->dev->flags & IFF_PROMISC)
 582                         dev_set_promiscuity(old_active->dev, -1);
 583
 584                 if (bond->dev->flags & IFF_ALLMULTI)
 585                         dev_set_allmulti(old_active->dev, -1);
 586
 587                 bond_hw_addr_flush(bond->dev, old_active->dev);
 588         }
 589
 590         if (new_active) {
 591                 /* FIXME: Signal errors upstream. */
 592                 if (bond->dev->flags & IFF_PROMISC)
 593                         dev_set_promiscuity(new_active->dev, 1);
 594
 595                 if (bond->dev->flags & IFF_ALLMULTI)
 596                         dev_set_allmulti(new_active->dev, 1);
 597
 598                 netif_addr_lock_bh(bond->dev);
 599                 dev_uc_sync(new_active->dev, bond->dev);
 600                 dev_mc_sync(new_active->dev, bond->dev);
 601                 netif_addr_unlock_bh(bond->dev);
 602         }
 603 }
 604
 605 /**
 606  * bond_set_dev_addr - clone slave's address to bond
 607  * @bond_dev: bond net device
 608  * @slave_dev: slave net device
 609  *
 610  * Should be called with RTNL held.
 611  */
 612 static void bond_set_dev_addr(struct net_device *bond_dev,
 613                               struct net_device *slave_dev)
 614 {
 615         netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n",
 616                    bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len);
 617         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 618         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 619         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 620 }
 621
 622 static struct slave *bond_get_old_active(struct bonding *bond,
 623                                          struct slave *new_active)
 624 {
 625         struct slave *slave;
 626         struct list_head *iter;
 627
 628         bond_for_each_slave(bond, slave, iter) {
 629                 if (slave == new_active)
 630                         continue;
 631
 632                 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
 633                         return slave;
 634         }
 635
 636         return NULL;
 637 }
 638
 639 /* bond_do_fail_over_mac
 640  *
 641  * Perform special MAC address swapping for fail_over_mac settings
 642  *
 643  * Called with RTNL
 644  */
 645 static void bond_do_fail_over_mac(struct bonding *bond,
 646                                   struct slave *new_active,
 647                                   struct slave *old_active)
 648 {
 649         u8 tmp_mac[MAX_ADDR_LEN];
 650         struct sockaddr_storage ss;
 651         int rv;
 652
 653         switch (bond->params.fail_over_mac) {
 654         case BOND_FOM_ACTIVE:
 655                 if (new_active)
 656                         bond_set_dev_addr(bond->dev, new_active->dev);
 657                 break;
 658         case BOND_FOM_FOLLOW:
 659                 /* if new_active && old_active, swap them
 660                  * if just old_active, do nothing (going to no active slave)
 661                  * if just new_active, set new_active to bond's MAC
 662                  */
 663                 if (!new_active)
 664                         return;
 665
 666                 if (!old_active)
 667                         old_active = bond_get_old_active(bond, new_active);
 668
 669                 if (old_active) {
 670                         bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
 671                                           new_active->dev->addr_len);
 672                         bond_hw_addr_copy(ss.__data,
 673                                           old_active->dev->dev_addr,
 674                                           old_active->dev->addr_len);
 675                         ss.ss_family = new_active->dev->type;
 676                 } else {
 677                         bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
 678                                           bond->dev->addr_len);
 679                         ss.ss_family = bond->dev->type;
 680                 }
 681
 682                 rv = dev_set_mac_address(new_active->dev,
 683                                          (struct sockaddr *)&ss);
 684                 if (rv) {
 685                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 686                                    -rv, new_active->dev->name);
 687                         goto out;
 688                 }
 689
 690                 if (!old_active)
 691                         goto out;
 692
 693                 bond_hw_addr_copy(ss.__data, tmp_mac,
 694                                   new_active->dev->addr_len);
 695                 ss.ss_family = old_active->dev->type;
 696
 697                 rv = dev_set_mac_address(old_active->dev,
 698                                          (struct sockaddr *)&ss);
 699                 if (rv)
 700                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 701                                    -rv, new_active->dev->name);
 702 out:
 703                 break;
 704         default:
 705                 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
 706                            bond->params.fail_over_mac);
 707                 break;
 708         }
 709
 710 }
 711
 712 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
 713 {
 714         struct slave *prim = rtnl_dereference(bond->primary_slave);
 715         struct slave *curr = rtnl_dereference(bond->curr_active_slave);
 716
 717         if (!prim || prim->link != BOND_LINK_UP) {
 718                 if (!curr || curr->link != BOND_LINK_UP)
 719                         return NULL;
 720                 return curr;
 721         }
 722
 723         if (bond->force_primary) {
 724                 bond->force_primary = false;
 725                 return prim;
 726         }
 727
 728         if (!curr || curr->link != BOND_LINK_UP)
 729                 return prim;
 730
 731         /* At this point, prim and curr are both up */
 732         switch (bond->params.primary_reselect) {
 733         case BOND_PRI_RESELECT_ALWAYS:
 734                 return prim;
 735         case BOND_PRI_RESELECT_BETTER:
 736                 if (prim->speed < curr->speed)
 737                         return curr;
 738                 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
 739                         return curr;
 740                 return prim;
 741         case BOND_PRI_RESELECT_FAILURE:
 742                 return curr;
 743         default:
 744                 netdev_err(bond->dev, "impossible primary_reselect %d\n",
 745                            bond->params.primary_reselect);
 746                 return curr;
 747         }
 748 }
 749
 750 /**
 751  * bond_find_best_slave - select the best available slave to be the active one
 752  * @bond: our bonding struct
 753  */
 754 static struct slave *bond_find_best_slave(struct bonding *bond)
 755 {
 756         struct slave *slave, *bestslave = NULL;
 757         struct list_head *iter;
 758         int mintime = bond->params.updelay;
 759
 760         slave = bond_choose_primary_or_current(bond);
 761         if (slave)
 762                 return slave;
 763
 764         bond_for_each_slave(bond, slave, iter) {
 765                 if (slave->link == BOND_LINK_UP)
 766                         return slave;
 767                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
 768                     slave->delay < mintime) {
 769                         mintime = slave->delay;
 770                         bestslave = slave;
 771                 }
 772         }
 773
 774         return bestslave;
 775 }
 776
 777 static bool bond_should_notify_peers(struct bonding *bond)
 778 {
 779         struct slave *slave;
 780
 781         rcu_read_lock();
 782         slave = rcu_dereference(bond->curr_active_slave);
 783         rcu_read_unlock();
 784
 785         if (!slave || !bond->send_peer_notif ||
 786             !netif_carrier_ok(bond->dev) ||
 787             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 788                 return false;
 789
 790         netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
 791                    slave ? slave->dev->name : "NULL");
 792
 793         return true;
 794 }
 795
 796 /**
 797  * change_active_interface - change the active slave into the specified one
 798  * @bond: our bonding struct
 799  * @new: the new slave to make the active one
 800  *
 801  * Set the new slave to the bond's settings and unset them on the old
 802  * curr_active_slave.
 803  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 804  *
 805  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 806  * because it is apparently the best available slave we have, even though its
 807  * updelay hasn't timed out yet.
 808  *
 809  * Caller must hold RTNL.
 810  */
 811 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 812 {
 813         struct slave *old_active;
 814
 815         ASSERT_RTNL();
 816
 817         old_active = rtnl_dereference(bond->curr_active_slave);
 818
 819         if (old_active == new_active)
 820                 return;
 821
 822         if (new_active) {
 823                 new_active->last_link_up = jiffies;
 824
 825                 if (new_active->link == BOND_LINK_BACK) {
 826                         if (bond_uses_primary(bond)) {
 827                                 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
 828                                             new_active->dev->name,
 829                                             (bond->params.updelay - new_active->delay) * bond->params.miimon);
 830                         }
 831
 832                         new_active->delay = 0;
 833                         bond_set_slave_link_state(new_active, BOND_LINK_UP,
 834                                                   BOND_SLAVE_NOTIFY_NOW);
 835
 836                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 837                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 838
 839                         if (bond_is_lb(bond))
 840                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 841                 } else {
 842                         if (bond_uses_primary(bond)) {
 843                                 netdev_info(bond->dev, "making interface %s the new active one\n",
 844                                             new_active->dev->name);
 845                         }
 846                 }
 847         }
 848
 849         if (bond_uses_primary(bond))
 850                 bond_hw_addr_swap(bond, new_active, old_active);
 851
 852         if (bond_is_lb(bond)) {
 853                 bond_alb_handle_active_change(bond, new_active);
 854                 if (old_active)
 855                         bond_set_slave_inactive_flags(old_active,
 856                                                       BOND_SLAVE_NOTIFY_NOW);
 857                 if (new_active)
 858                         bond_set_slave_active_flags(new_active,
 859                                                     BOND_SLAVE_NOTIFY_NOW);
 860         } else {
 861                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 862         }
 863
 864         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
 865                 if (old_active)
 866                         bond_set_slave_inactive_flags(old_active,
 867                                                       BOND_SLAVE_NOTIFY_NOW);
 868
 869                 if (new_active) {
 870                         bool should_notify_peers = false;
 871
 872                         bond_set_slave_active_flags(new_active,
 873                                                     BOND_SLAVE_NOTIFY_NOW);
 874
 875                         if (bond->params.fail_over_mac)
 876                                 bond_do_fail_over_mac(bond, new_active,
 877                                                       old_active);
 878
 879                         if (netif_running(bond->dev)) {
 880                                 bond->send_peer_notif =
 881                                         bond->params.num_peer_notif;
 882                                 should_notify_peers =
 883                                         bond_should_notify_peers(bond);
 884                         }
 885
 886                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 887                         if (should_notify_peers)
 888                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 889                                                          bond->dev);
 890                 }
 891         }
 892
 893         /* resend IGMP joins since active slave has changed or
 894          * all were sent on curr_active_slave.
 895          * resend only if bond is brought up with the affected
 896          * bonding modes and the retransmission is enabled
 897          */
 898         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 899             ((bond_uses_primary(bond) && new_active) ||
 900              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
 901                 bond->igmp_retrans = bond->params.resend_igmp;
 902                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 903         }
 904 }
 905
 906 /**
 907  * bond_select_active_slave - select a new active slave, if needed
 908  * @bond: our bonding struct
 909  *
 910  * This functions should be called when one of the following occurs:
 911  * - The old curr_active_slave has been released or lost its link.
 912  * - The primary_slave has got its link back.
 913  * - A slave has got its link back and there's no old curr_active_slave.
 914  *
 915  * Caller must hold RTNL.
 916  */
 917 void bond_select_active_slave(struct bonding *bond)
 918 {
 919         struct slave *best_slave;
 920         int rv;
 921
 922         ASSERT_RTNL();
 923
 924         best_slave = bond_find_best_slave(bond);
 925         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
 926                 bond_change_active_slave(bond, best_slave);
 927                 rv = bond_set_carrier(bond);
 928                 if (!rv)
 929                         return;
 930
 931                 if (netif_carrier_ok(bond->dev))
 932                         netdev_info(bond->dev, "first active interface up!\n");
 933                 else
 934                         netdev_info(bond->dev, "now running without any active interface!\n");
 935         }
 936 }
 937
 938 #ifdef CONFIG_NET_POLL_CONTROLLER
 939 static inline int slave_enable_netpoll(struct slave *slave)
 940 {
 941         struct netpoll *np;
 942         int err = 0;
 943
 944         np = kzalloc(sizeof(*np), GFP_KERNEL);
 945         err = -ENOMEM;
 946         if (!np)
 947                 goto out;
 948
 949         err = __netpoll_setup(np, slave->dev);
 950         if (err) {
 951                 kfree(np);
 952                 goto out;
 953         }
 954         slave->np = np;
 955 out:
 956         return err;
 957 }
 958 static inline void slave_disable_netpoll(struct slave *slave)
 959 {
 960         struct netpoll *np = slave->np;
 961
 962         if (!np)
 963                 return;
 964
 965         slave->np = NULL;
 966         __netpoll_free_async(np);
 967 }
 968
 969 static void bond_poll_controller(struct net_device *bond_dev)
 970 {
 971         struct bonding *bond = netdev_priv(bond_dev);
 972         struct slave *slave = NULL;
 973         struct list_head *iter;
 974         struct ad_info ad_info;
 975
 976         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 977                 if (bond_3ad_get_active_agg_info(bond, &ad_info))
 978                         return;
 979
 980         bond_for_each_slave_rcu(bond, slave, iter) {
 981                 if (!bond_slave_is_up(slave))
 982                         continue;
 983
 984                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 985                         struct aggregator *agg =
 986                             SLAVE_AD_INFO(slave)->port.aggregator;
 987
 988                         if (agg &&
 989                             agg->aggregator_identifier != ad_info.aggregator_id)
 990                                 continue;
 991                 }
 992
 993                 netpoll_poll_dev(slave->dev);
 994         }
 995 }
 996
 997 static void bond_netpoll_cleanup(struct net_device *bond_dev)
 998 {
 999         struct bonding *bond = netdev_priv(bond_dev);
1000         struct list_head *iter;
1001         struct slave *slave;
1002
1003         bond_for_each_slave(bond, slave, iter)
1004                 if (bond_slave_is_up(slave))
1005                         slave_disable_netpoll(slave);
1006 }
1007
1008 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1009 {
1010         struct bonding *bond = netdev_priv(dev);
1011         struct list_head *iter;
1012         struct slave *slave;
1013         int err = 0;
1014
1015         bond_for_each_slave(bond, slave, iter) {
1016                 err = slave_enable_netpoll(slave);
1017                 if (err) {
1018                         bond_netpoll_cleanup(dev);
1019                         break;
1020                 }
1021         }
1022         return err;
1023 }
1024 #else
1025 static inline int slave_enable_netpoll(struct slave *slave)
1026 {
1027         return 0;
1028 }
1029 static inline void slave_disable_netpoll(struct slave *slave)
1030 {
1031 }
1032 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1033 {
1034 }
1035 #endif
1036
1037 /*---------------------------------- IOCTL ----------------------------------*/
1038
1039 static netdev_features_t bond_fix_features(struct net_device *dev,
1040                                            netdev_features_t features)
1041 {
1042         struct bonding *bond = netdev_priv(dev);
1043         struct list_head *iter;
1044         netdev_features_t mask;
1045         struct slave *slave;
1046
1047         mask = features;
1048
1049         features &= ~NETIF_F_ONE_FOR_ALL;
1050         features |= NETIF_F_ALL_FOR_ALL;
1051
1052         bond_for_each_slave(bond, slave, iter) {
1053                 features = netdev_increment_features(features,
1054                                                      slave->dev->features,
1055                                                      mask);
1056         }
1057         features = netdev_add_tso_features(features, mask);
1058
1059         return features;
1060 }
1061
1062 #define BOND_VLAN_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1063                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1064                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1065
1066 #define BOND_ENC_FEATURES       (NETIF_F_HW_CSUM | NETIF_F_SG | \
1067                                  NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1068
1069 static void bond_compute_features(struct bonding *bond)
1070 {
1071         unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1072                                         IFF_XMIT_DST_RELEASE_PERM;
1073         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1074         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1075         struct net_device *bond_dev = bond->dev;
1076         struct list_head *iter;
1077         struct slave *slave;
1078         unsigned short max_hard_header_len = ETH_HLEN;
1079         unsigned int gso_max_size = GSO_MAX_SIZE;
1080         u16 gso_max_segs = GSO_MAX_SEGS;
1081
1082         if (!bond_has_slaves(bond))
1083                 goto done;
1084         vlan_features &= NETIF_F_ALL_FOR_ALL;
1085
1086         bond_for_each_slave(bond, slave, iter) {
1087                 vlan_features = netdev_increment_features(vlan_features,
1088                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1089
1090                 enc_features = netdev_increment_features(enc_features,
1091                                                          slave->dev->hw_enc_features,
1092                                                          BOND_ENC_FEATURES);
1093                 dst_release_flag &= slave->dev->priv_flags;
1094                 if (slave->dev->hard_header_len > max_hard_header_len)
1095                         max_hard_header_len = slave->dev->hard_header_len;
1096
1097                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1098                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1099         }
1100         bond_dev->hard_header_len = max_hard_header_len;
1101
1102 done:
1103         bond_dev->vlan_features = vlan_features;
1104         bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1105                                     NETIF_F_HW_VLAN_CTAG_TX |
1106                                     NETIF_F_HW_VLAN_STAG_TX |
1107                                     NETIF_F_GSO_UDP_L4;
1108         bond_dev->gso_max_segs = gso_max_segs;
1109         netif_set_gso_max_size(bond_dev, gso_max_size);
1110
1111         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1112         if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1113             dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1114                 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1115
1116         netdev_change_features(bond_dev);
1117 }
1118
1119 static void bond_setup_by_slave(struct net_device *bond_dev,
1120                                 struct net_device *slave_dev)
1121 {
1122         bond_dev->header_ops        = slave_dev->header_ops;
1123
1124         bond_dev->type              = slave_dev->type;
1125         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1126         bond_dev->needed_headroom   = slave_dev->needed_headroom;
1127         bond_dev->addr_len          = slave_dev->addr_len;
1128
1129         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1130                 slave_dev->addr_len);
1131 }
1132
1133 /* On bonding slaves other than the currently active slave, suppress
1134  * duplicates except for alb non-mcast/bcast.
