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_keys.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 /*---------------------------- Module parameters ----------------------------*/
  86
  87 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  88
  89 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  90 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  91 static int num_peer_notif = 1;
  92 static int miimon;
  93 static int updelay;
  94 static int downdelay;
  95 static int use_carrier  = 1;
  96 static char *mode;
  97 static char *primary;
  98 static char *primary_reselect;
  99 static char *lacp_rate;
 100 static int min_links;
 101 static char *ad_select;
 102 static char *xmit_hash_policy;
 103 static int arp_interval;
 104 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 105 static char *arp_validate;
 106 static char *arp_all_targets;
 107 static char *fail_over_mac;
 108 static int all_slaves_active;
 109 static struct bond_params bonding_defaults;
 110 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 111 static int packets_per_slave = 1;
 112 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
 113
 114 module_param(max_bonds, int, 0);
 115 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 116 module_param(tx_queues, int, 0);
 117 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 118 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 119 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 120                                "failover event (alias of num_unsol_na)");
 121 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 122 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 123                                "failover event (alias of num_grat_arp)");
 124 module_param(miimon, int, 0);
 125 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 126 module_param(updelay, int, 0);
 127 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 128 module_param(downdelay, int, 0);
 129 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 130                             "in milliseconds");
 131 module_param(use_carrier, int, 0);
 132 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 133                               "0 for off, 1 for on (default)");
 134 module_param(mode, charp, 0);
 135 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 136                        "1 for active-backup, 2 for balance-xor, "
 137                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 138                        "6 for balance-alb");
 139 module_param(primary, charp, 0);
 140 MODULE_PARM_DESC(primary, "Primary network device to use");
 141 module_param(primary_reselect, charp, 0);
 142 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 143                                    "once it comes up; "
 144                                    "0 for always (default), "
 145                                    "1 for only if speed of primary is "
 146                                    "better, "
 147                                    "2 for only on active slave "
 148                                    "failure");
 149 module_param(lacp_rate, charp, 0);
 150 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 151                             "0 for slow, 1 for fast");
 152 module_param(ad_select, charp, 0);
 153 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 154                             "0 for stable (default), 1 for bandwidth, "
 155                             "2 for count");
 156 module_param(min_links, int, 0);
 157 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 158
 159 module_param(xmit_hash_policy, charp, 0);
 160 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 161                                    "0 for layer 2 (default), 1 for layer 3+4, "
 162                                    "2 for layer 2+3, 3 for encap layer 2+3, "
 163                                    "4 for encap layer 3+4");
 164 module_param(arp_interval, int, 0);
 165 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 166 module_param_array(arp_ip_target, charp, NULL, 0);
 167 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 168 module_param(arp_validate, charp, 0);
 169 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 170                                "0 for none (default), 1 for active, "
 171                                "2 for backup, 3 for all");
 172 module_param(arp_all_targets, charp, 0);
 173 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 174 module_param(fail_over_mac, charp, 0);
 175 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 176                                 "the same MAC; 0 for none (default), "
 177                                 "1 for active, 2 for follow");
 178 module_param(all_slaves_active, int, 0);
 179 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
 180                                      "by setting active flag for all slaves; "
 181                                      "0 for never (default), 1 for always.");
 182 module_param(resend_igmp, int, 0);
 183 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 184                               "link failure");
 185 module_param(packets_per_slave, int, 0);
 186 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
 187                                     "mode; 0 for a random slave, 1 packet per "
 188                                     "slave (default), >1 packets per slave.");
 189 module_param(lp_interval, uint, 0);
 190 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
 191                               "the bonding driver sends learning packets to "
 192                               "each slaves peer switch. The default is 1.");
 193
 194 /*----------------------------- Global variables ----------------------------*/
 195
 196 #ifdef CONFIG_NET_POLL_CONTROLLER
 197 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 198 #endif
 199
 200 int bond_net_id __read_mostly;
 201
 202 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 203 static int arp_ip_count;
 204 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 205 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 206 static int lacp_fast;
 207
 208 /*-------------------------- Forward declarations ---------------------------*/
 209
 210 static int bond_init(struct net_device *bond_dev);
 211 static void bond_uninit(struct net_device *bond_dev);
 212 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
 213                                                 struct rtnl_link_stats64 *stats);
 214 static void bond_slave_arr_handler(struct work_struct *work);
 215
 216 /*---------------------------- General routines -----------------------------*/
 217
 218 const char *bond_mode_name(int mode)
 219 {
 220         static const char *names[] = {
 221                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 222                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 223                 [BOND_MODE_XOR] = "load balancing (xor)",
 224                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 225                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 226                 [BOND_MODE_TLB] = "transmit load balancing",
 227                 [BOND_MODE_ALB] = "adaptive load balancing",
 228         };
 229
 230         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 231                 return "unknown";
 232
 233         return names[mode];
 234 }
 235
 236 /*---------------------------------- VLAN -----------------------------------*/
 237
 238 /**
 239  * bond_dev_queue_xmit - Prepare skb for xmit.
 240  *
 241  * @bond: bond device that got this skb for tx.
 242  * @skb: hw accel VLAN tagged skb to transmit
 243  * @slave_dev: slave that is supposed to xmit this skbuff
 244  */
 245 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 246                         struct net_device *slave_dev)
 247 {
 248         skb->dev = slave_dev;
 249
 250         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 251                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 252         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 253
 254         if (unlikely(netpoll_tx_running(bond->dev)))
 255                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 256         else
 257                 dev_queue_xmit(skb);
 258 }
 259
 260 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 261  * We don't protect the slave list iteration with a lock because:
 262  * a. This operation is performed in IOCTL context,
 263  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 264  * c. Holding a lock with BH disabled while directly calling a base driver
 265  *    entry point is generally a BAD idea.
 266  *
 267  * The design of synchronization/protection for this operation in the 8021q
 268  * module is good for one or more VLAN devices over a single physical device
 269  * and cannot be extended for a teaming solution like bonding, so there is a
 270  * potential race condition here where a net device from the vlan group might
 271  * be referenced (either by a base driver or the 8021q code) while it is being
 272  * removed from the system. However, it turns out we're not making matters
 273  * worse, and if it works for regular VLAN usage it will work here too.
 274 */
 275
 276 /**
 277  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 278  * @bond_dev: bonding net device that got called
 279  * @vid: vlan id being added
 280  */
 281 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 282                                 __be16 proto, u16 vid)
 283 {
 284         struct bonding *bond = netdev_priv(bond_dev);
 285         struct slave *slave, *rollback_slave;
 286         struct list_head *iter;
 287         int res;
 288
 289         bond_for_each_slave(bond, slave, iter) {
 290                 res = vlan_vid_add(slave->dev, proto, vid);
 291                 if (res)
 292                         goto unwind;
 293         }
 294
 295         return 0;
 296
 297 unwind:
 298         /* unwind to the slave that failed */
 299         bond_for_each_slave(bond, rollback_slave, iter) {
 300                 if (rollback_slave == slave)
 301                         break;
 302
 303                 vlan_vid_del(rollback_slave->dev, proto, vid);
 304         }
 305
 306         return res;
 307 }
 308
 309 /**
 310  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 311  * @bond_dev: bonding net device that got called
 312  * @vid: vlan id being removed
 313  */
 314 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 315                                  __be16 proto, u16 vid)
 316 {
 317         struct bonding *bond = netdev_priv(bond_dev);
 318         struct list_head *iter;
 319         struct slave *slave;
 320
 321         bond_for_each_slave(bond, slave, iter)
 322                 vlan_vid_del(slave->dev, proto, vid);
 323
 324         if (bond_is_lb(bond))
 325                 bond_alb_clear_vlan(bond, vid);
 326
 327         return 0;
 328 }
 329
 330 /*------------------------------- Link status -------------------------------*/
 331
 332 /* Set the carrier state for the master according to the state of its
 333  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 334  * do special 802.3ad magic.
 335  *
 336  * Returns zero if carrier state does not change, nonzero if it does.
 337  */
 338 int bond_set_carrier(struct bonding *bond)
 339 {
 340         struct list_head *iter;
 341         struct slave *slave;
 342
 343         if (!bond_has_slaves(bond))
 344                 goto down;
 345
 346         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 347                 return bond_3ad_set_carrier(bond);
 348
 349         bond_for_each_slave(bond, slave, iter) {
 350                 if (slave->link == BOND_LINK_UP) {
 351                         if (!netif_carrier_ok(bond->dev)) {
 352                                 netif_carrier_on(bond->dev);
 353                                 return 1;
 354                         }
 355                         return 0;
 356                 }
 357         }
 358
 359 down:
 360         if (netif_carrier_ok(bond->dev)) {
 361                 netif_carrier_off(bond->dev);
 362                 return 1;
 363         }
 364         return 0;
 365 }
 366
 367 /* Get link speed and duplex from the slave's base driver
 368  * using ethtool. If for some reason the call fails or the
 369  * values are invalid, set speed and duplex to -1,
 370  * and return.
 371  */
 372 static void bond_update_speed_duplex(struct slave *slave)
 373 {
 374         struct net_device *slave_dev = slave->dev;
 375         struct ethtool_cmd ecmd;
 376         u32 slave_speed;
 377         int res;
 378
 379         slave->speed = SPEED_UNKNOWN;
 380         slave->duplex = DUPLEX_UNKNOWN;
 381
 382         res = __ethtool_get_settings(slave_dev, &ecmd);
 383         if (res < 0)
 384                 return;
 385
 386         slave_speed = ethtool_cmd_speed(&ecmd);
 387         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 388                 return;
 389
 390         switch (ecmd.duplex) {
 391         case DUPLEX_FULL:
 392         case DUPLEX_HALF:
 393                 break;
 394         default:
 395                 return;
 396         }
 397
 398         slave->speed = slave_speed;
 399         slave->duplex = ecmd.duplex;
 400
 401         return;
 402 }
 403
 404 const char *bond_slave_link_status(s8 link)
 405 {
 406         switch (link) {
 407         case BOND_LINK_UP:
 408                 return "up";
 409         case BOND_LINK_FAIL:
 410                 return "going down";
 411         case BOND_LINK_DOWN:
 412                 return "down";
 413         case BOND_LINK_BACK:
 414                 return "going back";
 415         default:
 416                 return "unknown";
 417         }
 418 }
 419
 420 /* if <dev> supports MII link status reporting, check its link status.
 421  *
 422  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 423  * depending upon the setting of the use_carrier parameter.