1135  */
1136 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1137                                             struct slave *slave,
1138                                             struct bonding *bond)
1139 {
1140         if (bond_is_slave_inactive(slave)) {
1141                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1142                     skb->pkt_type != PACKET_BROADCAST &&
1143                     skb->pkt_type != PACKET_MULTICAST)
1144                         return false;
1145                 return true;
1146         }
1147         return false;
1148 }
1149
1150 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1151 {
1152         struct sk_buff *skb = *pskb;
1153         struct slave *slave;
1154         struct bonding *bond;
1155         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1156                           struct slave *);
1157         int ret = RX_HANDLER_ANOTHER;
1158
1159         skb = skb_share_check(skb, GFP_ATOMIC);
1160         if (unlikely(!skb))
1161                 return RX_HANDLER_CONSUMED;
1162
1163         *pskb = skb;
1164
1165         slave = bond_slave_get_rcu(skb->dev);
1166         bond = slave->bond;
1167
1168         recv_probe = READ_ONCE(bond->recv_probe);
1169         if (recv_probe) {
1170                 ret = recv_probe(skb, bond, slave);
1171                 if (ret == RX_HANDLER_CONSUMED) {
1172                         consume_skb(skb);
1173                         return ret;
1174                 }
1175         }
1176
1177         /*
1178          * For packets determined by bond_should_deliver_exact_match() call to
1179          * be suppressed we want to make an exception for link-local packets.
1180          * This is necessary for e.g. LLDP daemons to be able to monitor
1181          * inactive slave links without being forced to bind to them
1182          * explicitly.
1183          *
1184          * At the same time, packets that are passed to the bonding master
1185          * (including link-local ones) can have their originating interface
1186          * determined via PACKET_ORIGDEV socket option.
1187          */
1188         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1189                 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1190                         return RX_HANDLER_PASS;
1191                 return RX_HANDLER_EXACT;
1192         }
1193
1194         skb->dev = bond->dev;
1195
1196         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1197             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1198             skb->pkt_type == PACKET_HOST) {
1199
1200                 if (unlikely(skb_cow_head(skb,
1201                                           skb->data - skb_mac_header(skb)))) {
1202                         kfree_skb(skb);
1203                         return RX_HANDLER_CONSUMED;
1204                 }
1205                 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1206                                   bond->dev->addr_len);
1207         }
1208
1209         return ret;
1210 }
1211
1212 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1213 {
1214         switch (BOND_MODE(bond)) {
1215         case BOND_MODE_ROUNDROBIN:
1216                 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1217         case BOND_MODE_ACTIVEBACKUP:
1218                 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1219         case BOND_MODE_BROADCAST:
1220                 return NETDEV_LAG_TX_TYPE_BROADCAST;
1221         case BOND_MODE_XOR:
1222         case BOND_MODE_8023AD:
1223                 return NETDEV_LAG_TX_TYPE_HASH;
1224         default:
1225                 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1226         }
1227 }
1228
1229 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1230                                                enum netdev_lag_tx_type type)
1231 {
1232         if (type != NETDEV_LAG_TX_TYPE_HASH)
1233                 return NETDEV_LAG_HASH_NONE;
1234
1235         switch (bond->params.xmit_policy) {
1236         case BOND_XMIT_POLICY_LAYER2:
1237                 return NETDEV_LAG_HASH_L2;
1238         case BOND_XMIT_POLICY_LAYER34:
1239                 return NETDEV_LAG_HASH_L34;
1240         case BOND_XMIT_POLICY_LAYER23:
1241                 return NETDEV_LAG_HASH_L23;
1242         case BOND_XMIT_POLICY_ENCAP23:
1243                 return NETDEV_LAG_HASH_E23;
1244         case BOND_XMIT_POLICY_ENCAP34:
1245                 return NETDEV_LAG_HASH_E34;
1246         default:
1247                 return NETDEV_LAG_HASH_UNKNOWN;
1248         }
1249 }
1250
1251 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1252                                       struct netlink_ext_ack *extack)
1253 {
1254         struct netdev_lag_upper_info lag_upper_info;
1255         enum netdev_lag_tx_type type;
1256
1257         type = bond_lag_tx_type(bond);
1258         lag_upper_info.tx_type = type;
1259         lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1260
1261         return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1262                                             &lag_upper_info, extack);
1263 }
1264
1265 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1266 {
1267         netdev_upper_dev_unlink(slave->dev, bond->dev);
1268         slave->dev->flags &= ~IFF_SLAVE;
1269 }
1270
1271 static void slave_kobj_release(struct kobject *kobj)
1272 {
1273         struct slave *slave = to_slave(kobj);
1274         struct bonding *bond = bond_get_bond_by_slave(slave);
1275
1276         cancel_delayed_work_sync(&slave->notify_work);
1277         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1278                 kfree(SLAVE_AD_INFO(slave));
1279
1280         kfree(slave);
1281 }
1282
1283 static struct kobj_type slave_ktype = {
1284         .release = slave_kobj_release,
1285 #ifdef CONFIG_SYSFS
1286         .sysfs_ops = &slave_sysfs_ops,
1287 #endif
1288 };
1289
1290 static int bond_kobj_init(struct slave *slave)
1291 {
1292         int err;
1293
1294         err = kobject_init_and_add(&slave->kobj, &slave_ktype,
1295                                    &(slave->dev->dev.kobj), "bonding_slave");
1296         if (err)
1297                 kobject_put(&slave->kobj);
1298
1299         return err;
1300 }
1301
1302 static struct slave *bond_alloc_slave(struct bonding *bond,
1303                                       struct net_device *slave_dev)
1304 {
1305         struct slave *slave = NULL;
1306
1307         slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1308         if (!slave)
1309                 return NULL;
1310
1311         slave->bond = bond;
1312         slave->dev = slave_dev;
1313         INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1314
1315         if (bond_kobj_init(slave))
1316                 return NULL;
1317
1318         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1319                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1320                                                GFP_KERNEL);
1321                 if (!SLAVE_AD_INFO(slave)) {
1322                         kobject_put(&slave->kobj);
1323                         return NULL;
1324                 }
1325         }
1326
1327         return slave;
1328 }
1329
1330 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1331 {
1332         info->bond_mode = BOND_MODE(bond);
1333         info->miimon = bond->params.miimon;
1334         info->num_slaves = bond->slave_cnt;
1335 }
1336
1337 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1338 {
1339         strcpy(info->slave_name, slave->dev->name);
1340         info->link = slave->link;
1341         info->state = bond_slave_state(slave);
1342         info->link_failure_count = slave->link_failure_count;
1343 }
1344
1345 static void bond_netdev_notify_work(struct work_struct *_work)
1346 {
1347         struct slave *slave = container_of(_work, struct slave,
1348                                            notify_work.work);
1349
1350         if (rtnl_trylock()) {
1351                 struct netdev_bonding_info binfo;
1352
1353                 bond_fill_ifslave(slave, &binfo.slave);
1354                 bond_fill_ifbond(slave->bond, &binfo.master);
1355                 netdev_bonding_info_change(slave->dev, &binfo);
1356                 rtnl_unlock();
1357         } else {
1358                 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1359         }
1360 }
1361
1362 void bond_queue_slave_event(struct slave *slave)
1363 {
1364         queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1365 }
1366
1367 void bond_lower_state_changed(struct slave *slave)
1368 {
1369         struct netdev_lag_lower_state_info info;
1370
1371         info.link_up = slave->link == BOND_LINK_UP ||
1372                        slave->link == BOND_LINK_FAIL;
1373         info.tx_enabled = bond_is_active_slave(slave);
1374         netdev_lower_state_changed(slave->dev, &info);
1375 }
1376
1377 /* enslave device <slave> to bond device <master> */
1378 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1379                  struct netlink_ext_ack *extack)
1380 {
1381         struct bonding *bond = netdev_priv(bond_dev);
1382         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1383         struct slave *new_slave = NULL, *prev_slave;
1384         struct sockaddr_storage ss;
1385         int link_reporting;
1386         int res = 0, i;
1387
1388         if (!bond->params.use_carrier &&
1389             slave_dev->ethtool_ops->get_link == NULL &&
1390             slave_ops->ndo_do_ioctl == NULL) {
1391                 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1392                             slave_dev->name);
1393         }
1394
1395         /* already in-use? */
1396         if (netdev_is_rx_handler_busy(slave_dev)) {
1397                 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1398                 netdev_err(bond_dev,
1399                            "Error: Device is in use and cannot be enslaved\n");
1400                 return -EBUSY;
1401         }
1402
1403         if (bond_dev == slave_dev) {
1404                 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1405                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1406                 return -EPERM;
1407         }
1408
1409         /* vlan challenged mutual exclusion */
1410         /* no need to lock since we're protected by rtnl_lock */
1411         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1412                 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1413                            slave_dev->name);
1414                 if (vlan_uses_dev(bond_dev)) {
1415                         NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1416                         netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1417                                    slave_dev->name, bond_dev->name);
1418                         return -EPERM;
1419                 } else {
1420                         netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1421                                     slave_dev->name, slave_dev->name,
1422                                     bond_dev->name);
1423                 }
1424         } else {
1425                 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1426                            slave_dev->name);
1427         }
1428
1429         /* Old ifenslave binaries are no longer supported.  These can
1430          * be identified with moderate accuracy by the state of the slave:
1431          * the current ifenslave will set the interface down prior to
1432          * enslaving it; the old ifenslave will not.
1433          */
1434         if (slave_dev->flags & IFF_UP) {
1435                 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
1436                 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1437                            slave_dev->name);
1438                 return -EPERM;
1439         }
1440
1441         /* set bonding device ether type by slave - bonding netdevices are
1442          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1443          * there is a need to override some of the type dependent attribs/funcs.
1444          *
1445          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1446          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1447          */
1448         if (!bond_has_slaves(bond)) {
1449                 if (bond_dev->type != slave_dev->type) {
1450                         netdev_dbg(bond_dev, "change device type from %d to %d\n",
1451                                    bond_dev->type, slave_dev->type);
1452
1453                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1454                                                        bond_dev);
1455                         res = notifier_to_errno(res);
1456                         if (res) {
1457                                 netdev_err(bond_dev, "refused to change device type\n");
1458                                 return -EBUSY;
1459                         }
1460
1461                         /* Flush unicast and multicast addresses */
1462                         dev_uc_flush(bond_dev);
1463                         dev_mc_flush(bond_dev);
1464
1465                         if (slave_dev->type != ARPHRD_ETHER)
1466                                 bond_setup_by_slave(bond_dev, slave_dev);
1467                         else {
1468                                 ether_setup(bond_dev);
1469                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1470                         }
1471
1472                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1473                                                  bond_dev);
1474                 }
1475         } else if (bond_dev->type != slave_dev->type) {
1476                 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1477                 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1478                            slave_dev->name, slave_dev->type, bond_dev->type);
1479                 return -EINVAL;
1480         }
1481
1482         if (slave_dev->type == ARPHRD_INFINIBAND &&
1483             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1484                 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1485                 netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n",
1486                             slave_dev->type);
1487                 res = -EOPNOTSUPP;
1488                 goto err_undo_flags;
1489         }
1490
1491         if (!slave_ops->ndo_set_mac_address ||
1492             slave_dev->type == ARPHRD_INFINIBAND) {
1493                 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1494                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1495                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1496                         if (!bond_has_slaves(bond)) {
1497                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1498                                 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1499                         } else {
1500                                 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1501                                 netdev_err(bond_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
1502                                 res = -EOPNOTSUPP;
1503                                 goto err_undo_flags;
1504                         }
1505                 }
1506         }
1507
1508         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1509
1510         /* If this is the first slave, then we need to set the master's hardware
1511          * address to be the same as the slave's.
1512          */
1513         if (!bond_has_slaves(bond) &&
1514             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1515                 bond_set_dev_addr(bond->dev, slave_dev);
1516
1517         new_slave = bond_alloc_slave(bond, slave_dev);
1518         if (!new_slave) {
1519                 res = -ENOMEM;
1520                 goto err_undo_flags;
1521         }
1522
1523         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1524          * is set via sysfs or module option if desired.
1525          */
1526         new_slave->queue_id = 0;
1527
1528         /* Save slave's original mtu and then set it to match the bond */
1529         new_slave->original_mtu = slave_dev->mtu;
1530         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1531         if (res) {
1532                 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1533                 goto err_free;
1534         }
1535
1536         /* Save slave's original ("permanent") mac address for modes
1537          * that need it, and for restoring it upon release, and then
1538          * set it to the master's address
1539          */
1540         bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1541                           slave_dev->addr_len);
1542
1543         if (!bond->params.fail_over_mac ||
1544             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1545                 /* Set slave to master's mac address.  The application already
1546                  * set the master's mac address to that of the first slave
1547                  */
1548                 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1549                 ss.ss_family = slave_dev->type;
1550                 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1551                 if (res) {
1552                         netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1553                         goto err_restore_mtu;
1554                 }
1555         }
1556
1557         /* set slave flag before open to prevent IPv6 addrconf */
1558         slave_dev->flags |= IFF_SLAVE;
1559
1560         /* open the slave since the application closed it */
1561         res = dev_open(slave_dev);
1562         if (res) {
1563                 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1564                 goto err_restore_mac;
1565         }
1566
1567         slave_dev->priv_flags |= IFF_BONDING;
1568         /* initialize slave stats */
1569         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1570
1571         if (bond_is_lb(bond)) {
1572                 /* bond_alb_init_slave() must be called before all other stages since
1573                  * it might fail and we do not want to have to undo everything
1574                  */
1575                 res = bond_alb_init_slave(bond, new_slave);
1576                 if (res)
1577                         goto err_close;
1578         }
1579
1580         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1581         if (res) {
1582                 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1583                            slave_dev->name);
1584                 goto err_close;
1585         }
1586
1587         prev_slave = bond_last_slave(bond);
1588
1589         new_slave->delay = 0;
1590         new_slave->link_failure_count = 0;
1591
1592         if (bond_update_speed_duplex(new_slave) &&
1593             bond_needs_speed_duplex(bond))
1594                 new_slave->link = BOND_LINK_DOWN;
1595
1596         new_slave->last_rx = jiffies -
1597                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1598         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1599                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1600
1601         if (bond->params.miimon && !bond->params.use_carrier) {
1602                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1603
1604                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1605                         /* miimon is set but a bonded network driver
1606                          * does not support ETHTOOL/MII and
1607                          * arp_interval is not set.  Note: if
1608                          * use_carrier is enabled, we will never go
1609                          * here (because netif_carrier is always
1610                          * supported); thus, we don't need to change
1611                          * the messages for netif_carrier.
1612                          */
1613                         netdev_warn(bond_dev, "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",
1614                                     slave_dev->name);
1615                 } else if (link_reporting == -1) {
1616                         /* unable get link status using mii/ethtool */
1617                         netdev_warn(bond_dev, "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",
1618                                     slave_dev->name);
1619                 }
1620         }
1621
1622         /* check for initial state */
1623         new_slave->link = BOND_LINK_NOCHANGE;
1624         if (bond->params.miimon) {
1625                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1626                         if (bond->params.updelay) {
1627                                 bond_set_slave_link_state(new_slave,
1628                                                           BOND_LINK_BACK,
1629                                                           BOND_SLAVE_NOTIFY_NOW);
1630                                 new_slave->delay = bond->params.updelay;
1631                         } else {
1632                                 bond_set_slave_link_state(new_slave,
1633                                                           BOND_LINK_UP,
1634                                                           BOND_SLAVE_NOTIFY_NOW);
1635                         }
1636                 } else {
1637                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1638                                                   BOND_SLAVE_NOTIFY_NOW);
1639                 }
1640         } else if (bond->params.arp_interval) {
1641                 bond_set_slave_link_state(new_slave,
1642                                           (netif_carrier_ok(slave_dev) ?
1643                                           BOND_LINK_UP : BOND_LINK_DOWN),
1644                                           BOND_SLAVE_NOTIFY_NOW);
1645         } else {
1646                 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1647                                           BOND_SLAVE_NOTIFY_NOW);
1648         }
1649
1650         if (new_slave->link != BOND_LINK_DOWN)
1651                 new_slave->last_link_up = jiffies;
1652         netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1653                    new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1654                    (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1655
1656         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1657                 /* if there is a primary slave, remember it */
1658                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1659                         rcu_assign_pointer(bond->primary_slave, new_slave);
1660                         bond->force_primary = true;
1661                 }
1662         }
1663
1664         switch (BOND_MODE(bond)) {
1665         case BOND_MODE_ACTIVEBACKUP:
1666                 bond_set_slave_inactive_flags(new_slave,
1667                                               BOND_SLAVE_NOTIFY_NOW);
1668                 break;
1669         case BOND_MODE_8023AD:
1670                 /* in 802.3ad mode, the internal mechanism
1671                  * will activate the slaves in the selected
1672                  * aggregator
1673                  */
1674                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1675                 /* if this is the first slave */
1676                 if (!prev_slave) {
1677                         SLAVE_AD_INFO(new_slave)->id = 1;
1678                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1679                          * can be called only after the mac address of the bond is set
1680                          */
1681                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1682                 } else {
1683                         SLAVE_AD_INFO(new_slave)->id =
1684                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1685                 }
1686
1687                 bond_3ad_bind_slave(new_slave);
1688                 break;
1689         case BOND_MODE_TLB:
1690         case BOND_MODE_ALB:
1691                 bond_set_active_slave(new_slave);
1692                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1693                 break;
1694         default:
1695                 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1696
1697                 /* always active in trunk mode */
1698                 bond_set_active_slave(new_slave);
1699
1700                 /* In trunking mode there is little meaning to curr_active_slave
1701                  * anyway (it holds no special properties of the bond device),
1702                  * so we can change it without calling change_active_interface()
1703                  */
1704                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1705                     new_slave->link == BOND_LINK_UP)
1706                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1707
1708                 break;
1709         } /* switch(bond_mode) */
1710
1711 #ifdef CONFIG_NET_POLL_CONTROLLER
1712         if (bond->dev->npinfo) {
1713                 if (slave_enable_netpoll(new_slave)) {
1714                         netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1715                         res = -EBUSY;
1716                         goto err_detach;
1717                 }
1718         }
1719 #endif
1720
1721         if (!(bond_dev->features & NETIF_F_LRO))
1722                 dev_disable_lro(slave_dev);
1723
1724         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1725                                          new_slave);
1726         if (res) {
1727                 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1728                 goto err_detach;
1729         }
1730
1731         res = bond_master_upper_dev_link(bond, new_slave, extack);
1732         if (res) {
1733                 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1734                 goto err_unregister;
1735         }
1736
1737         res = bond_sysfs_slave_add(new_slave);
1738         if (res) {
1739                 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1740                 goto err_upper_unlink;
1741         }
1742
1743         bond->nest_level = dev_get_nest_level(bond_dev) + 1;
1744
1745         /* If the mode uses primary, then the following is handled by
1746          * bond_change_active_slave().