 424  *
 425  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 426  * can't tell and just pretend it is), or 0, meaning that the link is
 427  * down.
 428  *
 429  * If reporting is non-zero, instead of faking link up, return -1 if
 430  * both ETHTOOL and MII ioctls fail (meaning the device does not
 431  * support them).  If use_carrier is set, return whatever it says.
 432  * It'd be nice if there was a good way to tell if a driver supports
 433  * netif_carrier, but there really isn't.
 434  */
 435 static int bond_check_dev_link(struct bonding *bond,
 436                                struct net_device *slave_dev, int reporting)
 437 {
 438         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 439         int (*ioctl)(struct net_device *, struct ifreq *, int);
 440         struct ifreq ifr;
 441         struct mii_ioctl_data *mii;
 442
 443         if (!reporting && !netif_running(slave_dev))
 444                 return 0;
 445
 446         if (bond->params.use_carrier)
 447                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 448
 449         /* Try to get link status using Ethtool first. */
 450         if (slave_dev->ethtool_ops->get_link)
 451                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 452                         BMSR_LSTATUS : 0;
 453
 454         /* Ethtool can't be used, fallback to MII ioctls. */
 455         ioctl = slave_ops->ndo_do_ioctl;
 456         if (ioctl) {
 457                 /* TODO: set pointer to correct ioctl on a per team member
 458                  *       bases to make this more efficient. that is, once
 459                  *       we determine the correct ioctl, we will always
 460                  *       call it and not the others for that team
 461                  *       member.
 462                  */
 463
 464                 /* We cannot assume that SIOCGMIIPHY will also read a
 465                  * register; not all network drivers (e.g., e100)
 466                  * support that.
 467                  */
 468
 469                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 470                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 471                 mii = if_mii(&ifr);
 472                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 473                         mii->reg_num = MII_BMSR;
 474                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 475                                 return mii->val_out & BMSR_LSTATUS;
 476                 }
 477         }
 478
 479         /* If reporting, report that either there's no dev->do_ioctl,
 480          * or both SIOCGMIIREG and get_link failed (meaning that we
 481          * cannot report link status).  If not reporting, pretend
 482          * we're ok.
 483          */
 484         return reporting ? -1 : BMSR_LSTATUS;
 485 }
 486
 487 /*----------------------------- Multicast list ------------------------------*/
 488
 489 /* Push the promiscuity flag down to appropriate slaves */
 490 static int bond_set_promiscuity(struct bonding *bond, int inc)
 491 {
 492         struct list_head *iter;
 493         int err = 0;
 494
 495         if (bond_uses_primary(bond)) {
 496                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 497
 498                 if (curr_active)
 499                         err = dev_set_promiscuity(curr_active->dev, inc);
 500         } else {
 501                 struct slave *slave;
 502
 503                 bond_for_each_slave(bond, slave, iter) {
 504                         err = dev_set_promiscuity(slave->dev, inc);
 505                         if (err)
 506                                 return err;
 507                 }
 508         }
 509         return err;
 510 }
 511
 512 /* Push the allmulti flag down to all slaves */
 513 static int bond_set_allmulti(struct bonding *bond, int inc)
 514 {
 515         struct list_head *iter;
 516         int err = 0;
 517
 518         if (bond_uses_primary(bond)) {
 519                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 520
 521                 if (curr_active)
 522                         err = dev_set_allmulti(curr_active->dev, inc);
 523         } else {
 524                 struct slave *slave;
 525
 526                 bond_for_each_slave(bond, slave, iter) {
 527                         err = dev_set_allmulti(slave->dev, inc);
 528                         if (err)
 529                                 return err;
 530                 }
 531         }
 532         return err;
 533 }
 534
 535 /* Retrieve the list of registered multicast addresses for the bonding
 536  * device and retransmit an IGMP JOIN request to the current active
 537  * slave.
 538  */
 539 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 540 {
 541         struct bonding *bond = container_of(work, struct bonding,
 542                                             mcast_work.work);
 543
 544         if (!rtnl_trylock()) {
 545                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 546                 return;
 547         }
 548         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 549
 550         if (bond->igmp_retrans > 1) {
 551                 bond->igmp_retrans--;
 552                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 553         }
 554         rtnl_unlock();
 555 }
 556
 557 /* Flush bond's hardware addresses from slave */
 558 static void bond_hw_addr_flush(struct net_device *bond_dev,
 559                                struct net_device *slave_dev)
 560 {
 561         struct bonding *bond = netdev_priv(bond_dev);
 562
 563         dev_uc_unsync(slave_dev, bond_dev);
 564         dev_mc_unsync(slave_dev, bond_dev);
 565
 566         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 567                 /* del lacpdu mc addr from mc list */
 568                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 569
 570                 dev_mc_del(slave_dev, lacpdu_multicast);
 571         }
 572 }
 573
 574 /*--------------------------- Active slave change ---------------------------*/
 575
 576 /* Update the hardware address list and promisc/allmulti for the new and
 577  * old active slaves (if any).  Modes that are not using primary keep all
 578  * slaves up date at all times; only the modes that use primary need to call
 579  * this function to swap these settings during a failover.
 580  */
 581 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 582                               struct slave *old_active)
 583 {
 584         if (old_active) {
 585                 if (bond->dev->flags & IFF_PROMISC)
 586                         dev_set_promiscuity(old_active->dev, -1);
 587
 588                 if (bond->dev->flags & IFF_ALLMULTI)
 589                         dev_set_allmulti(old_active->dev, -1);
 590
 591                 bond_hw_addr_flush(bond->dev, old_active->dev);
 592         }
 593
 594         if (new_active) {
 595                 /* FIXME: Signal errors upstream. */
 596                 if (bond->dev->flags & IFF_PROMISC)
 597                         dev_set_promiscuity(new_active->dev, 1);
 598
 599                 if (bond->dev->flags & IFF_ALLMULTI)
 600                         dev_set_allmulti(new_active->dev, 1);
 601
 602                 netif_addr_lock_bh(bond->dev);
 603                 dev_uc_sync(new_active->dev, bond->dev);
 604                 dev_mc_sync(new_active->dev, bond->dev);
 605                 netif_addr_unlock_bh(bond->dev);
 606         }
 607 }
 608
 609 /**
 610  * bond_set_dev_addr - clone slave's address to bond
 611  * @bond_dev: bond net device
 612  * @slave_dev: slave net device
 613  *
 614  * Should be called with RTNL held.
 615  */
 616 static void bond_set_dev_addr(struct net_device *bond_dev,
 617                               struct net_device *slave_dev)
 618 {
 619         netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 620                    bond_dev, slave_dev, slave_dev->addr_len);
 621         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 622         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 623         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 624 }
 625
 626 /* bond_do_fail_over_mac
 627  *
 628  * Perform special MAC address swapping for fail_over_mac settings
 629  *
 630  * Called with RTNL
 631  */
 632 static void bond_do_fail_over_mac(struct bonding *bond,
 633                                   struct slave *new_active,
 634                                   struct slave *old_active)
 635 {
 636         u8 tmp_mac[ETH_ALEN];
 637         struct sockaddr saddr;
 638         int rv;
 639
 640         switch (bond->params.fail_over_mac) {
 641         case BOND_FOM_ACTIVE:
 642                 if (new_active)
 643                         bond_set_dev_addr(bond->dev, new_active->dev);
 644                 break;
 645         case BOND_FOM_FOLLOW:
 646                 /* if new_active && old_active, swap them
 647                  * if just old_active, do nothing (going to no active slave)
 648                  * if just new_active, set new_active to bond's MAC
 649                  */
 650                 if (!new_active)
 651                         return;
 652
 653                 if (old_active) {
 654                         ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
 655                         ether_addr_copy(saddr.sa_data,
 656                                         old_active->dev->dev_addr);
 657                         saddr.sa_family = new_active->dev->type;
 658                 } else {
 659                         ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
 660                         saddr.sa_family = bond->dev->type;
 661                 }
 662
 663                 rv = dev_set_mac_address(new_active->dev, &saddr);
 664                 if (rv) {
 665                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 666                                    -rv, new_active->dev->name);
 667                         goto out;
 668                 }
 669
 670                 if (!old_active)
 671                         goto out;
 672
 673                 ether_addr_copy(saddr.sa_data, tmp_mac);
 674                 saddr.sa_family = old_active->dev->type;
 675
 676                 rv = dev_set_mac_address(old_active->dev, &saddr);
 677                 if (rv)
 678                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 679                                    -rv, new_active->dev->name);
 680 out:
 681                 break;
 682         default:
 683                 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
 684                            bond->params.fail_over_mac);
 685                 break;
 686         }
 687
 688 }
 689
 690 static bool bond_should_change_active(struct bonding *bond)
 691 {
 692         struct slave *prim = rtnl_dereference(bond->primary_slave);
 693         struct slave *curr = rtnl_dereference(bond->curr_active_slave);
 694
 695         if (!prim || !curr || curr->link != BOND_LINK_UP)
 696                 return true;
 697         if (bond->force_primary) {
 698                 bond->force_primary = false;
 699                 return true;
 700         }
 701         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 702             (prim->speed < curr->speed ||
 703              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 704                 return false;
 705         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 706                 return false;
 707         return true;
 708 }
 709
 710 /**
 711  * find_best_interface - select the best available slave to be the active one
 712  * @bond: our bonding struct
 713  */
 714 static struct slave *bond_find_best_slave(struct bonding *bond)
 715 {
 716         struct slave *slave, *bestslave = NULL, *primary;
 717         struct list_head *iter;
 718         int mintime = bond->params.updelay;
 719
 720         primary = rtnl_dereference(bond->primary_slave);
 721         if (primary && primary->link == BOND_LINK_UP &&
 722             bond_should_change_active(bond))
 723                 return primary;
 724
 725         bond_for_each_slave(bond, slave, iter) {
 726                 if (slave->link == BOND_LINK_UP)
 727                         return slave;
 728                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
 729                     slave->delay < mintime) {
 730                         mintime = slave->delay;
 731                         bestslave = slave;
 732                 }
 733         }
 734
 735         return bestslave;
 736 }
 737
 738 static bool bond_should_notify_peers(struct bonding *bond)
 739 {
 740         struct slave *slave;
 741
 742         rcu_read_lock();
 743         slave = rcu_dereference(bond->curr_active_slave);
 744         rcu_read_unlock();
 745
 746         netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
 747                    slave ? slave->dev->name : "NULL");
 748
 749         if (!slave || !bond->send_peer_notif ||
 750             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 751                 return false;
 752
 753         return true;
 754 }
 755
 756 /**
 757  * change_active_interface - change the active slave into the specified one
 758  * @bond: our bonding struct
 759  * @new: the new slave to make the active one
 760  *
 761  * Set the new slave to the bond's settings and unset them on the old
 762  * curr_active_slave.