1747          */
1748         if (!bond_uses_primary(bond)) {
1749                 /* set promiscuity level to new slave */
1750                 if (bond_dev->flags & IFF_PROMISC) {
1751                         res = dev_set_promiscuity(slave_dev, 1);
1752                         if (res)
1753                                 goto err_sysfs_del;
1754                 }
1755
1756                 /* set allmulti level to new slave */
1757                 if (bond_dev->flags & IFF_ALLMULTI) {
1758                         res = dev_set_allmulti(slave_dev, 1);
1759                         if (res) {
1760                                 if (bond_dev->flags & IFF_PROMISC)
1761                                         dev_set_promiscuity(slave_dev, -1);
1762                                 goto err_sysfs_del;
1763                         }
1764                 }
1765
1766                 netif_addr_lock_bh(bond_dev);
1767                 dev_mc_sync_multiple(slave_dev, bond_dev);
1768                 dev_uc_sync_multiple(slave_dev, bond_dev);
1769                 netif_addr_unlock_bh(bond_dev);
1770
1771                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1772                         /* add lacpdu mc addr to mc list */
1773                         u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1774
1775                         dev_mc_add(slave_dev, lacpdu_multicast);
1776                 }
1777         }
1778
1779         bond->slave_cnt++;
1780         bond_compute_features(bond);
1781         bond_set_carrier(bond);
1782
1783         if (bond_uses_primary(bond)) {
1784                 block_netpoll_tx();
1785                 bond_select_active_slave(bond);
1786                 unblock_netpoll_tx();
1787         }
1788
1789         if (bond_mode_can_use_xmit_hash(bond))
1790                 bond_update_slave_arr(bond, NULL);
1791
1792
1793         netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1794                     slave_dev->name,
1795                     bond_is_active_slave(new_slave) ? "an active" : "a backup",
1796                     new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1797
1798         /* enslave is successful */
1799         bond_queue_slave_event(new_slave);
1800         return 0;
1801
1802 /* Undo stages on error */
1803 err_sysfs_del:
1804         bond_sysfs_slave_del(new_slave);
1805
1806 err_upper_unlink:
1807         bond_upper_dev_unlink(bond, new_slave);
1808
1809 err_unregister:
1810         netdev_rx_handler_unregister(slave_dev);
1811
1812 err_detach:
1813         vlan_vids_del_by_dev(slave_dev, bond_dev);
1814         if (rcu_access_pointer(bond->primary_slave) == new_slave)
1815                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1816         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1817                 block_netpoll_tx();
1818                 bond_change_active_slave(bond, NULL);
1819                 bond_select_active_slave(bond);
1820                 unblock_netpoll_tx();
1821         }
1822         /* either primary_slave or curr_active_slave might've changed */
1823         synchronize_rcu();
1824         slave_disable_netpoll(new_slave);
1825
1826 err_close:
1827         if (!netif_is_bond_master(slave_dev))
1828                 slave_dev->priv_flags &= ~IFF_BONDING;
1829         dev_close(slave_dev);
1830
1831 err_restore_mac:
1832         slave_dev->flags &= ~IFF_SLAVE;
1833         if (!bond->params.fail_over_mac ||
1834             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1835                 /* XXX TODO - fom follow mode needs to change master's
1836                  * MAC if this slave's MAC is in use by the bond, or at
1837                  * least print a warning.
1838                  */
1839                 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
1840                                   new_slave->dev->addr_len);
1841                 ss.ss_family = slave_dev->type;
1842                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1843         }
1844
1845 err_restore_mtu:
1846         dev_set_mtu(slave_dev, new_slave->original_mtu);
1847
1848 err_free:
1849         kobject_put(&new_slave->kobj);
1850
1851 err_undo_flags:
1852         /* Enslave of first slave has failed and we need to fix master's mac */
1853         if (!bond_has_slaves(bond)) {
1854                 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1855                                             slave_dev->dev_addr))
1856                         eth_hw_addr_random(bond_dev);
1857                 if (bond_dev->type != ARPHRD_ETHER) {
1858                         dev_close(bond_dev);
1859                         ether_setup(bond_dev);
1860                         bond_dev->flags |= IFF_MASTER;
1861                         bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1862                 }
1863         }
1864
1865         return res;
1866 }
1867
1868 /* Try to release the slave device <slave> from the bond device <master>
1869  * It is legal to access curr_active_slave without a lock because all the function
1870  * is RTNL-locked. If "all" is true it means that the function is being called
1871  * while destroying a bond interface and all slaves are being released.
1872  *
1873  * The rules for slave state should be:
1874  *   for Active/Backup:
1875  *     Active stays on all backups go down
1876  *   for Bonded connections:
1877  *     The first up interface should be left on and all others downed.
1878  */
1879 static int __bond_release_one(struct net_device *bond_dev,
1880                               struct net_device *slave_dev,
1881                               bool all, bool unregister)
1882 {
1883         struct bonding *bond = netdev_priv(bond_dev);
1884         struct slave *slave, *oldcurrent;
1885         struct sockaddr_storage ss;
1886         int old_flags = bond_dev->flags;
1887         netdev_features_t old_features = bond_dev->features;
1888
1889         /* slave is not a slave or master is not master of this slave */
1890         if (!(slave_dev->flags & IFF_SLAVE) ||
1891             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1892                 netdev_dbg(bond_dev, "cannot release %s\n",
1893                            slave_dev->name);
1894                 return -EINVAL;
1895         }
1896
1897         block_netpoll_tx();
1898
1899         slave = bond_get_slave_by_dev(bond, slave_dev);
1900         if (!slave) {
1901                 /* not a slave of this bond */
1902                 netdev_info(bond_dev, "%s not enslaved\n",
1903                             slave_dev->name);
1904                 unblock_netpoll_tx();
1905                 return -EINVAL;
1906         }
1907
1908         bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1909
1910         bond_sysfs_slave_del(slave);
1911
1912         /* recompute stats just before removing the slave */
1913         bond_get_stats(bond->dev, &bond->bond_stats);
1914
1915         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1916          * for this slave anymore.
1917          */
1918         netdev_rx_handler_unregister(slave_dev);
1919
1920         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1921                 bond_3ad_unbind_slave(slave);
1922
1923         bond_upper_dev_unlink(bond, slave);
1924
1925         if (bond_mode_can_use_xmit_hash(bond))
1926                 bond_update_slave_arr(bond, slave);
1927
1928         netdev_info(bond_dev, "Releasing %s interface %s\n",
1929                     bond_is_active_slave(slave) ? "active" : "backup",
1930                     slave_dev->name);
1931
1932         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1933
1934         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1935
1936         if (!all && (!bond->params.fail_over_mac ||
1937                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1938                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1939                     bond_has_slaves(bond))
1940                         netdev_warn(bond_dev, "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",
1941                                     slave_dev->name, slave->perm_hwaddr,
1942                                     bond_dev->name, slave_dev->name);
1943         }
1944
1945         if (rtnl_dereference(bond->primary_slave) == slave)
1946                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1947
1948         if (oldcurrent == slave)
1949                 bond_change_active_slave(bond, NULL);
1950
1951         if (bond_is_lb(bond)) {
1952                 /* Must be called only after the slave has been
1953                  * detached from the list and the curr_active_slave
1954                  * has been cleared (if our_slave == old_current),
1955                  * but before a new active slave is selected.
1956                  */
1957                 bond_alb_deinit_slave(bond, slave);
1958         }
1959
1960         if (all) {
1961                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1962         } else if (oldcurrent == slave) {
1963                 /* Note that we hold RTNL over this sequence, so there
1964                  * is no concern that another slave add/remove event
1965                  * will interfere.
1966                  */
1967                 bond_select_active_slave(bond);
1968         }
1969
1970         if (!bond_has_slaves(bond)) {
1971                 bond_set_carrier(bond);
1972                 eth_hw_addr_random(bond_dev);
1973                 bond->nest_level = SINGLE_DEPTH_NESTING;
1974         } else {
1975                 bond->nest_level = dev_get_nest_level(bond_dev) + 1;
1976         }
1977
1978         unblock_netpoll_tx();
1979         synchronize_rcu();
1980         bond->slave_cnt--;
1981
1982         if (!bond_has_slaves(bond)) {
1983                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1984                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1985         }
1986
1987         bond_compute_features(bond);
1988         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1989             (old_features & NETIF_F_VLAN_CHALLENGED))
1990                 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1991                             slave_dev->name, bond_dev->name);
1992
1993         vlan_vids_del_by_dev(slave_dev, bond_dev);
1994
1995         /* If the mode uses primary, then this case was handled above by
1996          * bond_change_active_slave(..., NULL)
1997          */
1998         if (!bond_uses_primary(bond)) {
1999                 /* unset promiscuity level from slave
2000                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
2001                  * of the IFF_PROMISC flag in the bond_dev, but we need the
2002                  * value of that flag before that change, as that was the value
2003                  * when this slave was attached, so we cache at the start of the
2004                  * function and use it here. Same goes for ALLMULTI below
2005                  */
2006                 if (old_flags & IFF_PROMISC)
2007                         dev_set_promiscuity(slave_dev, -1);
2008
2009                 /* unset allmulti level from slave */
2010                 if (old_flags & IFF_ALLMULTI)
2011                         dev_set_allmulti(slave_dev, -1);
2012
2013                 bond_hw_addr_flush(bond_dev, slave_dev);
2014         }
2015
2016         slave_disable_netpoll(slave);
2017
2018         /* close slave before restoring its mac address */
2019         dev_close(slave_dev);
2020
2021         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2022             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2023                 /* restore original ("permanent") mac address */
2024                 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2025                                   slave->dev->addr_len);
2026                 ss.ss_family = slave_dev->type;
2027                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
2028         }
2029
2030         if (unregister)
2031                 __dev_set_mtu(slave_dev, slave->original_mtu);
2032         else
2033                 dev_set_mtu(slave_dev, slave->original_mtu);
2034
2035         if (!netif_is_bond_master(slave_dev))
2036                 slave_dev->priv_flags &= ~IFF_BONDING;
2037
2038         kobject_put(&slave->kobj);
2039
2040         return 0;
2041 }
2042
2043 /* A wrapper used because of ndo_del_link */
2044 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2045 {
2046         return __bond_release_one(bond_dev, slave_dev, false, false);
2047 }
2048
2049 /* First release a slave and then destroy the bond if no more slaves are left.
2050  * Must be under rtnl_lock when this function is called.
2051  */
2052 static int  bond_release_and_destroy(struct net_device *bond_dev,
2053                                      struct net_device *slave_dev)
2054 {
2055         struct bonding *bond = netdev_priv(bond_dev);
2056         int ret;
2057
2058         ret = __bond_release_one(bond_dev, slave_dev, false, true);
2059         if (ret == 0 && !bond_has_slaves(bond) &&
2060             bond_dev->reg_state != NETREG_UNREGISTERING) {
2061                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2062                 netdev_info(bond_dev, "Destroying bond %s\n",
2063                             bond_dev->name);
2064                 bond_remove_proc_entry(bond);
2065                 unregister_netdevice(bond_dev);
2066         }
2067         return ret;
2068 }
2069
2070 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2071 {
2072         struct bonding *bond = netdev_priv(bond_dev);
2073         bond_fill_ifbond(bond, info);
2074 }
2075
2076 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2077 {
2078         struct bonding *bond = netdev_priv(bond_dev);
2079         struct list_head *iter;
2080         int i = 0, res = -ENODEV;
2081         struct slave *slave;
2082
2083         bond_for_each_slave(bond, slave, iter) {
2084                 if (i++ == (int)info->slave_id) {
2085                         res = 0;
2086                         bond_fill_ifslave(slave, info);
2087                         break;
2088                 }
2089         }
2090
2091         return res;
2092 }
2093
2094 /*-------------------------------- Monitoring -------------------------------*/
2095
2096 /* called with rcu_read_lock() */
2097 static int bond_miimon_inspect(struct bonding *bond)
2098 {
2099         int link_state, commit = 0;
2100         struct list_head *iter;
2101         struct slave *slave;
2102         bool ignore_updelay;
2103
2104         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2105
2106         bond_for_each_slave_rcu(bond, slave, iter) {
2107                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2108
2109                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2110
2111                 switch (slave->link) {
2112                 case BOND_LINK_UP:
2113                         if (link_state)
2114                                 continue;
2115
2116                         bond_propose_link_state(slave, BOND_LINK_FAIL);
2117                         commit++;
2118                         slave->delay = bond->params.downdelay;
2119                         if (slave->delay) {
2120                                 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2121                                             (BOND_MODE(bond) ==
2122                                              BOND_MODE_ACTIVEBACKUP) ?
2123                                              (bond_is_active_slave(slave) ?
2124                                               "active " : "backup ") : "",
2125                                             slave->dev->name,
2126                                             bond->params.downdelay * bond->params.miimon);
2127                         }
2128                         /*FALLTHRU*/
2129                 case BOND_LINK_FAIL:
2130                         if (link_state) {
2131                                 /* recovered before downdelay expired */
2132                                 bond_propose_link_state(slave, BOND_LINK_UP);
2133                                 slave->last_link_up = jiffies;
2134                                 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2135                                             (bond->params.downdelay - slave->delay) *
2136                                             bond->params.miimon,
2137                                             slave->dev->name);
2138                                 commit++;
2139                                 continue;
2140                         }
2141
2142                         if (slave->delay <= 0) {
2143                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2144                                 commit++;
2145                                 continue;
2146                         }
2147
2148                         slave->delay--;
2149                         break;
2150
2151                 case BOND_LINK_DOWN:
2152                         if (!link_state)
2153                                 continue;
2154
2155                         bond_propose_link_state(slave, BOND_LINK_BACK);
2156                         commit++;
2157                         slave->delay = bond->params.updelay;
2158
2159                         if (slave->delay) {
2160                                 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2161                                             slave->dev->name,
2162                                             ignore_updelay ? 0 :
2163                                             bond->params.updelay *
2164                                             bond->params.miimon);
2165                         }
2166                         /*FALLTHRU*/
2167                 case BOND_LINK_BACK:
2168                         if (!link_state) {
2169                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2170                                 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2171                                             (bond->params.updelay - slave->delay) *
2172                                             bond->params.miimon,
2173                                             slave->dev->name);
2174                                 commit++;
2175                                 continue;
2176                         }
2177
2178                         if (ignore_updelay)
2179                                 slave->delay = 0;
2180
2181                         if (slave->delay <= 0) {
2182                                 bond_propose_link_state(slave, BOND_LINK_UP);
2183                                 commit++;
2184                                 ignore_updelay = false;
2185                                 continue;
2186                         }
2187
2188                         slave->delay--;
2189                         break;
2190                 }
2191         }
2192
2193         return commit;
2194 }
2195
2196 static void bond_miimon_link_change(struct bonding *bond,
2197                                     struct slave *slave,
2198                                     char link)
2199 {
2200         switch (BOND_MODE(bond)) {
2201         case BOND_MODE_8023AD:
2202                 bond_3ad_handle_link_change(slave, link);
2203                 break;
2204         case BOND_MODE_TLB:
2205         case BOND_MODE_ALB:
2206                 bond_alb_handle_link_change(bond, slave, link);
2207                 break;
2208         case BOND_MODE_XOR:
2209                 bond_update_slave_arr(bond, NULL);
2210                 break;
2211         }
2212 }
2213
2214 static void bond_miimon_commit(struct bonding *bond)
2215 {
2216         struct list_head *iter;
2217         struct slave *slave, *primary;
2218
2219         bond_for_each_slave(bond, slave, iter) {
2220                 switch (slave->link_new_state) {
2221                 case BOND_LINK_NOCHANGE:
2222                         /* For 802.3ad mode, check current slave speed and
2223                          * duplex again in case its port was disabled after
2224                          * invalid speed/duplex reporting but recovered before
2225                          * link monitoring could make a decision on the actual
2226                          * link status
2227                          */
2228                         if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2229                             slave->link == BOND_LINK_UP)
2230                                 bond_3ad_adapter_speed_duplex_changed(slave);
2231                         continue;
2232
2233                 case BOND_LINK_UP:
2234                         if (bond_update_speed_duplex(slave) &&
2235                             bond_needs_speed_duplex(bond)) {
2236                                 slave->link = BOND_LINK_DOWN;
2237                                 if (net_ratelimit())
2238                                         netdev_warn(bond->dev,
2239                                                     "failed to get link speed/duplex for %s\n",
2240                                                     slave->dev->name);
2241                                 continue;
2242                         }
2243                         bond_set_slave_link_state(slave, BOND_LINK_UP,
2244                                                   BOND_SLAVE_NOTIFY_NOW);
2245                         slave->last_link_up = jiffies;
2246
2247                         primary = rtnl_dereference(bond->primary_slave);
2248                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2249                                 /* prevent it from being the active one */
2250                                 bond_set_backup_slave(slave);
2251                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2252                                 /* make it immediately active */
2253                                 bond_set_active_slave(slave);
2254                         }
2255
2256                         netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2257                                     slave->dev->name,
2258                                     slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2259                                     slave->duplex ? "full" : "half");
2260
2261                         bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2262
2263                         if (!bond->curr_active_slave || slave == primary)
2264                                 goto do_failover;
2265
2266                         continue;
2267
2268                 case BOND_LINK_DOWN:
2269                         if (slave->link_failure_count < UINT_MAX)
2270                                 slave->link_failure_count++;
2271
2272                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2273                                                   BOND_SLAVE_NOTIFY_NOW);
2274
2275                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2276                             BOND_MODE(bond) == BOND_MODE_8023AD)
2277                                 bond_set_slave_inactive_flags(slave,
2278                                                               BOND_SLAVE_NOTIFY_NOW);
2279
2280                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2281                                     slave->dev->name);
2282
2283                         bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2284
2285                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2286                                 goto do_failover;
2287
2288                         continue;
2289
2290                 default:
2291                         netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2292                                    slave->link_new_state, slave->dev->name);
2293                         bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2294
2295                         continue;
2296                 }
2297
2298 do_failover:
2299                 block_netpoll_tx();
2300                 bond_select_active_slave(bond);
2301                 unblock_netpoll_tx();
2302         }
2303
2304         bond_set_carrier(bond);
2305 }
2306
2307 /* bond_mii_monitor
2308  *
2309  * Really a wrapper that splits the mii monitor into two phases: an
2310  * inspection, then (if inspection indicates something needs to be done)
2311  * an acquisition of appropriate locks followed by a commit phase to
2312  * implement whatever link state changes are indicated.
2313  */
2314 static void bond_mii_monitor(struct work_struct *work)
2315 {
2316         struct bonding *bond = container_of(work, struct bonding,
2317                                             mii_work.work);
2318         bool should_notify_peers = false;
2319         unsigned long delay;
2320         struct slave *slave;
2321         struct list_head *iter;
2322
2323         delay = msecs_to_jiffies(bond->params.miimon);
2324
2325         if (!bond_has_slaves(bond))
2326                 goto re_arm;
2327
2328         rcu_read_lock();
2329
2330         should_notify_peers = bond_should_notify_peers(bond);
2331
2332         if (bond_miimon_inspect(bond)) {
2333                 rcu_read_unlock();
2334
2335                 /* Race avoidance with bond_close cancel of workqueue */
2336                 if (!rtnl_trylock()) {
2337                         delay = 1;
2338                         should_notify_peers = false;
2339                         goto re_arm;
2340                 }
2341
2342                 bond_for_each_slave(bond, slave, iter) {
2343                         bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2344                 }
2345                 bond_miimon_commit(bond);
2346
2347                 rtnl_unlock();  /* might sleep, hold no other locks */
2348         } else
2349                 rcu_read_unlock();
2350
2351 re_arm:
2352         if (bond->params.miimon)
2353                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2354
2355         if (should_notify_peers) {
2356                 if (!rtnl_trylock())
2357                         return;
2358                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2359                 rtnl_unlock();
2360         }
2361 }
2362
2363 static int bond_upper_dev_walk(struct net_device *upper, void *data)
2364 {
2365         __be32 ip = *((__be32 *)data);
2366
2367         return ip == bond_confirm_addr(upper, 0, ip);
2368 }
2369
2370 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2371 {
2372         bool ret = false;
2373
2374         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2375                 return true;
2376
2377         rcu_read_lock();
2378         if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip))
2379                 ret = true;
2380         rcu_read_unlock();
2381
2382         return ret;
2383 }
2384
2385 /* We go to the (large) trouble of VLAN tagging ARP frames because
2386  * switches in VLAN mode (especially if ports are configured as
2387  * "native" to a VLAN) might not pass non-tagged frames.
2388  */
2389 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2390                           __be32 dest_ip, __be32 src_ip,
2391                           struct bond_vlan_tag *tags)
2392 {
2393         struct sk_buff *skb;
2394         struct bond_vlan_tag *outer_tag = tags;
2395
2396         netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2397                    arp_op, slave_dev->name, &dest_ip, &src_ip);
2398
2399         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2400                          NULL, slave_dev->dev_addr, NULL);
2401
2402         if (!skb) {
2403                 net_err_ratelimited("ARP packet allocation failed\n");
2404                 return;
2405         }
2406
2407         if (!tags || tags->vlan_proto == VLAN_N_VID)
2408                 goto xmit;
2409
2410         tags++;
2411
2412         /* Go through all the tags backwards and add them to the packet */
2413         while (tags->vlan_proto != VLAN_N_VID) {
2414                 if (!tags->vlan_id) {
2415                         tags++;
2416                         continue;
2417                 }
2418
2419                 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2420                            ntohs(outer_tag->vlan_proto), tags->vlan_id);
2421                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2422                                                 tags->vlan_id);
2423                 if (!skb) {
2424                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2425                         return;
2426                 }
2427
2428                 tags++;
2429         }
2430         /* Set the outer tag */
2431         if (outer_tag->vlan_id) {
2432                 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2433                            ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2434                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2435                                        outer_tag->vlan_id);
2436         }
2437
2438 xmit:
2439         arp_xmit(skb);
2440 }
2441
2442 /* Validate the device path between the @start_dev and the @end_dev.
2443  * The path is valid if the @end_dev is reachable through device
2444  * stacking.
2445  * When the path is validated, collect any vlan information in the
2446  * path.
2447  */
2448 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2449                                               struct net_device *end_dev,
2450                                               int level)
2451 {
2452         struct bond_vlan_tag *tags;
2453         struct net_device *upper;
2454         struct list_head  *iter;
2455
2456         if (start_dev == end_dev) {
2457                 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2458                 if (!tags)
2459                         return ERR_PTR(-ENOMEM);
2460                 tags[level].vlan_proto = VLAN_N_VID;
2461                 return tags;
2462         }
2463
2464         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2465                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2466                 if (IS_ERR_OR_NULL(tags)) {
2467                         if (IS_ERR(tags))
2468                                 return tags;
2469                         continue;
2470                 }
2471                 if (is_vlan_dev(upper)) {
2472                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2473                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2474                 }
2475
2476                 return tags;
2477         }
2478
2479         return NULL;
2480 }
2481
2482 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2483 {
2484         struct rtable *rt;
2485         struct bond_vlan_tag *tags;
2486         __be32 *targets = bond->params.arp_targets, addr;
2487         int i;
2488
2489         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2490                 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2491                 tags = NULL;
2492
2493                 /* Find out through which dev should the packet go */
2494                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2495                                      RTO_ONLINK, 0);
2496                 if (IS_ERR(rt)) {
2497                         /* there's no route to target - try to send arp
2498                          * probe to generate any traffic (arp_validate=0)
2499                          */
2500                         if (bond->params.arp_validate)
2501                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2502                                                      bond->dev->name,
2503                                                      &targets[i]);
2504                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2505                                       0, tags);
2506                         continue;
2507                 }
2508
2509                 /* bond device itself */
2510                 if (rt->dst.dev == bond->dev)
2511                         goto found;
2512
2513                 rcu_read_lock();
2514                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2515                 rcu_read_unlock();
2516
2517                 if (!IS_ERR_OR_NULL(tags))
2518                         goto found;
2519
2520                 /* Not our device - skip */
2521                 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2522                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2523
2524                 ip_rt_put(rt);
2525                 continue;
2526
2527 found:
2528                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2529                 ip_rt_put(rt);
2530                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2531                               addr, tags);
2532                 kfree(tags);
2533         }
2534 }
2535
2536 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2537 {
2538         int i;
2539
2540         if (!sip || !bond_has_this_ip(bond, tip)) {
2541                 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2542                            &sip, &tip);
2543                 return;
2544         }
2545
2546         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2547         if (i == -1) {
2548                 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2549                            &sip);
2550                 return;
2551         }
2552         slave->last_rx = jiffies;
2553         slave->target_last_arp_rx[i] = jiffies;
2554 }
2555
2556 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2557                  struct slave *slave)
2558 {
2559         struct arphdr *arp = (struct arphdr *)skb->data;
2560         struct slave *curr_active_slave, *curr_arp_slave;
2561         unsigned char *arp_ptr;
2562         __be32 sip, tip;
2563         int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2564         unsigned int alen;
2565
2566         if (!slave_do_arp_validate(bond, slave)) {
2567                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2568                     !slave_do_arp_validate_only(bond))
2569                         slave->last_rx = jiffies;
2570                 return RX_HANDLER_ANOTHER;
2571         } else if (!is_arp) {
2572                 return RX_HANDLER_ANOTHER;
2573         }
2574
2575         alen = arp_hdr_len(bond->dev);
2576
2577         netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2578                    skb->dev->name);
2579
2580         if (alen > skb_headlen(skb)) {
2581                 arp = kmalloc(alen, GFP_ATOMIC);
2582                 if (!arp)
2583                         goto out_unlock;
2584                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2585                         goto out_unlock;
2586         }
2587
2588         if (arp->ar_hln != bond->dev->addr_len ||
2589             skb->pkt_type == PACKET_OTHERHOST ||
2590             skb->pkt_type == PACKET_LOOPBACK ||
2591             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2592             arp->ar_pro != htons(ETH_P_IP) ||
2593             arp->ar_pln != 4)
2594                 goto out_unlock;
2595
2596         arp_ptr = (unsigned char *)(arp + 1);
2597         arp_ptr += bond->dev->addr_len;
2598         memcpy(&sip, arp_ptr, 4);
2599         arp_ptr += 4 + bond->dev->addr_len;
2600         memcpy(&tip, arp_ptr, 4);
2601
2602         netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2603                    slave->dev->name, bond_slave_state(slave),
2604                      bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2605                      &sip, &tip);
2606
2607         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2608         curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2609
2610         /* We 'trust' the received ARP enough to validate it if:
2611          *
2612          * (a) the slave receiving the ARP is active (which includes the
2613          * current ARP slave, if any), or
2614          *
2615          * (b) the receiving slave isn't active, but there is a currently
2616          * active slave and it received valid arp reply(s) after it became
2617          * the currently active slave, or
2618          *
2619          * (c) there is an ARP slave that sent an ARP during the prior ARP
2620          * interval, and we receive an ARP reply on any slave.  We accept
2621          * these because switch FDB update delays may deliver the ARP
2622          * reply to a slave other than the sender of the ARP request.
2623          *
2624          * Note: for (b), backup slaves are receiving the broadcast ARP
2625          * request, not a reply.  This request passes from the sending
2626          * slave through the L2 switch(es) to the receiving slave.  Since
2627          * this is checking the request, sip/tip are swapped for
2628          * validation.
2629          *
2630          * This is done to avoid endless looping when we can't reach the
2631          * arp_ip_target and fool ourselves with our own arp requests.
2632          */
2633         if (bond_is_active_slave(slave))
2634                 bond_validate_arp(bond, slave, sip, tip);
2635         else if (curr_active_slave &&
2636                  time_after(slave_last_rx(bond, curr_active_slave),
2637                             curr_active_slave->last_link_up))
2638                 bond_validate_arp(bond, slave, tip, sip);
2639         else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2640                  bond_time_in_interval(bond,
2641                                        dev_trans_start(curr_arp_slave->dev), 1))
2642                 bond_validate_arp(bond, slave, sip, tip);
2643
2644 out_unlock:
2645         if (arp != (struct arphdr *)skb->data)
2646                 kfree(arp);
2647         return RX_HANDLER_ANOTHER;
2648 }
2649
2650 /* function to verify if we're in the arp_interval timeslice, returns true if
2651  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2652  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2653  */
2654 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2655                                   int mod)
2656 {
2657         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2658
2659         return time_in_range(jiffies,
2660                              last_act - delta_in_ticks,
2661                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2662 }
2663
2664 /* This function is called regularly to monitor each slave's link
2665  * ensuring that traffic is being sent and received when arp monitoring
2666  * is used in load-balancing mode. if the adapter has been dormant, then an
2667  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2668  * arp monitoring in active backup mode.
2669  */
2670 static void bond_loadbalance_arp_mon(struct bonding *bond)
2671 {
2672         struct slave *slave, *oldcurrent;
2673         struct list_head *iter;
2674         int do_failover = 0, slave_state_changed = 0;
2675
2676         if (!bond_has_slaves(bond))
2677                 goto re_arm;
2678
2679         rcu_read_lock();
2680
2681         oldcurrent = rcu_dereference(bond->curr_active_slave);
2682         /* see if any of the previous devices are up now (i.e. they have
2683          * xmt and rcv traffic). the curr_active_slave does not come into
2684          * the picture unless it is null. also, slave->last_link_up is not
2685          * needed here because we send an arp on each slave and give a slave
2686          * as long as it needs to get the tx/rx within the delta.
2687          * TODO: what about up/down delay in arp mode? it wasn't here before
2688          *       so it can wait
2689          */
2690         bond_for_each_slave_rcu(bond, slave, iter) {
2691                 unsigned long trans_start = dev_trans_start(slave->dev);
2692
2693                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2694
2695                 if (slave->link != BOND_LINK_UP) {
2696                         if (bond_time_in_interval(bond, trans_start, 1) &&
2697                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2698
2699                                 bond_propose_link_state(slave, BOND_LINK_UP);
2700                                 slave_state_changed = 1;
2701
2702                                 /* primary_slave has no meaning in round-robin
2703                                  * mode. the window of a slave being up and
2704                                  * curr_active_slave being null after enslaving
2705                                  * is closed.
2706                                  */
2707                                 if (!oldcurrent) {
2708                                         netdev_info(bond->dev, "link status definitely up for interface %s\n",
2709                                                     slave->dev->name);
2710                                         do_failover = 1;
2711                                 } else {
2712                                         netdev_info(bond->dev, "interface %s is now up\n",
2713                                                     slave->dev->name);
2714                                 }
2715                         }
2716                 } else {
2717                         /* slave->link == BOND_LINK_UP */
2718
2719                         /* not all switches will respond to an arp request
2720                          * when the source ip is 0, so don't take the link down
2721                          * if we don't know our ip yet
2722                          */
2723                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2724                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2725
2726                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2727                                 slave_state_changed = 1;
2728
2729                                 if (slave->link_failure_count < UINT_MAX)
2730                                         slave->link_failure_count++;
2731
2732                                 netdev_info(bond->dev, "interface %s is now down\n",
2733                                             slave->dev->name);
2734
2735                                 if (slave == oldcurrent)
2736                                         do_failover = 1;
2737                         }
2738                 }
2739
2740                 /* note: if switch is in round-robin mode, all links
2741                  * must tx arp to ensure all links rx an arp - otherwise
2742                  * links may oscillate or not come up at all; if switch is
2743                  * in something like xor mode, there is nothing we can
2744                  * do - all replies will be rx'ed on same link causing slaves
2745                  * to be unstable during low/no traffic periods
2746                  */
2747                 if (bond_slave_is_up(slave))
2748                         bond_arp_send_all(bond, slave);
2749         }
2750
2751         rcu_read_unlock();
2752
2753         if (do_failover || slave_state_changed) {
2754                 if (!rtnl_trylock())
2755                         goto re_arm;
2756
2757                 bond_for_each_slave(bond, slave, iter) {
2758                         if (slave->link_new_state != BOND_LINK_NOCHANGE)
2759                                 slave->link = slave->link_new_state;
2760                 }
2761
2762                 if (slave_state_changed) {
2763                         bond_slave_state_change(bond);
2764                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2765                                 bond_update_slave_arr(bond, NULL);
2766                 }
2767                 if (do_failover) {
2768                         block_netpoll_tx();
2769                         bond_select_active_slave(bond);
2770                         unblock_netpoll_tx();
2771                 }
2772                 rtnl_unlock();
2773         }
2774
2775 re_arm:
2776         if (bond->params.arp_interval)
2777                 queue_delayed_work(bond->wq, &bond->arp_work,
2778                                    msecs_to_jiffies(bond->params.arp_interval));
2779 }
2780
2781 /* Called to inspect slaves for active-backup mode ARP monitor link state
2782  * changes.  Sets proposed link state in slaves to specify what action
2783  * should take place for the slave.  Returns 0 if no changes are found, >0
2784  * if changes to link states must be committed.
2785  *
2786  * Called with rcu_read_lock held.
2787  */
2788 static int bond_ab_arp_inspect(struct bonding *bond)
2789 {
2790         unsigned long trans_start, last_rx;
2791         struct list_head *iter;
2792         struct slave *slave;
2793         int commit = 0;
2794
2795         bond_for_each_slave_rcu(bond, slave, iter) {
2796                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2797                 last_rx = slave_last_rx(bond, slave);
2798
2799                 if (slave->link != BOND_LINK_UP) {
2800                         if (bond_time_in_interval(bond, last_rx, 1)) {
2801                                 bond_propose_link_state(slave, BOND_LINK_UP);
2802                                 commit++;
2803                         } else if (slave->link == BOND_LINK_BACK) {
2804                                 bond_propose_link_state(slave, BOND_LINK_FAIL);
2805                                 commit++;
2806                         }
2807                         continue;
2808                 }
2809
2810                 /* Give slaves 2*delta after being enslaved or made
2811                  * active.  This avoids bouncing, as the last receive
2812                  * times need a full ARP monitor cycle to be updated.