 763  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 764  *
 765  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 766  * because it is apparently the best available slave we have, even though its
 767  * updelay hasn't timed out yet.
 768  *
 769  * Caller must hold RTNL.
 770  */
 771 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 772 {
 773         struct slave *old_active;
 774
 775         ASSERT_RTNL();
 776
 777         old_active = rtnl_dereference(bond->curr_active_slave);
 778
 779         if (old_active == new_active)
 780                 return;
 781
 782         if (new_active) {
 783                 new_active->last_link_up = jiffies;
 784
 785                 if (new_active->link == BOND_LINK_BACK) {
 786                         if (bond_uses_primary(bond)) {
 787                                 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
 788                                             new_active->dev->name,
 789                                             (bond->params.updelay - new_active->delay) * bond->params.miimon);
 790                         }
 791
 792                         new_active->delay = 0;
 793                         bond_set_slave_link_state(new_active, BOND_LINK_UP);
 794
 795                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 796                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 797
 798                         if (bond_is_lb(bond))
 799                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 800                 } else {
 801                         if (bond_uses_primary(bond)) {
 802                                 netdev_info(bond->dev, "making interface %s the new active one\n",
 803                                             new_active->dev->name);
 804                         }
 805                 }
 806         }
 807
 808         if (bond_uses_primary(bond))
 809                 bond_hw_addr_swap(bond, new_active, old_active);
 810
 811         if (bond_is_lb(bond)) {
 812                 bond_alb_handle_active_change(bond, new_active);
 813                 if (old_active)
 814                         bond_set_slave_inactive_flags(old_active,
 815                                                       BOND_SLAVE_NOTIFY_NOW);
 816                 if (new_active)
 817                         bond_set_slave_active_flags(new_active,
 818                                                     BOND_SLAVE_NOTIFY_NOW);
 819         } else {
 820                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 821         }
 822
 823         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
 824                 if (old_active)
 825                         bond_set_slave_inactive_flags(old_active,
 826                                                       BOND_SLAVE_NOTIFY_NOW);
 827
 828                 if (new_active) {
 829                         bool should_notify_peers = false;
 830
 831                         bond_set_slave_active_flags(new_active,
 832                                                     BOND_SLAVE_NOTIFY_NOW);
 833
 834                         if (bond->params.fail_over_mac)
 835                                 bond_do_fail_over_mac(bond, new_active,
 836                                                       old_active);
 837
 838                         if (netif_running(bond->dev)) {
 839                                 bond->send_peer_notif =
 840                                         bond->params.num_peer_notif;
 841                                 should_notify_peers =
 842                                         bond_should_notify_peers(bond);
 843                         }
 844
 845                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 846                         if (should_notify_peers)
 847                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 848                                                          bond->dev);
 849                 }
 850         }
 851
 852         /* resend IGMP joins since active slave has changed or
 853          * all were sent on curr_active_slave.
 854          * resend only if bond is brought up with the affected
 855          * bonding modes and the retransmission is enabled
 856          */
 857         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 858             ((bond_uses_primary(bond) && new_active) ||
 859              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
 860                 bond->igmp_retrans = bond->params.resend_igmp;
 861                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 862         }
 863 }
 864
 865 /**
 866  * bond_select_active_slave - select a new active slave, if needed
 867  * @bond: our bonding struct
 868  *
 869  * This functions should be called when one of the following occurs:
 870  * - The old curr_active_slave has been released or lost its link.
 871  * - The primary_slave has got its link back.
 872  * - A slave has got its link back and there's no old curr_active_slave.
 873  *
 874  * Caller must hold RTNL.