2813                  */
2814                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2815                         continue;
2816
2817                 /* Backup slave is down if:
2818                  * - No current_arp_slave AND
2819                  * - more than 3*delta since last receive AND
2820                  * - the bond has an IP address
2821                  *
2822                  * Note: a non-null current_arp_slave indicates
2823                  * the curr_active_slave went down and we are
2824                  * searching for a new one; under this condition
2825                  * we only take the curr_active_slave down - this
2826                  * gives each slave a chance to tx/rx traffic
2827                  * before being taken out
2828                  */
2829                 if (!bond_is_active_slave(slave) &&
2830                     !rcu_access_pointer(bond->current_arp_slave) &&
2831                     !bond_time_in_interval(bond, last_rx, 3)) {
2832                         bond_propose_link_state(slave, BOND_LINK_DOWN);
2833                         commit++;
2834                 }
2835
2836                 /* Active slave is down if:
2837                  * - more than 2*delta since transmitting OR
2838                  * - (more than 2*delta since receive AND
2839                  *    the bond has an IP address)
2840                  */
2841                 trans_start = dev_trans_start(slave->dev);
2842                 if (bond_is_active_slave(slave) &&
2843                     (!bond_time_in_interval(bond, trans_start, 2) ||
2844                      !bond_time_in_interval(bond, last_rx, 2))) {
2845                         bond_propose_link_state(slave, BOND_LINK_DOWN);
2846                         commit++;
2847                 }
2848         }
2849
2850         return commit;
2851 }
2852
2853 /* Called to commit link state changes noted by inspection step of
2854  * active-backup mode ARP monitor.
2855  *
2856  * Called with RTNL hold.
2857  */
2858 static void bond_ab_arp_commit(struct bonding *bond)
2859 {
2860         unsigned long trans_start;
2861         struct list_head *iter;
2862         struct slave *slave;
2863
2864         bond_for_each_slave(bond, slave, iter) {
2865                 switch (slave->link_new_state) {
2866                 case BOND_LINK_NOCHANGE:
2867                         continue;
2868
2869                 case BOND_LINK_UP:
2870                         trans_start = dev_trans_start(slave->dev);
2871                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
2872                             (!rtnl_dereference(bond->curr_active_slave) &&
2873                              bond_time_in_interval(bond, trans_start, 1))) {
2874                                 struct slave *current_arp_slave;
2875
2876                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2877                                 bond_set_slave_link_state(slave, BOND_LINK_UP,
2878                                                           BOND_SLAVE_NOTIFY_NOW);
2879                                 if (current_arp_slave) {
2880                                         bond_set_slave_inactive_flags(
2881                                                 current_arp_slave,
2882                                                 BOND_SLAVE_NOTIFY_NOW);
2883                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2884                                 }
2885
2886                                 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2887                                             slave->dev->name);
2888
2889                                 if (!rtnl_dereference(bond->curr_active_slave) ||
2890                                     slave == rtnl_dereference(bond->primary_slave))
2891                                         goto do_failover;
2892
2893                         }
2894
2895                         continue;
2896
2897                 case BOND_LINK_DOWN:
2898                         if (slave->link_failure_count < UINT_MAX)
2899                                 slave->link_failure_count++;
2900
2901                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2902                                                   BOND_SLAVE_NOTIFY_NOW);
2903                         bond_set_slave_inactive_flags(slave,
2904                                                       BOND_SLAVE_NOTIFY_NOW);
2905
2906                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2907                                     slave->dev->name);
2908
2909                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
2910                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2911                                 goto do_failover;
2912                         }
2913
2914                         continue;
2915
2916                 case BOND_LINK_FAIL:
2917                         bond_set_slave_link_state(slave, BOND_LINK_FAIL,
2918                                                   BOND_SLAVE_NOTIFY_NOW);
2919                         bond_set_slave_inactive_flags(slave,
2920                                                       BOND_SLAVE_NOTIFY_NOW);
2921
2922                         /* A slave has just been enslaved and has become
2923                          * the current active slave.
2924                          */
2925                         if (rtnl_dereference(bond->curr_active_slave))
2926                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2927                         continue;
2928
2929                 default:
2930                         netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2931                                    slave->link_new_state, slave->dev->name);
2932                         continue;
2933                 }
2934
2935 do_failover:
2936                 block_netpoll_tx();
2937                 bond_select_active_slave(bond);
2938                 unblock_netpoll_tx();
2939         }
2940
2941         bond_set_carrier(bond);
2942 }
2943
2944 /* Send ARP probes for active-backup mode ARP monitor.
2945  *
2946  * Called with rcu_read_lock held.
2947  */
2948 static bool bond_ab_arp_probe(struct bonding *bond)
2949 {
2950         struct slave *slave, *before = NULL, *new_slave = NULL,
2951                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2952                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2953         struct list_head *iter;
2954         bool found = false;
2955         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2956
2957         if (curr_arp_slave && curr_active_slave)
2958                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2959                             curr_arp_slave->dev->name,
2960                             curr_active_slave->dev->name);
2961
2962         if (curr_active_slave) {
2963                 bond_arp_send_all(bond, curr_active_slave);
2964                 return should_notify_rtnl;
2965         }
2966
2967         /* if we don't have a curr_active_slave, search for the next available
2968          * backup slave from the current_arp_slave and make it the candidate
2969          * for becoming the curr_active_slave
2970          */
2971
2972         if (!curr_arp_slave) {
2973                 curr_arp_slave = bond_first_slave_rcu(bond);
2974                 if (!curr_arp_slave)
2975                         return should_notify_rtnl;
2976         }
2977
2978         bond_for_each_slave_rcu(bond, slave, iter) {
2979                 if (!found && !before && bond_slave_is_up(slave))
2980                         before = slave;
2981
2982                 if (found && !new_slave && bond_slave_is_up(slave))
2983                         new_slave = slave;
2984                 /* if the link state is up at this point, we
2985                  * mark it down - this can happen if we have
2986                  * simultaneous link failures and
2987                  * reselect_active_interface doesn't make this
2988                  * one the current slave so it is still marked
2989                  * up when it is actually down
2990                  */
2991                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2992                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2993                                                   BOND_SLAVE_NOTIFY_LATER);
2994                         if (slave->link_failure_count < UINT_MAX)
2995                                 slave->link_failure_count++;
2996
2997                         bond_set_slave_inactive_flags(slave,
2998                                                       BOND_SLAVE_NOTIFY_LATER);
2999
3000                         netdev_info(bond->dev, "backup interface %s is now down\n",
3001                                     slave->dev->name);
3002                 }
3003                 if (slave == curr_arp_slave)
3004                         found = true;
3005         }
3006
3007         if (!new_slave && before)
3008                 new_slave = before;
3009
3010         if (!new_slave)
3011                 goto check_state;
3012
3013         bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
3014                                   BOND_SLAVE_NOTIFY_LATER);
3015         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
3016         bond_arp_send_all(bond, new_slave);
3017         new_slave->last_link_up = jiffies;
3018         rcu_assign_pointer(bond->current_arp_slave, new_slave);
3019
3020 check_state:
3021         bond_for_each_slave_rcu(bond, slave, iter) {
3022                 if (slave->should_notify || slave->should_notify_link) {
3023                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3024                         break;
3025                 }
3026         }
3027         return should_notify_rtnl;
3028 }
3029
3030 static void bond_activebackup_arp_mon(struct bonding *bond)
3031 {
3032         bool should_notify_peers = false;
3033         bool should_notify_rtnl = false;
3034         int delta_in_ticks;
3035
3036         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3037
3038         if (!bond_has_slaves(bond))
3039                 goto re_arm;
3040
3041         rcu_read_lock();
3042
3043         should_notify_peers = bond_should_notify_peers(bond);
3044
3045         if (bond_ab_arp_inspect(bond)) {
3046                 rcu_read_unlock();
3047
3048                 /* Race avoidance with bond_close flush of workqueue */
3049                 if (!rtnl_trylock()) {
3050                         delta_in_ticks = 1;
3051                         should_notify_peers = false;
3052                         goto re_arm;
3053                 }
3054
3055                 bond_ab_arp_commit(bond);
3056
3057                 rtnl_unlock();
3058                 rcu_read_lock();
3059         }
3060
3061         should_notify_rtnl = bond_ab_arp_probe(bond);
3062         rcu_read_unlock();
3063
3064 re_arm:
3065         if (bond->params.arp_interval)
3066                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3067
3068         if (should_notify_peers || should_notify_rtnl) {
3069                 if (!rtnl_trylock())
3070                         return;
3071
3072                 if (should_notify_peers)
3073                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3074                                                  bond->dev);
3075                 if (should_notify_rtnl) {
3076                         bond_slave_state_notify(bond);
3077                         bond_slave_link_notify(bond);
3078                 }
3079
3080                 rtnl_unlock();
3081         }
3082 }
3083
3084 static void bond_arp_monitor(struct work_struct *work)
3085 {
3086         struct bonding *bond = container_of(work, struct bonding,
3087                                             arp_work.work);
3088
3089         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3090                 bond_activebackup_arp_mon(bond);
3091         else
3092                 bond_loadbalance_arp_mon(bond);
3093 }
3094
3095 /*-------------------------- netdev event handling --------------------------*/
3096
3097 /* Change device name */
3098 static int bond_event_changename(struct bonding *bond)
3099 {
3100         bond_remove_proc_entry(bond);
3101         bond_create_proc_entry(bond);
3102
3103         bond_debug_reregister(bond);
3104
3105         return NOTIFY_DONE;
3106 }
3107
3108 static int bond_master_netdev_event(unsigned long event,
3109                                     struct net_device *bond_dev)
3110 {
3111         struct bonding *event_bond = netdev_priv(bond_dev);
3112
3113         switch (event) {
3114         case NETDEV_CHANGENAME:
3115                 return bond_event_changename(event_bond);
3116         case NETDEV_UNREGISTER:
3117                 bond_remove_proc_entry(event_bond);
3118                 break;
3119         case NETDEV_REGISTER:
3120                 bond_create_proc_entry(event_bond);
3121                 break;
3122         case NETDEV_NOTIFY_PEERS:
3123                 if (event_bond->send_peer_notif)
3124                         event_bond->send_peer_notif--;
3125                 break;
3126         default:
3127                 break;
3128         }
3129
3130         return NOTIFY_DONE;
3131 }
3132
3133 static int bond_slave_netdev_event(unsigned long event,
3134                                    struct net_device *slave_dev)
3135 {
3136         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3137         struct bonding *bond;
3138         struct net_device *bond_dev;
3139
3140         /* A netdev event can be generated while enslaving a device
3141          * before netdev_rx_handler_register is called in which case
3142          * slave will be NULL
3143          */
3144         if (!slave)
3145                 return NOTIFY_DONE;
3146         bond_dev = slave->bond->dev;
3147         bond = slave->bond;
3148         primary = rtnl_dereference(bond->primary_slave);
3149
3150         switch (event) {
3151         case NETDEV_UNREGISTER:
3152                 if (bond_dev->type != ARPHRD_ETHER)
3153                         bond_release_and_destroy(bond_dev, slave_dev);
3154                 else
3155                         __bond_release_one(bond_dev, slave_dev, false, true);
3156                 break;
3157         case NETDEV_UP:
3158         case NETDEV_CHANGE:
3159                 /* For 802.3ad mode only:
3160                  * Getting invalid Speed/Duplex values here will put slave
3161                  * in weird state. Mark it as link-fail if the link was
3162                  * previously up or link-down if it hasn't yet come up, and
3163                  * let link-monitoring (miimon) set it right when correct
3164                  * speeds/duplex are available.
3165                  */
3166                 if (bond_update_speed_duplex(slave) &&
3167                     BOND_MODE(bond) == BOND_MODE_8023AD) {
3168                         if (slave->last_link_up)
3169                                 slave->link = BOND_LINK_FAIL;
3170                         else
3171                                 slave->link = BOND_LINK_DOWN;
3172                 }
3173
3174                 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3175                         bond_3ad_adapter_speed_duplex_changed(slave);
3176                 /* Fallthrough */
3177         case NETDEV_DOWN:
3178                 /* Refresh slave-array if applicable!
3179                  * If the setup does not use miimon or arpmon (mode-specific!),
3180                  * then these events will not cause the slave-array to be
3181                  * refreshed. This will cause xmit to use a slave that is not
3182                  * usable. Avoid such situation by refeshing the array at these
3183                  * events. If these (miimon/arpmon) parameters are configured
3184                  * then array gets refreshed twice and that should be fine!
3185                  */
3186                 if (bond_mode_can_use_xmit_hash(bond))
3187                         bond_update_slave_arr(bond, NULL);
3188                 break;
3189         case NETDEV_CHANGEMTU:
3190                 /* TODO: Should slaves be allowed to
3191                  * independently alter their MTU?  For
3192                  * an active-backup bond, slaves need
3193                  * not be the same type of device, so
3194                  * MTUs may vary.  For other modes,
3195                  * slaves arguably should have the
3196                  * same MTUs. To do this, we'd need to
3197                  * take over the slave's change_mtu
3198                  * function for the duration of their
3199                  * servitude.
3200                  */
3201                 break;
3202         case NETDEV_CHANGENAME:
3203                 /* we don't care if we don't have primary set */
3204                 if (!bond_uses_primary(bond) ||
3205                     !bond->params.primary[0])
3206                         break;
3207
3208                 if (slave == primary) {
3209                         /* slave's name changed - he's no longer primary */
3210                         RCU_INIT_POINTER(bond->primary_slave, NULL);
3211                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3212                         /* we have a new primary slave */
3213                         rcu_assign_pointer(bond->primary_slave, slave);
3214                 } else { /* we didn't change primary - exit */
3215                         break;
3216                 }
3217
3218                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3219                             primary ? slave_dev->name : "none");
3220
3221                 block_netpoll_tx();
3222                 bond_select_active_slave(bond);
3223                 unblock_netpoll_tx();
3224                 break;
3225         case NETDEV_FEAT_CHANGE:
3226                 bond_compute_features(bond);
3227                 break;
3228         case NETDEV_RESEND_IGMP:
3229                 /* Propagate to master device */
3230                 call_netdevice_notifiers(event, slave->bond->dev);
3231                 break;
3232         default:
3233                 break;
3234         }
3235
3236         return NOTIFY_DONE;
3237 }
3238
3239 /* bond_netdev_event: handle netdev notifier chain events.
3240  *
3241  * This function receives events for the netdev chain.  The caller (an
3242  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3243  * locks for us to safely manipulate the slave devices (RTNL lock,
3244  * dev_probe_lock).
3245  */
3246 static int bond_netdev_event(struct notifier_block *this,
3247                              unsigned long event, void *ptr)
3248 {
3249         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3250
3251         netdev_dbg(event_dev, "event: %lx\n", event);
3252
3253         if (!(event_dev->priv_flags & IFF_BONDING))
3254                 return NOTIFY_DONE;
3255
3256         if (event_dev->flags & IFF_MASTER) {
3257                 int ret;
3258
3259                 netdev_dbg(event_dev, "IFF_MASTER\n");
3260                 ret = bond_master_netdev_event(event, event_dev);
3261                 if (ret != NOTIFY_DONE)
3262                         return ret;
3263         }
3264
3265         if (event_dev->flags & IFF_SLAVE) {
3266                 netdev_dbg(event_dev, "IFF_SLAVE\n");
3267                 return bond_slave_netdev_event(event, event_dev);
3268         }
3269
3270         return NOTIFY_DONE;
3271 }
3272
3273 static struct notifier_block bond_netdev_notifier = {
3274         .notifier_call = bond_netdev_event,
3275 };
3276
3277 /*---------------------------- Hashing Policies -----------------------------*/
3278
3279 /* L2 hash helper */
3280 static inline u32 bond_eth_hash(struct sk_buff *skb)
3281 {
3282         struct ethhdr *ep, hdr_tmp;
3283
3284         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3285         if (ep)
3286                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3287         return 0;
3288 }
3289
3290 /* Extract the appropriate headers based on bond's xmit policy */
3291 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3292                               struct flow_keys *fk)
3293 {
3294         const struct ipv6hdr *iph6;
3295         const struct iphdr *iph;
3296         int noff, proto = -1;
3297
3298         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3299                 return skb_flow_dissect_flow_keys(skb, fk, 0);
3300
3301         fk->ports.ports = 0;
3302         noff = skb_network_offset(skb);
3303         if (skb->protocol == htons(ETH_P_IP)) {
3304                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3305                         return false;
3306                 iph = ip_hdr(skb);
3307                 iph_to_flow_copy_v4addrs(fk, iph);
3308                 noff += iph->ihl << 2;
3309                 if (!ip_is_fragment(iph))
3310                         proto = iph->protocol;
3311         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3312                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3313                         return false;
3314                 iph6 = ipv6_hdr(skb);
3315                 iph_to_flow_copy_v6addrs(fk, iph6);
3316                 noff += sizeof(*iph6);
3317                 proto = iph6->nexthdr;
3318         } else {
3319                 return false;
3320         }
3321         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3322                 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3323
3324         return true;
3325 }
3326
3327 /**
3328  * bond_xmit_hash - generate a hash value based on the xmit policy
3329  * @bond: bonding device
3330  * @skb: buffer to use for headers
3331  *
3332  * This function will extract the necessary headers from the skb buffer and use
3333  * them to generate a hash based on the xmit_policy set in the bonding device
3334  */
3335 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3336 {
3337         struct flow_keys flow;
3338         u32 hash;
3339
3340         if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3341             skb->l4_hash)
3342                 return skb->hash;
3343
3344         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3345             !bond_flow_dissect(bond, skb, &flow))
3346                 return bond_eth_hash(skb);
3347
3348         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3349             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3350                 hash = bond_eth_hash(skb);
3351         else
3352                 hash = (__force u32)flow.ports.ports;
3353         hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3354                 (__force u32)flow_get_u32_src(&flow);
3355         hash ^= (hash >> 16);
3356         hash ^= (hash >> 8);
3357
3358         return hash >> 1;
3359 }
3360
3361 /*-------------------------- Device entry points ----------------------------*/
3362
3363 void bond_work_init_all(struct bonding *bond)
3364 {
3365         INIT_DELAYED_WORK(&bond->mcast_work,
3366                           bond_resend_igmp_join_requests_delayed);
3367         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3368         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3369         INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3370         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3371         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3372 }
3373
3374 static void bond_work_cancel_all(struct bonding *bond)
3375 {
3376         cancel_delayed_work_sync(&bond->mii_work);
3377         cancel_delayed_work_sync(&bond->arp_work);
3378         cancel_delayed_work_sync(&bond->alb_work);
3379         cancel_delayed_work_sync(&bond->ad_work);
3380         cancel_delayed_work_sync(&bond->mcast_work);
3381         cancel_delayed_work_sync(&bond->slave_arr_work);
3382 }
3383
3384 static int bond_open(struct net_device *bond_dev)
3385 {
3386         struct bonding *bond = netdev_priv(bond_dev);
3387         struct list_head *iter;
3388         struct slave *slave;
3389
3390         /* reset slave->backup and slave->inactive */
3391         if (bond_has_slaves(bond)) {
3392                 bond_for_each_slave(bond, slave, iter) {
3393                         if (bond_uses_primary(bond) &&
3394                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3395                                 bond_set_slave_inactive_flags(slave,
3396                                                               BOND_SLAVE_NOTIFY_NOW);
3397                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3398                                 bond_set_slave_active_flags(slave,
3399                                                             BOND_SLAVE_NOTIFY_NOW);
3400                         }
3401                 }
3402         }
3403
3404         if (bond_is_lb(bond)) {
3405                 /* bond_alb_initialize must be called before the timer
3406                  * is started.