 875  */
 876 void bond_select_active_slave(struct bonding *bond)
 877 {
 878         struct slave *best_slave;
 879         int rv;
 880
 881         ASSERT_RTNL();
 882
 883         best_slave = bond_find_best_slave(bond);
 884         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
 885                 bond_change_active_slave(bond, best_slave);
 886                 rv = bond_set_carrier(bond);
 887                 if (!rv)
 888                         return;
 889
 890                 if (netif_carrier_ok(bond->dev)) {
 891                         netdev_info(bond->dev, "first active interface up!\n");
 892                 } else {
 893                         netdev_info(bond->dev, "now running without any active interface!\n");
 894                 }
 895         }
 896 }
 897
 898 #ifdef CONFIG_NET_POLL_CONTROLLER
 899 static inline int slave_enable_netpoll(struct slave *slave)
 900 {
 901         struct netpoll *np;
 902         int err = 0;
 903
 904         np = kzalloc(sizeof(*np), GFP_KERNEL);
 905         err = -ENOMEM;
 906         if (!np)
 907                 goto out;
 908
 909         err = __netpoll_setup(np, slave->dev);
 910         if (err) {
 911                 kfree(np);
 912                 goto out;
 913         }
 914         slave->np = np;
 915 out:
 916         return err;
 917 }
 918 static inline void slave_disable_netpoll(struct slave *slave)
 919 {
 920         struct netpoll *np = slave->np;
 921
 922         if (!np)
 923                 return;
 924
 925         slave->np = NULL;
 926         __netpoll_free_async(np);
 927 }
 928
 929 static void bond_poll_controller(struct net_device *bond_dev)
 930 {
 931 }
 932
 933 static void bond_netpoll_cleanup(struct net_device *bond_dev)
 934 {
 935         struct bonding *bond = netdev_priv(bond_dev);
 936         struct list_head *iter;
 937         struct slave *slave;
 938
 939         bond_for_each_slave(bond, slave, iter)
 940                 if (bond_slave_is_up(slave))
 941                         slave_disable_netpoll(slave);
 942 }
 943
 944 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
 945 {
 946         struct bonding *bond = netdev_priv(dev);
 947         struct list_head *iter;
 948         struct slave *slave;
 949         int err = 0;
 950
 951         bond_for_each_slave(bond, slave, iter) {
 952                 err = slave_enable_netpoll(slave);
 953                 if (err) {
 954                         bond_netpoll_cleanup(dev);
 955                         break;
 956                 }
 957         }
 958         return err;
 959 }
 960 #else
 961 static inline int slave_enable_netpoll(struct slave *slave)
 962 {
 963         return 0;
 964 }
 965 static inline void slave_disable_netpoll(struct slave *slave)
 966 {
 967 }
 968 static void bond_netpoll_cleanup(struct net_device *bond_dev)
 969 {
 970 }
 971 #endif
 972
 973 /*---------------------------------- IOCTL ----------------------------------*/
 974
 975 static netdev_features_t bond_fix_features(struct net_device *dev,
 976                                            netdev_features_t features)
 977 {
 978         struct bonding *bond = netdev_priv(dev);
 979         struct list_head *iter;
 980         netdev_features_t mask;
 981         struct slave *slave;
 982
 983         /* If any slave has the offload feature flag set,
 984          * set the offload flag on the bond.
 985          */
 986         mask = features | NETIF_F_HW_SWITCH_OFFLOAD;
 987
 988         features &= ~NETIF_F_ONE_FOR_ALL;
 989         features |= NETIF_F_ALL_FOR_ALL;
 990
 991         bond_for_each_slave(bond, slave, iter) {
 992                 features = netdev_increment_features(features,
 993                                                      slave->dev->features,
 994                                                      mask);
 995         }
 996         features = netdev_add_tso_features(features, mask);
 997
 998         return features;
 999 }
1000
1001 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1002                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1003                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1004
1005 #define BOND_ENC_FEATURES       (NETIF_F_ALL_CSUM | NETIF_F_SG | NETIF_F_RXCSUM |\
1006                                  NETIF_F_TSO)
1007
1008 static void bond_compute_features(struct bonding *bond)
1009 {
1010         unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1011                                         IFF_XMIT_DST_RELEASE_PERM;
1012         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1013         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1014         struct net_device *bond_dev = bond->dev;
1015         struct list_head *iter;
1016         struct slave *slave;
1017         unsigned short max_hard_header_len = ETH_HLEN;
1018         unsigned int gso_max_size = GSO_MAX_SIZE;
1019         u16 gso_max_segs = GSO_MAX_SEGS;
1020
1021         if (!bond_has_slaves(bond))
1022                 goto done;
1023         vlan_features &= NETIF_F_ALL_FOR_ALL;
1024
1025         bond_for_each_slave(bond, slave, iter) {
1026                 vlan_features = netdev_increment_features(vlan_features,
1027                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1028
1029                 enc_features = netdev_increment_features(enc_features,
1030                                                          slave->dev->hw_enc_features,
1031                                                          BOND_ENC_FEATURES);
1032                 dst_release_flag &= slave->dev->priv_flags;
1033                 if (slave->dev->hard_header_len > max_hard_header_len)
1034                         max_hard_header_len = slave->dev->hard_header_len;
1035
1036                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1037                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1038         }
1039
1040 done:
1041         bond_dev->vlan_features = vlan_features;
1042         bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
1043         bond_dev->hard_header_len = max_hard_header_len;
1044         bond_dev->gso_max_segs = gso_max_segs;
1045         netif_set_gso_max_size(bond_dev, gso_max_size);
1046
1047         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1048         if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1049             dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1050                 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1051
1052         netdev_change_features(bond_dev);
1053 }
1054
1055 static void bond_setup_by_slave(struct net_device *bond_dev,
1056                                 struct net_device *slave_dev)
1057 {
1058         bond_dev->header_ops        = slave_dev->header_ops;
1059
1060         bond_dev->type              = slave_dev->type;
1061         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1062         bond_dev->addr_len          = slave_dev->addr_len;
1063
1064         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1065                 slave_dev->addr_len);
1066 }
1067
1068 /* On bonding slaves other than the currently active slave, suppress
1069  * duplicates except for alb non-mcast/bcast.
1070  */
1071 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1072                                             struct slave *slave,
1073                                             struct bonding *bond)
1074 {
1075         if (bond_is_slave_inactive(slave)) {
1076                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1077                     skb->pkt_type != PACKET_BROADCAST &&
1078                     skb->pkt_type != PACKET_MULTICAST)
1079                         return false;
1080                 return true;
1081         }
1082         return false;
1083 }
1084
1085 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1086 {
1087         struct sk_buff *skb = *pskb;
1088         struct slave *slave;
1089         struct bonding *bond;
1090         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1091                           struct slave *);
1092         int ret = RX_HANDLER_ANOTHER;
1093
1094         skb = skb_share_check(skb, GFP_ATOMIC);
1095         if (unlikely(!skb))
1096                 return RX_HANDLER_CONSUMED;
1097
1098         *pskb = skb;
1099
1100         slave = bond_slave_get_rcu(skb->dev);
1101         bond = slave->bond;
1102
1103         recv_probe = ACCESS_ONCE(bond->recv_probe);
1104         if (recv_probe) {
1105                 ret = recv_probe(skb, bond, slave);
1106                 if (ret == RX_HANDLER_CONSUMED) {
1107                         consume_skb(skb);
1108                         return ret;
1109                 }
1110         }
1111
1112         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1113                 return RX_HANDLER_EXACT;
1114         }
1115
1116         skb->dev = bond->dev;
1117
1118         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1119             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1120             skb->pkt_type == PACKET_HOST) {
1121
1122                 if (unlikely(skb_cow_head(skb,
1123                                           skb->data - skb_mac_header(skb)))) {
1124                         kfree_skb(skb);
1125                         return RX_HANDLER_CONSUMED;
1126                 }
1127                 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1128         }
1129
1130         return ret;
1131 }
1132
1133 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1134                                       struct net_device *slave_dev,
1135                                       struct slave *slave)
1136 {
1137         int err;
1138
1139         err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1140         if (err)
1141                 return err;
1142         slave_dev->flags |= IFF_SLAVE;
1143         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1144         return 0;
1145 }
1146
1147 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1148                                   struct net_device *slave_dev)
1149 {
1150         netdev_upper_dev_unlink(slave_dev, bond_dev);
1151         slave_dev->flags &= ~IFF_SLAVE;
1152         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1153 }
1154
1155 static struct slave *bond_alloc_slave(struct bonding *bond)
1156 {
1157         struct slave *slave = NULL;
1158
1159         slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1160         if (!slave)
1161                 return NULL;
1162
1163         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1164                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1165                                                GFP_KERNEL);
1166                 if (!SLAVE_AD_INFO(slave)) {
1167                         kfree(slave);
1168                         return NULL;
1169                 }
1170         }
1171         return slave;
1172 }
1173
1174 static void bond_free_slave(struct slave *slave)
1175 {
1176         struct bonding *bond = bond_get_bond_by_slave(slave);
1177
1178         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1179                 kfree(SLAVE_AD_INFO(slave));
1180
1181         kfree(slave);
1182 }
1183
1184 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1185 {
1186         info->bond_mode = BOND_MODE(bond);
1187         info->miimon = bond->params.miimon;
1188         info->num_slaves = bond->slave_cnt;
1189 }
1190
1191 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1192 {
1193         strcpy(info->slave_name, slave->dev->name);
1194         info->link = slave->link;
1195         info->state = bond_slave_state(slave);
1196         info->link_failure_count = slave->link_failure_count;
1197 }
1198
1199 static void bond_netdev_notify(struct net_device *dev,
1200                                struct netdev_bonding_info *info)
1201 {
1202         rtnl_lock();
1203         netdev_bonding_info_change(dev, info);
1204         rtnl_unlock();
1205 }
1206
1207 static void bond_netdev_notify_work(struct work_struct *_work)
1208 {
1209         struct netdev_notify_work *w =
1210                 container_of(_work, struct netdev_notify_work, work.work);
1211
1212         bond_netdev_notify(w->dev, &w->bonding_info);
1213         dev_put(w->dev);
1214         kfree(w);
1215 }
1216
1217 void bond_queue_slave_event(struct slave *slave)
1218 {
1219         struct bonding *bond = slave->bond;
1220         struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1221
1222         if (!nnw)
1223                 return;
1224
1225         dev_hold(slave->dev);
1226         nnw->dev = slave->dev;
1227         bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1228         bond_fill_ifbond(bond, &nnw->bonding_info.master);
1229         INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1230
1231         queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1232 }
1233
1234 /* enslave device <slave> to bond device <master> */
1235 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1236 {
1237         struct bonding *bond = netdev_priv(bond_dev);
1238         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1239         struct slave *new_slave = NULL, *prev_slave;
1240         struct sockaddr addr;
1241         int link_reporting;
1242         int res = 0, i;
1243
1244         if (!bond->params.use_carrier &&
1245             slave_dev->ethtool_ops->get_link == NULL &&
1246             slave_ops->ndo_do_ioctl == NULL) {
1247                 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1248                             slave_dev->name);
1249         }
1250
1251         /* already enslaved */
1252         if (slave_dev->flags & IFF_SLAVE) {
1253                 netdev_dbg(bond_dev, "Error: Device was already enslaved\n");
1254                 return -EBUSY;
1255         }
1256
1257         if (bond_dev == slave_dev) {
1258                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1259                 return -EPERM;
1260         }
1261
1262         /* vlan challenged mutual exclusion */
1263         /* no need to lock since we're protected by rtnl_lock */
1264         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1265                 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1266                            slave_dev->name);
1267                 if (vlan_uses_dev(bond_dev)) {
1268                         netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1269                                    slave_dev->name, bond_dev->name);
1270                         return -EPERM;
1271                 } else {
1272                         netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1273                                     slave_dev->name, slave_dev->name,
1274                                     bond_dev->name);
1275                 }
1276         } else {
1277                 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1278                            slave_dev->name);
1279         }
1280
1281         /* Old ifenslave binaries are no longer supported.  These can
1282          * be identified with moderate accuracy by the state of the slave:
1283          * the current ifenslave will set the interface down prior to
1284          * enslaving it; the old ifenslave will not.