3407                  */
3408                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3409                         return -ENOMEM;
3410                 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
3411                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3412         }
3413
3414         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3415                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3416
3417         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3418                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3419                 bond->recv_probe = bond_arp_rcv;
3420         }
3421
3422         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3423                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3424                 /* register to receive LACPDUs */
3425                 bond->recv_probe = bond_3ad_lacpdu_recv;
3426                 bond_3ad_initiate_agg_selection(bond, 1);
3427         }
3428
3429         if (bond_mode_can_use_xmit_hash(bond))
3430                 bond_update_slave_arr(bond, NULL);
3431
3432         return 0;
3433 }
3434
3435 static int bond_close(struct net_device *bond_dev)
3436 {
3437         struct bonding *bond = netdev_priv(bond_dev);
3438
3439         bond_work_cancel_all(bond);
3440         bond->send_peer_notif = 0;
3441         if (bond_is_lb(bond))
3442                 bond_alb_deinitialize(bond);
3443         bond->recv_probe = NULL;
3444
3445         return 0;
3446 }
3447
3448 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3449  * that some drivers can provide 32bit values only.
3450  */
3451 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3452                             const struct rtnl_link_stats64 *_new,
3453                             const struct rtnl_link_stats64 *_old)
3454 {
3455         const u64 *new = (const u64 *)_new;
3456         const u64 *old = (const u64 *)_old;
3457         u64 *res = (u64 *)_res;
3458         int i;
3459
3460         for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3461                 u64 nv = new[i];
3462                 u64 ov = old[i];
3463                 s64 delta = nv - ov;
3464
3465                 /* detects if this particular field is 32bit only */
3466                 if (((nv | ov) >> 32) == 0)
3467                         delta = (s64)(s32)((u32)nv - (u32)ov);
3468
3469                 /* filter anomalies, some drivers reset their stats
3470                  * at down/up events.
3471                  */
3472                 if (delta > 0)
3473                         res[i] += delta;
3474         }
3475 }
3476
3477 static int bond_get_nest_level(struct net_device *bond_dev)
3478 {
3479         struct bonding *bond = netdev_priv(bond_dev);
3480
3481         return bond->nest_level;
3482 }
3483
3484 static void bond_get_stats(struct net_device *bond_dev,
3485                            struct rtnl_link_stats64 *stats)
3486 {
3487         struct bonding *bond = netdev_priv(bond_dev);
3488         struct rtnl_link_stats64 temp;
3489         struct list_head *iter;
3490         struct slave *slave;
3491
3492         spin_lock_nested(&bond->stats_lock, bond_get_nest_level(bond_dev));
3493         memcpy(stats, &bond->bond_stats, sizeof(*stats));
3494
3495         rcu_read_lock();
3496         bond_for_each_slave_rcu(bond, slave, iter) {
3497                 const struct rtnl_link_stats64 *new =
3498                         dev_get_stats(slave->dev, &temp);
3499
3500                 bond_fold_stats(stats, new, &slave->slave_stats);
3501
3502                 /* save off the slave stats for the next run */
3503                 memcpy(&slave->slave_stats, new, sizeof(*new));
3504         }
3505         rcu_read_unlock();
3506
3507         memcpy(&bond->bond_stats, stats, sizeof(*stats));
3508         spin_unlock(&bond->stats_lock);
3509 }
3510
3511 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3512 {
3513         struct bonding *bond = netdev_priv(bond_dev);
3514         struct net_device *slave_dev = NULL;
3515         struct ifbond k_binfo;
3516         struct ifbond __user *u_binfo = NULL;
3517         struct ifslave k_sinfo;
3518         struct ifslave __user *u_sinfo = NULL;
3519         struct mii_ioctl_data *mii = NULL;
3520         struct bond_opt_value newval;
3521         struct net *net;
3522         int res = 0;
3523
3524         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3525
3526         switch (cmd) {
3527         case SIOCGMIIPHY:
3528                 mii = if_mii(ifr);
3529                 if (!mii)
3530                         return -EINVAL;
3531
3532                 mii->phy_id = 0;
3533                 /* Fall Through */
3534         case SIOCGMIIREG:
3535                 /* We do this again just in case we were called by SIOCGMIIREG
3536                  * instead of SIOCGMIIPHY.
3537                  */
3538                 mii = if_mii(ifr);
3539                 if (!mii)
3540                         return -EINVAL;
3541
3542                 if (mii->reg_num == 1) {
3543                         mii->val_out = 0;
3544                         if (netif_carrier_ok(bond->dev))
3545                                 mii->val_out = BMSR_LSTATUS;
3546                 }
3547
3548                 return 0;
3549         case BOND_INFO_QUERY_OLD:
3550         case SIOCBONDINFOQUERY:
3551                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3552
3553                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3554                         return -EFAULT;
3555
3556                 bond_info_query(bond_dev, &k_binfo);
3557                 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3558                         return -EFAULT;
3559
3560                 return 0;
3561         case BOND_SLAVE_INFO_QUERY_OLD:
3562         case SIOCBONDSLAVEINFOQUERY:
3563                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3564
3565                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3566                         return -EFAULT;
3567
3568                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3569                 if (res == 0 &&
3570                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3571                         return -EFAULT;
3572
3573                 return res;
3574         default:
3575                 break;
3576         }
3577
3578         net = dev_net(bond_dev);
3579
3580         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3581                 return -EPERM;
3582
3583         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3584
3585         netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3586
3587         if (!slave_dev)
3588                 return -ENODEV;
3589
3590         netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3591         switch (cmd) {
3592         case BOND_ENSLAVE_OLD:
3593         case SIOCBONDENSLAVE:
3594                 res = bond_enslave(bond_dev, slave_dev, NULL);
3595                 break;
3596         case BOND_RELEASE_OLD:
3597         case SIOCBONDRELEASE:
3598                 res = bond_release(bond_dev, slave_dev);
3599                 break;
3600         case BOND_SETHWADDR_OLD:
3601         case SIOCBONDSETHWADDR:
3602                 bond_set_dev_addr(bond_dev, slave_dev);
3603                 res = 0;
3604                 break;
3605         case BOND_CHANGE_ACTIVE_OLD:
3606         case SIOCBONDCHANGEACTIVE:
3607                 bond_opt_initstr(&newval, slave_dev->name);
3608                 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
3609                                             &newval);
3610                 break;
3611         default:
3612                 res = -EOPNOTSUPP;
3613         }
3614
3615         return res;
3616 }
3617
3618 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3619 {
3620         struct bonding *bond = netdev_priv(bond_dev);
3621
3622         if (change & IFF_PROMISC)
3623                 bond_set_promiscuity(bond,
3624                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3625
3626         if (change & IFF_ALLMULTI)
3627                 bond_set_allmulti(bond,
3628                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3629 }
3630
3631 static void bond_set_rx_mode(struct net_device *bond_dev)
3632 {
3633         struct bonding *bond = netdev_priv(bond_dev);
3634         struct list_head *iter;
3635         struct slave *slave;
3636
3637         rcu_read_lock();
3638         if (bond_uses_primary(bond)) {
3639                 slave = rcu_dereference(bond->curr_active_slave);
3640                 if (slave) {
3641                         dev_uc_sync(slave->dev, bond_dev);
3642                         dev_mc_sync(slave->dev, bond_dev);
3643                 }
3644         } else {
3645                 bond_for_each_slave_rcu(bond, slave, iter) {
3646                         dev_uc_sync_multiple(slave->dev, bond_dev);
3647                         dev_mc_sync_multiple(slave->dev, bond_dev);
3648                 }
3649         }
3650         rcu_read_unlock();
3651 }
3652
3653 static int bond_neigh_init(struct neighbour *n)
3654 {
3655         struct bonding *bond = netdev_priv(n->dev);
3656         const struct net_device_ops *slave_ops;
3657         struct neigh_parms parms;
3658         struct slave *slave;
3659         int ret;
3660
3661         slave = bond_first_slave(bond);
3662         if (!slave)
3663                 return 0;
3664         slave_ops = slave->dev->netdev_ops;
3665         if (!slave_ops->ndo_neigh_setup)
3666                 return 0;
3667
3668         parms.neigh_setup = NULL;
3669         parms.neigh_cleanup = NULL;
3670         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3671         if (ret)
3672                 return ret;
3673
3674         /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3675          * after the last slave has been detached.  Assumes that all slaves
3676          * utilize the same neigh_cleanup (true at this writing as only user
3677          * is ipoib).
3678          */
3679         n->parms->neigh_cleanup = parms.neigh_cleanup;
3680
3681         if (!parms.neigh_setup)
3682                 return 0;
3683
3684         return parms.neigh_setup(n);
3685 }
3686
3687 /* The bonding ndo_neigh_setup is called at init time beofre any
3688  * slave exists. So we must declare proxy setup function which will
3689  * be used at run time to resolve the actual slave neigh param setup.
3690  *
3691  * It's also called by master devices (such as vlans) to setup their
3692  * underlying devices. In that case - do nothing, we're already set up from
3693  * our init.
3694  */
3695 static int bond_neigh_setup(struct net_device *dev,
3696                             struct neigh_parms *parms)
3697 {
3698         /* modify only our neigh_parms */
3699         if (parms->dev == dev)
3700                 parms->neigh_setup = bond_neigh_init;
3701
3702         return 0;
3703 }
3704
3705 /* Change the MTU of all of a master's slaves to match the master */
3706 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3707 {
3708         struct bonding *bond = netdev_priv(bond_dev);
3709         struct slave *slave, *rollback_slave;
3710         struct list_head *iter;
3711         int res = 0;
3712
3713         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3714
3715         bond_for_each_slave(bond, slave, iter) {
3716                 netdev_dbg(bond_dev, "s %p c_m %p\n",
3717                            slave, slave->dev->netdev_ops->ndo_change_mtu);
3718
3719                 res = dev_set_mtu(slave->dev, new_mtu);
3720
3721                 if (res) {
3722                         /* If we failed to set the slave's mtu to the new value
3723                          * we must abort the operation even in ACTIVE_BACKUP
3724                          * mode, because if we allow the backup slaves to have
3725                          * different mtu values than the active slave we'll
3726                          * need to change their mtu when doing a failover. That
3727                          * means changing their mtu from timer context, which
3728                          * is probably not a good idea.
3729                          */
3730                         netdev_dbg(bond_dev, "err %d %s\n", res,
3731                                    slave->dev->name);
3732                         goto unwind;
3733                 }
3734         }
3735
3736         bond_dev->mtu = new_mtu;
3737
3738         return 0;
3739
3740 unwind:
3741         /* unwind from head to the slave that failed */
3742         bond_for_each_slave(bond, rollback_slave, iter) {
3743                 int tmp_res;
3744
3745                 if (rollback_slave == slave)
3746                         break;
3747
3748                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3749                 if (tmp_res) {
3750                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3751                                    tmp_res, rollback_slave->dev->name);
3752                 }
3753         }
3754
3755         return res;
3756 }
3757
3758 /* Change HW address
3759  *
3760  * Note that many devices must be down to change the HW address, and
3761  * downing the master releases all slaves.  We can make bonds full of
3762  * bonding devices to test this, however.
3763  */
3764 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3765 {
3766         struct bonding *bond = netdev_priv(bond_dev);
3767         struct slave *slave, *rollback_slave;
3768         struct sockaddr_storage *ss = addr, tmp_ss;
3769         struct list_head *iter;
3770         int res = 0;
3771
3772         if (BOND_MODE(bond) == BOND_MODE_ALB)
3773                 return bond_alb_set_mac_address(bond_dev, addr);
3774
3775
3776         netdev_dbg(bond_dev, "bond=%p\n", bond);
3777
3778         /* If fail_over_mac is enabled, do nothing and return success.
3779          * Returning an error causes ifenslave to fail.
3780          */
3781         if (bond->params.fail_over_mac &&
3782             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3783                 return 0;
3784
3785         if (!is_valid_ether_addr(ss->__data))
3786                 return -EADDRNOTAVAIL;
3787
3788         bond_for_each_slave(bond, slave, iter) {
3789                 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3790                 res = dev_set_mac_address(slave->dev, addr);
3791                 if (res) {
3792                         /* TODO: consider downing the slave
3793                          * and retry ?
3794                          * User should expect communications
3795                          * breakage anyway until ARP finish
3796                          * updating, so...
3797                          */
3798                         netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3799                         goto unwind;
3800                 }
3801         }
3802
3803         /* success */
3804         memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
3805         return 0;
3806
3807 unwind:
3808         memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
3809         tmp_ss.ss_family = bond_dev->type;
3810
3811         /* unwind from head to the slave that failed */
3812         bond_for_each_slave(bond, rollback_slave, iter) {
3813                 int tmp_res;
3814
3815                 if (rollback_slave == slave)
3816                         break;
3817
3818                 tmp_res = dev_set_mac_address(rollback_slave->dev,
3819                                               (struct sockaddr *)&tmp_ss);
3820                 if (tmp_res) {
3821                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3822                                    tmp_res, rollback_slave->dev->name);
3823                 }
3824         }
3825
3826         return res;
3827 }
3828
3829 /**
3830  * bond_xmit_slave_id - transmit skb through slave with slave_id
3831  * @bond: bonding device that is transmitting
3832  * @skb: buffer to transmit
3833  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3834  *
3835  * This function tries to transmit through slave with slave_id but in case
3836  * it fails, it tries to find the first available slave for transmission.
3837  * The skb is consumed in all cases, thus the function is void.
3838  */
3839 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3840 {
3841         struct list_head *iter;
3842         struct slave *slave;
3843         int i = slave_id;
3844
3845         /* Here we start from the slave with slave_id */
3846         bond_for_each_slave_rcu(bond, slave, iter) {
3847                 if (--i < 0) {
3848                         if (bond_slave_can_tx(slave)) {
3849                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3850                                 return;
3851                         }
3852                 }
3853         }
3854
3855         /* Here we start from the first slave up to slave_id */
3856         i = slave_id;
3857         bond_for_each_slave_rcu(bond, slave, iter) {
3858                 if (--i < 0)
3859                         break;
3860                 if (bond_slave_can_tx(slave)) {
3861                         bond_dev_queue_xmit(bond, skb, slave->dev);
3862                         return;
3863                 }
3864         }
3865         /* no slave that can tx has been found */
3866         bond_tx_drop(bond->dev, skb);
3867 }
3868
3869 /**
3870  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3871  * @bond: bonding device to use
3872  *
3873  * Based on the value of the bonding device's packets_per_slave parameter
3874  * this function generates a slave id, which is usually used as the next
3875  * slave to transmit through.
3876  */
3877 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3878 {
3879         u32 slave_id;
3880         struct reciprocal_value reciprocal_packets_per_slave;
3881         int packets_per_slave = bond->params.packets_per_slave;
3882
3883         switch (packets_per_slave) {
3884         case 0:
3885                 slave_id = prandom_u32();
3886                 break;
3887         case 1:
3888                 slave_id = bond->rr_tx_counter;
3889                 break;
3890         default:
3891                 reciprocal_packets_per_slave =
3892                         bond->params.reciprocal_packets_per_slave;
3893                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3894                                              reciprocal_packets_per_slave);
3895                 break;
3896         }
3897         bond->rr_tx_counter++;
3898
3899         return slave_id;
3900 }
3901
3902 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
3903                                         struct net_device *bond_dev)
3904 {
3905         struct bonding *bond = netdev_priv(bond_dev);
3906         struct slave *slave;
3907         int slave_cnt;
3908         u32 slave_id;
3909
3910         /* Start with the curr_active_slave that joined the bond as the
3911          * default for sending IGMP traffic.  For failover purposes one
3912          * needs to maintain some consistency for the interface that will
3913          * send the join/membership reports.  The curr_active_slave found
3914          * will send all of this type of traffic.
3915          */
3916         if (skb->protocol == htons(ETH_P_IP)) {
3917                 int noff = skb_network_offset(skb);
3918                 struct iphdr *iph;
3919
3920                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3921                         goto non_igmp;
3922
3923                 iph = ip_hdr(skb);
3924                 if (iph->protocol == IPPROTO_IGMP) {
3925                         slave = rcu_dereference(bond->curr_active_slave);
3926                         if (slave)
3927                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3928                         else
3929                                 bond_xmit_slave_id(bond, skb, 0);
3930                         return NETDEV_TX_OK;
3931                 }
3932         }
3933
3934 non_igmp:
3935         slave_cnt = READ_ONCE(bond->slave_cnt);
3936         if (likely(slave_cnt)) {
3937                 slave_id = bond_rr_gen_slave_id(bond);
3938                 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3939         } else {
3940                 bond_tx_drop(bond_dev, skb);
3941         }
3942         return NETDEV_TX_OK;
3943 }
3944
3945 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3946  * the bond has a usable interface.
3947  */
3948 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
3949                                           struct net_device *bond_dev)
3950 {
3951         struct bonding *bond = netdev_priv(bond_dev);
3952         struct slave *slave;
3953
3954         slave = rcu_dereference(bond->curr_active_slave);
3955         if (slave)
3956                 bond_dev_queue_xmit(bond, skb, slave->dev);
3957         else
3958                 bond_tx_drop(bond_dev, skb);
3959
3960         return NETDEV_TX_OK;
3961 }
3962
3963 /* Use this to update slave_array when (a) it's not appropriate to update
3964  * slave_array right away (note that update_slave_array() may sleep)
3965  * and / or (b) RTNL is not held.
3966  */
3967 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3968 {
3969         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3970 }
3971
3972 /* Slave array work handler. Holds only RTNL */
3973 static void bond_slave_arr_handler(struct work_struct *work)
3974 {
3975         struct bonding *bond = container_of(work, struct bonding,
3976                                             slave_arr_work.work);
3977         int ret;
3978
3979         if (!rtnl_trylock())
3980                 goto err;
3981
3982         ret = bond_update_slave_arr(bond, NULL);
3983         rtnl_unlock();
3984         if (ret) {
3985                 pr_warn_ratelimited("Failed to update slave array from WT\n");
3986                 goto err;
3987         }
3988         return;
3989
3990 err:
3991         bond_slave_arr_work_rearm(bond, 1);
3992 }
3993
3994 /* Build the usable slaves array in control path for modes that use xmit-hash
3995  * to determine the slave interface -
3996  * (a) BOND_MODE_8023AD
3997  * (b) BOND_MODE_XOR
3998  * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
3999  *
4000  * The caller is expected to hold RTNL only and NO other lock!
4001  */
4002 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
4003 {
4004         struct slave *slave;
4005         struct list_head *iter;
4006         struct bond_up_slave *new_arr, *old_arr;
4007         int agg_id = 0;
4008         int ret = 0;
4009
4010 #ifdef CONFIG_LOCKDEP
4011         WARN_ON(lockdep_is_held(&bond->mode_lock));
4012 #endif
4013
4014         new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
4015                           GFP_KERNEL);
4016         if (!new_arr) {
4017                 ret = -ENOMEM;
4018                 pr_err("Failed to build slave-array.\n");
4019                 goto out;
4020         }
4021         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4022                 struct ad_info ad_info;
4023
4024                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
4025                         pr_debug("bond_3ad_get_active_agg_info failed\n");
4026                         kfree_rcu(new_arr, rcu);
4027                         /* No active aggragator means it's not safe to use
4028                          * the previous array.
4029                          */
4030                         old_arr = rtnl_dereference(bond->slave_arr);
4031                         if (old_arr) {
4032                                 RCU_INIT_POINTER(bond->slave_arr, NULL);
4033                                 kfree_rcu(old_arr, rcu);
4034                         }
4035                         goto out;
4036                 }
4037                 agg_id = ad_info.aggregator_id;
4038         }
4039         bond_for_each_slave(bond, slave, iter) {
4040                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4041                         struct aggregator *agg;
4042
4043                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
4044                         if (!agg || agg->aggregator_identifier != agg_id)
4045                                 continue;
4046                 }
4047                 if (!bond_slave_can_tx(slave))
4048                         continue;
4049                 if (skipslave == slave)
4050                         continue;
4051
4052                 netdev_dbg(bond->dev,
4053                            "Adding slave dev %s to tx hash array[%d]\n",
4054                            slave->dev->name, new_arr->count);
4055
4056                 new_arr->arr[new_arr->count++] = slave;
4057         }
4058
4059         old_arr = rtnl_dereference(bond->slave_arr);
4060         rcu_assign_pointer(bond->slave_arr, new_arr);
4061         if (old_arr)
4062                 kfree_rcu(old_arr, rcu);
4063 out:
4064         if (ret != 0 && skipslave) {
4065                 int idx;
4066
4067                 /* Rare situation where caller has asked to skip a specific
4068                  * slave but allocation failed (most likely!). BTW this is
4069                  * only possible when the call is initiated from
4070                  * __bond_release_one(). In this situation; overwrite the
4071                  * skipslave entry in the array with the last entry from the
4072                  * array to avoid a situation where the xmit path may choose
4073                  * this to-be-skipped slave to send a packet out.
4074                  */
4075                 old_arr = rtnl_dereference(bond->slave_arr);
4076                 for (idx = 0; old_arr != NULL && idx < old_arr->count; idx++) {
4077                         if (skipslave == old_arr->arr[idx]) {
4078                                 old_arr->arr[idx] =
4079                                     old_arr->arr[old_arr->count-1];
4080                                 old_arr->count--;
4081                                 break;
4082                         }
4083                 }
4084         }
4085         return ret;
4086 }
4087
4088 /* Use this Xmit function for 3AD as well as XOR modes. The current
4089  * usable slave array is formed in the control path. The xmit function
4090  * just calculates hash and sends the packet out.
4091  */
4092 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
4093                                      struct net_device *dev)
4094 {
4095         struct bonding *bond = netdev_priv(dev);
4096         struct slave *slave;
4097         struct bond_up_slave *slaves;
4098         unsigned int count;
4099
4100         slaves = rcu_dereference(bond->slave_arr);
4101         count = slaves ? READ_ONCE(slaves->count) : 0;
4102         if (likely(count)) {
4103                 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
4104                 bond_dev_queue_xmit(bond, skb, slave->dev);
4105         } else {
4106                 bond_tx_drop(dev, skb);
4107         }
4108
4109         return NETDEV_TX_OK;
4110 }
4111
4112 /* in broadcast mode, we send everything to all usable interfaces. */
4113 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
4114                                        struct net_device *bond_dev)
4115 {
4116         struct bonding *bond = netdev_priv(bond_dev);
4117         struct slave *slave = NULL;
4118         struct list_head *iter;
4119
4120         bond_for_each_slave_rcu(bond, slave, iter) {
4121                 if (bond_is_last_slave(bond, slave))
4122                         break;
4123                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4124                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4125
4126                         if (!skb2) {
4127                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4128                                                     bond_dev->name, __func__);
4129                                 continue;
4130                         }
4131                         bond_dev_queue_xmit(bond, skb2, slave->dev);
4132                 }
4133         }
4134         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4135                 bond_dev_queue_xmit(bond, skb, slave->dev);
4136         else
4137                 bond_tx_drop(bond_dev, skb);
4138
4139         return NETDEV_TX_OK;
4140 }
4141
4142 /*------------------------- Device initialization ---------------------------*/
4143
4144 /* Lookup the slave that corresponds to a qid */
4145 static inline int bond_slave_override(struct bonding *bond,
4146                                       struct sk_buff *skb)
4147 {
4148         struct slave *slave = NULL;
4149         struct list_head *iter;
4150
4151         if (!skb_rx_queue_recorded(skb))
4152                 return 1;
4153
4154         /* Find out if any slaves have the same mapping as this skb. */
4155         bond_for_each_slave_rcu(bond, slave, iter) {
4156                 if (slave->queue_id == skb_get_queue_mapping(skb)) {
4157                         if (bond_slave_is_up(slave) &&
4158                             slave->link == BOND_LINK_UP) {
4159                                 bond_dev_queue_xmit(bond, skb, slave->dev);
4160                                 return 0;
4161                         }
4162                         /* If the slave isn't UP, use default transmit policy. */
4163                         break;
4164                 }
4165         }
4166
4167         return 1;
4168 }
4169
4170
4171 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4172                              struct net_device *sb_dev,
4173                              select_queue_fallback_t fallback)
4174 {
4175         /* This helper function exists to help dev_pick_tx get the correct
4176          * destination queue.  Using a helper function skips a call to
4177          * skb_tx_hash and will put the skbs in the queue we expect on their
4178          * way down to the bonding driver.
4179          */
4180         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4181
4182         /* Save the original txq to restore before passing to the driver */
4183         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
4184
4185         if (unlikely(txq >= dev->real_num_tx_queues)) {
4186                 do {
4187                         txq -= dev->real_num_tx_queues;
4188                 } while (txq >= dev->real_num_tx_queues);
4189         }
4190         return txq;
4191 }
4192
4193 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4194 {
4195         struct bonding *bond = netdev_priv(dev);
4196
4197         if (bond_should_override_tx_queue(bond) &&
4198             !bond_slave_override(bond, skb))
4199                 return NETDEV_TX_OK;
4200
4201         switch (BOND_MODE(bond)) {
4202         case BOND_MODE_ROUNDROBIN:
4203                 return bond_xmit_roundrobin(skb, dev);
4204         case BOND_MODE_ACTIVEBACKUP:
4205                 return bond_xmit_activebackup(skb, dev);
4206         case BOND_MODE_8023AD:
4207         case BOND_MODE_XOR:
4208                 return bond_3ad_xor_xmit(skb, dev);
4209         case BOND_MODE_BROADCAST:
4210                 return bond_xmit_broadcast(skb, dev);
4211         case BOND_MODE_ALB:
4212                 return bond_alb_xmit(skb, dev);
4213         case BOND_MODE_TLB:
4214                 return bond_tlb_xmit(skb, dev);
4215         default:
4216                 /* Should never happen, mode already checked */
4217                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4218                 WARN_ON_ONCE(1);
4219                 bond_tx_drop(dev, skb);
4220                 return NETDEV_TX_OK;
4221         }
4222 }
4223
4224 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4225 {
4226         struct bonding *bond = netdev_priv(dev);
4227         netdev_tx_t ret = NETDEV_TX_OK;
4228
4229         /* If we risk deadlock from transmitting this in the
4230          * netpoll path, tell netpoll to queue the frame for later tx
4231          */
4232         if (unlikely(is_netpoll_tx_blocked(dev)))
4233                 return NETDEV_TX_BUSY;
4234
4235         rcu_read_lock();
4236         if (bond_has_slaves(bond))
4237                 ret = __bond_start_xmit(skb, dev);
4238         else
4239                 bond_tx_drop(dev, skb);
4240         rcu_read_unlock();
4241
4242         return ret;
4243 }
4244
4245 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed)
4246 {
4247         if (speed == 0 || speed == SPEED_UNKNOWN)
4248                 speed = slave->speed;
4249         else
4250                 speed = min(speed, slave->speed);
4251
4252         return speed;
4253 }
4254
4255 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4256                                            struct ethtool_link_ksettings *cmd)
4257 {
4258         struct bonding *bond = netdev_priv(bond_dev);
4259         struct list_head *iter;
4260         struct slave *slave;
4261         u32 speed = 0;
4262
4263         cmd->base.duplex = DUPLEX_UNKNOWN;
4264         cmd->base.port = PORT_OTHER;
4265
4266         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4267          * do not need to check mode.  Though link speed might not represent
4268          * the true receive or transmit bandwidth (not all modes are symmetric)
4269          * this is an accurate maximum.
4270          */
4271         bond_for_each_slave(bond, slave, iter) {
4272                 if (bond_slave_can_tx(slave)) {
4273                         if (slave->speed != SPEED_UNKNOWN) {
4274                                 if (BOND_MODE(bond) == BOND_MODE_BROADCAST)
4275                                         speed = bond_mode_bcast_speed(slave,
4276                                                                       speed);
4277                                 else
4278                                         speed += slave->speed;
4279                         }
4280                         if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4281                             slave->duplex != DUPLEX_UNKNOWN)
4282                                 cmd->base.duplex = slave->duplex;
4283                 }
4284         }
4285         cmd->base.speed = speed ? : SPEED_UNKNOWN;
4286
4287         return 0;
4288 }
4289
4290 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4291                                      struct ethtool_drvinfo *drvinfo)
4292 {
4293         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4294         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4295         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4296                  BOND_ABI_VERSION);
4297 }
4298
4299 static const struct ethtool_ops bond_ethtool_ops = {
4300         .get_drvinfo            = bond_ethtool_get_drvinfo,
4301         .get_link               = ethtool_op_get_link,
4302         .get_link_ksettings     = bond_ethtool_get_link_ksettings,
4303 };
4304
4305 static const struct net_device_ops bond_netdev_ops = {
4306         .ndo_init               = bond_init,
4307         .ndo_uninit             = bond_uninit,
4308         .ndo_open               = bond_open,
4309         .ndo_stop               = bond_close,
4310         .ndo_start_xmit         = bond_start_xmit,
4311         .ndo_select_queue       = bond_select_queue,
4312         .ndo_get_stats64        = bond_get_stats,
4313         .ndo_do_ioctl           = bond_do_ioctl,
4314         .ndo_change_rx_flags    = bond_change_rx_flags,
4315         .ndo_set_rx_mode        = bond_set_rx_mode,
4316         .ndo_change_mtu         = bond_change_mtu,
4317         .ndo_set_mac_address    = bond_set_mac_address,
4318         .ndo_neigh_setup        = bond_neigh_setup,
4319         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4320         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4321         .ndo_get_lock_subclass  = bond_get_nest_level,
4322 #ifdef CONFIG_NET_POLL_CONTROLLER
4323         .ndo_netpoll_setup      = bond_netpoll_setup,
4324         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4325         .ndo_poll_controller    = bond_poll_controller,
4326 #endif
4327         .ndo_add_slave          = bond_enslave,
4328         .ndo_del_slave          = bond_release,
4329         .ndo_fix_features       = bond_fix_features,
4330         .ndo_features_check     = passthru_features_check,
4331 };
4332
4333 static const struct device_type bond_type = {
4334         .name = "bond",
4335 };
4336
4337 static void bond_destructor(struct net_device *bond_dev)
4338 {
4339         struct bonding *bond = netdev_priv(bond_dev);
4340         if (bond->wq)
4341                 destroy_workqueue(bond->wq);
4342 }
4343
4344 void bond_setup(struct net_device *bond_dev)
4345 {
4346         struct bonding *bond = netdev_priv(bond_dev);
4347
4348         spin_lock_init(&bond->mode_lock);
4349         spin_lock_init(&bond->stats_lock);
4350         bond->params = bonding_defaults;
4351
4352         /* Initialize pointers */
4353         bond->dev = bond_dev;
4354
4355         /* Initialize the device entry points */
4356         ether_setup(bond_dev);
4357         bond_dev->max_mtu = ETH_MAX_MTU;
4358         bond_dev->netdev_ops = &bond_netdev_ops;
4359         bond_dev->ethtool_ops = &bond_ethtool_ops;
4360
4361         bond_dev->needs_free_netdev = true;
4362         bond_dev->priv_destructor = bond_destructor;
4363
4364         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4365
4366         /* Initialize the device options */
4367         bond_dev->flags |= IFF_MASTER;
4368         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4369         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4370
4371         /* don't acquire bond device's netif_tx_lock when transmitting */
4372         bond_dev->features |= NETIF_F_LLTX;
4373
4374         /* By default, we declare the bond to be fully
4375          * VLAN hardware accelerated capable. Special
4376          * care is taken in the various xmit functions
4377          * when there are slaves that are not hw accel
4378          * capable
4379          */
4380
4381         /* Don't allow bond devices to change network namespaces. */
4382         bond_dev->features |= NETIF_F_NETNS_LOCAL;
4383
4384         bond_dev->hw_features = BOND_VLAN_FEATURES |
4385                                 NETIF_F_HW_VLAN_CTAG_RX |
4386                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4387
4388         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4;
4389         bond_dev->features |= bond_dev->hw_features;
4390         bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
4391 }
4392
4393 /* Destroy a bonding device.
4394  * Must be under rtnl_lock when this function is called.
4395  */
4396 static void bond_uninit(struct net_device *bond_dev)
4397 {
4398         struct bonding *bond = netdev_priv(bond_dev);
4399         struct list_head *iter;
4400         struct slave *slave;
4401         struct bond_up_slave *arr;
4402
4403         bond_netpoll_cleanup(bond_dev);
4404
4405         /* Release the bonded slaves */
4406         bond_for_each_slave(bond, slave, iter)
4407                 __bond_release_one(bond_dev, slave->dev, true, true);
4408         netdev_info(bond_dev, "Released all slaves\n");
4409
4410         arr = rtnl_dereference(bond->slave_arr);
4411         if (arr) {
4412                 RCU_INIT_POINTER(bond->slave_arr, NULL);
4413                 kfree_rcu(arr, rcu);
4414         }
4415
4416         list_del(&bond->bond_list);
4417
4418         bond_debug_unregister(bond);
4419 }
4420
4421 /*------------------------- Module initialization ---------------------------*/
4422
4423 static int bond_check_params(struct bond_params *params)
4424 {
4425         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4426         struct bond_opt_value newval;
4427         const struct bond_opt_value *valptr;
4428         int arp_all_targets_value = 0;
4429         u16 ad_actor_sys_prio = 0;
4430         u16 ad_user_port_key = 0;
4431         __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
4432         int arp_ip_count;
4433         int bond_mode   = BOND_MODE_ROUNDROBIN;
4434         int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4435         int lacp_fast = 0;
4436         int tlb_dynamic_lb;
4437
4438         /* Convert string parameters. */
4439         if (mode) {
4440                 bond_opt_initstr(&newval, mode);
4441                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4442                 if (!valptr) {
4443                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4444                         return -EINVAL;
4445                 }
4446                 bond_mode = valptr->value;
4447         }
4448
4449         if (xmit_hash_policy) {
4450                 if (bond_mode == BOND_MODE_ROUNDROBIN ||
4451                     bond_mode == BOND_MODE_ACTIVEBACKUP ||
4452                     bond_mode == BOND_MODE_BROADCAST) {
4453                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4454                                 bond_mode_name(bond_mode));
4455                 } else {
4456                         bond_opt_initstr(&newval, xmit_hash_policy);
4457                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4458                                                 &newval);
4459                         if (!valptr) {
4460                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4461                                        xmit_hash_policy);
4462                                 return -EINVAL;
4463                         }
4464                         xmit_hashtype = valptr->value;
4465                 }
4466         }
4467
4468         if (lacp_rate) {
4469                 if (bond_mode != BOND_MODE_8023AD) {
4470                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4471                                 bond_mode_name(bond_mode));
4472                 } else {
4473                         bond_opt_initstr(&newval, lacp_rate);
4474                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4475                                                 &newval);
4476                         if (!valptr) {
4477                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4478                                        lacp_rate);
4479                                 return -EINVAL;
4480                         }
4481                         lacp_fast = valptr->value;
4482                 }
4483         }
4484
4485         if (ad_select) {
4486                 bond_opt_initstr(&newval, ad_select);
4487                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4488                                         &newval);
4489                 if (!valptr) {
4490                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4491                         return -EINVAL;
4492                 }
4493                 params->ad_select = valptr->value;
4494                 if (bond_mode != BOND_MODE_8023AD)
4495                         pr_warn("ad_select param only affects 802.3ad mode\n");
4496         } else {
4497                 params->ad_select = BOND_AD_STABLE;
4498         }
4499
4500         if (max_bonds < 0) {
4501                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4502                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4503                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4504         }
4505
4506         if (miimon < 0) {
4507                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4508                         miimon, INT_MAX);
4509                 miimon = 0;
4510         }
4511
4512         if (updelay < 0) {
4513                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4514                         updelay, INT_MAX);
4515                 updelay = 0;
4516         }
4517
4518         if (downdelay < 0) {
4519                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4520                         downdelay, INT_MAX);
4521                 downdelay = 0;
4522         }
4523
4524         if ((use_carrier != 0) && (use_carrier != 1)) {
4525                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4526                         use_carrier);
4527                 use_carrier = 1;
4528         }
4529
4530         if (num_peer_notif < 0 || num_peer_notif > 255) {
4531                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4532                         num_peer_notif);
4533                 num_peer_notif = 1;
4534         }
4535
4536         /* reset values for 802.3ad/TLB/ALB */
4537         if (!bond_mode_uses_arp(bond_mode)) {
4538                 if (!miimon) {
4539                         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");
4540                         pr_warn("Forcing miimon to 100msec\n");
4541                         miimon = BOND_DEFAULT_MIIMON;
4542                 }
4543         }
4544
4545         if (tx_queues < 1 || tx_queues > 255) {
4546                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4547                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4548                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4549         }
4550
4551         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4552                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4553                         all_slaves_active);
4554                 all_slaves_active = 0;
4555         }
4556
4557         if (resend_igmp < 0 || resend_igmp > 255) {
4558                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4559                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4560                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4561         }
4562
4563         bond_opt_initval(&newval, packets_per_slave);
4564         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4565                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4566                         packets_per_slave, USHRT_MAX);
4567                 packets_per_slave = 1;
4568         }
4569
4570         if (bond_mode == BOND_MODE_ALB) {
4571                 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",
4572                           updelay);
4573         }
4574
4575         if (!miimon) {
4576                 if (updelay || downdelay) {
4577                         /* just warn the user the up/down delay will have
4578                          * no effect since miimon is zero...
4579                          */
4580                         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",
4581                                 updelay, downdelay);
4582                 }
4583         } else {
4584                 /* don't allow arp monitoring */
4585                 if (arp_interval) {
4586                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4587                                 miimon, arp_interval);
4588                         arp_interval = 0;
4589                 }
4590
4591                 if ((updelay % miimon) != 0) {
4592                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4593                                 updelay, miimon, (updelay / miimon) * miimon);
4594                 }
4595
4596                 updelay /= miimon;
4597
4598                 if ((downdelay % miimon) != 0) {
4599                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4600                                 downdelay, miimon,
4601                                 (downdelay / miimon) * miimon);
4602                 }
4603
4604                 downdelay /= miimon;
4605         }
4606
4607         if (arp_interval < 0) {
4608                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4609                         arp_interval, INT_MAX);
4610                 arp_interval = 0;
4611         }
4612
4613         for (arp_ip_count = 0, i = 0;
4614              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4615                 __be32 ip;
4616
4617                 /* not a complete check, but good enough to catch mistakes */
4618                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4619                     !bond_is_ip_target_ok(ip)) {
4620                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4621                                 arp_ip_target[i]);
4622                         arp_interval = 0;
4623                 } else {
4624                         if (bond_get_targets_ip(arp_target, ip) == -1)
4625                                 arp_target[arp_ip_count++] = ip;
4626                         else
4627                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4628                                         &ip);
4629                 }
4630         }
4631
4632         if (arp_interval && !arp_ip_count) {
4633                 /* don't allow arping if no arp_ip_target given... */
4634                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4635                         arp_interval);
4636                 arp_interval = 0;
4637         }
4638
4639         if (arp_validate) {
4640                 if (!arp_interval) {
4641                         pr_err("arp_validate requires arp_interval\n");
4642                         return -EINVAL;
4643                 }
4644
4645                 bond_opt_initstr(&newval, arp_validate);
4646                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4647                                         &newval);
4648                 if (!valptr) {
4649                         pr_err("Error: invalid arp_validate \"%s\"\n",
4650                                arp_validate);
4651                         return -EINVAL;
4652                 }
4653                 arp_validate_value = valptr->value;
4654         } else {
4655                 arp_validate_value = 0;
4656         }
4657
4658         if (arp_all_targets) {
4659                 bond_opt_initstr(&newval, arp_all_targets);
4660                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4661                                         &newval);
4662                 if (!valptr) {
4663                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4664                                arp_all_targets);
4665                         arp_all_targets_value = 0;
4666                 } else {
4667                         arp_all_targets_value = valptr->value;
4668                 }
4669         }
4670
4671         if (miimon) {
4672                 pr_info("MII link monitoring set to %d ms\n", miimon);
4673         } else if (arp_interval) {
4674                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4675                                           arp_validate_value);
4676                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4677                         arp_interval, valptr->string, arp_ip_count);
4678
4679                 for (i = 0; i < arp_ip_count; i++)
4680                         pr_cont(" %s", arp_ip_target[i]);
4681
4682                 pr_cont("\n");
4683
4684         } else if (max_bonds) {
4685                 /* miimon and arp_interval not set, we need one so things
4686                  * work as expected, see bonding.txt for details
4687                  */
4688                 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");
4689         }
4690
4691         if (primary && !bond_mode_uses_primary(bond_mode)) {
4692                 /* currently, using a primary only makes sense
4693                  * in active backup, TLB or ALB modes
4694                  */
4695                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4696                         primary, bond_mode_name(bond_mode));
4697                 primary = NULL;
4698         }
4699
4700         if (primary && primary_reselect) {
4701                 bond_opt_initstr(&newval, primary_reselect);
4702                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4703                                         &newval);
4704                 if (!valptr) {
4705                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4706                                primary_reselect);
4707                         return -EINVAL;
4708                 }
4709                 primary_reselect_value = valptr->value;
4710         } else {
4711                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4712         }
4713
4714         if (fail_over_mac) {
4715                 bond_opt_initstr(&newval, fail_over_mac);
4716                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4717                                         &newval);
4718                 if (!valptr) {
4719                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4720                                fail_over_mac);
4721                         return -EINVAL;
4722                 }
4723                 fail_over_mac_value = valptr->value;
4724                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4725                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4726         } else {
4727                 fail_over_mac_value = BOND_FOM_NONE;
4728         }
4729
4730         bond_opt_initstr(&newval, "default");
4731         valptr = bond_opt_parse(
4732                         bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4733                                      &newval);
4734         if (!valptr) {
4735                 pr_err("Error: No ad_actor_sys_prio default value");
4736                 return -EINVAL;
4737         }
4738         ad_actor_sys_prio = valptr->value;
4739
4740         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4741                                 &newval);
4742         if (!valptr) {
4743                 pr_err("Error: No ad_user_port_key default value");
4744                 return -EINVAL;
4745         }
4746         ad_user_port_key = valptr->value;
4747
4748         bond_opt_initstr(&newval, "default");
4749         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
4750         if (!valptr) {
4751                 pr_err("Error: No tlb_dynamic_lb default value");
4752                 return -EINVAL;
4753         }
4754         tlb_dynamic_lb = valptr->value;
4755
4756         if (lp_interval == 0) {
4757                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4758                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4759                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4760         }
4761
4762         /* fill params struct with the proper values */
4763         params->mode = bond_mode;
4764         params->xmit_policy = xmit_hashtype;
4765         params->miimon = miimon;
4766         params->num_peer_notif = num_peer_notif;
4767         params->arp_interval = arp_interval;
4768         params->arp_validate = arp_validate_value;
4769         params->arp_all_targets = arp_all_targets_value;
4770         params->updelay = updelay;
4771         params->downdelay = downdelay;
4772         params->use_carrier = use_carrier;
4773         params->lacp_fast = lacp_fast;
4774         params->primary[0] = 0;
4775         params->primary_reselect = primary_reselect_value;
4776         params->fail_over_mac = fail_over_mac_value;
4777         params->tx_queues = tx_queues;
4778         params->all_slaves_active = all_slaves_active;
4779         params->resend_igmp = resend_igmp;
4780         params->min_links = min_links;
4781         params->lp_interval = lp_interval;
4782         params->packets_per_slave = packets_per_slave;
4783         params->tlb_dynamic_lb = tlb_dynamic_lb;
4784         params->ad_actor_sys_prio = ad_actor_sys_prio;
4785         eth_zero_addr(params->ad_actor_system);
4786         params->ad_user_port_key = ad_user_port_key;
4787         if (packets_per_slave > 0) {
4788                 params->reciprocal_packets_per_slave =
4789                         reciprocal_value(packets_per_slave);
4790         } else {
4791                 /* reciprocal_packets_per_slave is unused if
4792                  * packets_per_slave is 0 or 1, just initialize it
4793                  */
4794                 params->reciprocal_packets_per_slave =
4795                         (struct reciprocal_value) { 0 };
4796         }
4797
4798         if (primary) {
4799                 strncpy(params->primary, primary, IFNAMSIZ);
4800                 params->primary[IFNAMSIZ - 1] = 0;
4801         }
4802
4803         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4804
4805         return 0;
4806 }
4807
4808 /* Called from registration process */
4809 static int bond_init(struct net_device *bond_dev)
4810 {
4811         struct bonding *bond = netdev_priv(bond_dev);
4812         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4813
4814         netdev_dbg(bond_dev, "Begin bond_init\n");
4815
4816         bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
4817         if (!bond->wq)
4818                 return -ENOMEM;
4819
4820         bond->nest_level = SINGLE_DEPTH_NESTING;
4821         netdev_lockdep_set_classes(bond_dev);
4822
4823         list_add_tail(&bond->bond_list, &bn->dev_list);
4824
4825         bond_prepare_sysfs_group(bond);
4826
4827         bond_debug_register(bond);
4828
4829         /* Ensure valid dev_addr */
4830         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4831             bond_dev->addr_assign_type == NET_ADDR_PERM)
4832                 eth_hw_addr_random(bond_dev);
4833
4834         return 0;
4835 }
4836
4837 unsigned int bond_get_num_tx_queues(void)
4838 {
4839         return tx_queues;
4840 }
4841
4842 /* Create a new bond based on the specified name and bonding parameters.
4843  * If name is NULL, obtain a suitable "bond%d" name for us.
4844  * Caller must NOT hold rtnl_lock; we need to release it here before we
4845  * set up our sysfs entries.
4846  */
4847 int bond_create(struct net *net, const char *name)
4848 {
4849         struct net_device *bond_dev;
4850         struct bonding *bond;
4851         struct alb_bond_info *bond_info;
4852         int res;
4853
4854         rtnl_lock();
4855
4856         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4857                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
4858                                    bond_setup, tx_queues);
4859         if (!bond_dev) {
4860                 pr_err("%s: eek! can't alloc netdev!\n", name);
4861                 rtnl_unlock();
4862                 return -ENOMEM;
4863         }
4864
4865         /*
4866          * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4867          * It is set to 0 by default which is wrong.
4868          */
4869         bond = netdev_priv(bond_dev);
4870         bond_info = &(BOND_ALB_INFO(bond));
4871         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4872
4873         dev_net_set(bond_dev, net);
4874         bond_dev->rtnl_link_ops = &bond_link_ops;
4875
4876         res = register_netdevice(bond_dev);
4877         if (res < 0) {
4878                 free_netdev(bond_dev);
4879                 rtnl_unlock();
4880
4881                 return res;
4882         }
4883
4884         netif_carrier_off(bond_dev);
4885
4886         bond_work_init_all(bond);
4887
4888         rtnl_unlock();
4889         return 0;
4890 }
4891
4892 static int __net_init bond_net_init(struct net *net)
4893 {
4894         struct bond_net *bn = net_generic(net, bond_net_id);
4895
4896         bn->net = net;
4897         INIT_LIST_HEAD(&bn->dev_list);
4898
4899         bond_create_proc_dir(bn);
4900         bond_create_sysfs(bn);
4901
4902         return 0;
4903 }
4904
4905 static void __net_exit bond_net_exit(struct net *net)
4906 {
4907         struct bond_net *bn = net_generic(net, bond_net_id);
4908         struct bonding *bond, *tmp_bond;
4909         LIST_HEAD(list);
4910
4911         bond_destroy_sysfs(bn);
4912
4913         /* Kill off any bonds created after unregistering bond rtnl ops */
4914         rtnl_lock();
4915         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4916                 unregister_netdevice_queue(bond->dev, &list);
4917         unregister_netdevice_many(&list);
4918         rtnl_unlock();
4919
4920         bond_destroy_proc_dir(bn);
4921 }
4922
4923 static struct pernet_operations bond_net_ops = {
4924         .init = bond_net_init,
4925         .exit = bond_net_exit,
4926         .id   = &bond_net_id,
4927         .size = sizeof(struct bond_net),
4928 };
4929
4930 static int __init bonding_init(void)
4931 {
4932         int i;
4933         int res;
4934
4935         pr_info("%s", bond_version);
4936
4937         res = bond_check_params(&bonding_defaults);
4938         if (res)
4939                 goto out;
4940
4941         res = register_pernet_subsys(&bond_net_ops);
4942         if (res)
4943                 goto out;
4944
4945         res = bond_netlink_init();
4946         if (res)
4947                 goto err_link;
4948
4949         bond_create_debugfs();
4950
4951         for (i = 0; i < max_bonds; i++) {
4952                 res = bond_create(&init_net, NULL);
4953                 if (res)
4954                         goto err;
4955         }
4956
4957         register_netdevice_notifier(&bond_netdev_notifier);
4958 out:
4959         return res;
4960 err:
4961         bond_destroy_debugfs();
4962         bond_netlink_fini();
4963 err_link:
4964         unregister_pernet_subsys(&bond_net_ops);
4965         goto out;
4966
4967 }
4968
4969 static void __exit bonding_exit(void)
4970 {
4971         unregister_netdevice_notifier(&bond_netdev_notifier);
4972
4973         bond_destroy_debugfs();
4974
4975         bond_netlink_fini();
4976         unregister_pernet_subsys(&bond_net_ops);
4977
4978 #ifdef CONFIG_NET_POLL_CONTROLLER
4979         /* Make sure we don't have an imbalance on our netpoll blocking */
4980         WARN_ON(atomic_read(&netpoll_block_tx));
4981 #endif
4982 }
4983
4984 module_init(bonding_init);
4985 module_exit(bonding_exit);
4986 MODULE_LICENSE("GPL");
4987 MODULE_VERSION(DRV_VERSION);
4988 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4989 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");