1285          */
1286         if ((slave_dev->flags & IFF_UP)) {
1287                 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1288                            slave_dev->name);
1289                 res = -EPERM;
1290                 goto err_undo_flags;
1291         }
1292
1293         /* set bonding device ether type by slave - bonding netdevices are
1294          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1295          * there is a need to override some of the type dependent attribs/funcs.
1296          *
1297          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1298          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1299          */
1300         if (!bond_has_slaves(bond)) {
1301                 if (bond_dev->type != slave_dev->type) {
1302                         netdev_dbg(bond_dev, "change device type from %d to %d\n",
1303                                    bond_dev->type, slave_dev->type);
1304
1305                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1306                                                        bond_dev);
1307                         res = notifier_to_errno(res);
1308                         if (res) {
1309                                 netdev_err(bond_dev, "refused to change device type\n");
1310                                 res = -EBUSY;
1311                                 goto err_undo_flags;
1312                         }
1313
1314                         /* Flush unicast and multicast addresses */
1315                         dev_uc_flush(bond_dev);
1316                         dev_mc_flush(bond_dev);
1317
1318                         if (slave_dev->type != ARPHRD_ETHER)
1319                                 bond_setup_by_slave(bond_dev, slave_dev);
1320                         else {
1321                                 ether_setup(bond_dev);
1322                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1323                         }
1324
1325                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1326                                                  bond_dev);
1327                 }
1328         } else if (bond_dev->type != slave_dev->type) {
1329                 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1330                            slave_dev->name, slave_dev->type, bond_dev->type);
1331                 res = -EINVAL;
1332                 goto err_undo_flags;
1333         }
1334
1335         if (slave_ops->ndo_set_mac_address == NULL) {
1336                 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1337                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1338                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1339                         if (!bond_has_slaves(bond)) {
1340                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1341                                 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1342                         } else {
1343                                 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");
1344                                 res = -EOPNOTSUPP;
1345                                 goto err_undo_flags;
1346                         }
1347                 }
1348         }
1349
1350         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1351
1352         /* If this is the first slave, then we need to set the master's hardware
1353          * address to be the same as the slave's.
1354          */
1355         if (!bond_has_slaves(bond) &&
1356             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1357                 bond_set_dev_addr(bond->dev, slave_dev);
1358
1359         new_slave = bond_alloc_slave(bond);
1360         if (!new_slave) {
1361                 res = -ENOMEM;
1362                 goto err_undo_flags;
1363         }
1364
1365         new_slave->bond = bond;
1366         new_slave->dev = slave_dev;
1367         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1368          * is set via sysfs or module option if desired.
1369          */
1370         new_slave->queue_id = 0;
1371
1372         /* Save slave's original mtu and then set it to match the bond */
1373         new_slave->original_mtu = slave_dev->mtu;
1374         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1375         if (res) {
1376                 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1377                 goto err_free;
1378         }
1379
1380         /* Save slave's original ("permanent") mac address for modes
1381          * that need it, and for restoring it upon release, and then
1382          * set it to the master's address
1383          */
1384         ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1385
1386         if (!bond->params.fail_over_mac ||
1387             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1388                 /* Set slave to master's mac address.  The application already
1389                  * set the master's mac address to that of the first slave
1390                  */
1391                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1392                 addr.sa_family = slave_dev->type;
1393                 res = dev_set_mac_address(slave_dev, &addr);
1394                 if (res) {
1395                         netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1396                         goto err_restore_mtu;
1397                 }
1398         }
1399
1400         /* open the slave since the application closed it */
1401         res = dev_open(slave_dev);
1402         if (res) {
1403                 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1404                 goto err_restore_mac;
1405         }
1406
1407         slave_dev->priv_flags |= IFF_BONDING;
1408         /* initialize slave stats */
1409         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1410
1411         if (bond_is_lb(bond)) {
1412                 /* bond_alb_init_slave() must be called before all other stages since
1413                  * it might fail and we do not want to have to undo everything
1414                  */
1415                 res = bond_alb_init_slave(bond, new_slave);
1416                 if (res)
1417                         goto err_close;
1418         }
1419
1420         /* If the mode uses primary, then the following is handled by
1421          * bond_change_active_slave().
1422          */
1423         if (!bond_uses_primary(bond)) {
1424                 /* set promiscuity level to new slave */
1425                 if (bond_dev->flags & IFF_PROMISC) {
1426                         res = dev_set_promiscuity(slave_dev, 1);
1427                         if (res)
1428                                 goto err_close;
1429                 }
1430
1431                 /* set allmulti level to new slave */
1432                 if (bond_dev->flags & IFF_ALLMULTI) {
1433                         res = dev_set_allmulti(slave_dev, 1);
1434                         if (res)
1435                                 goto err_close;
1436                 }
1437
1438                 netif_addr_lock_bh(bond_dev);
1439
1440                 dev_mc_sync_multiple(slave_dev, bond_dev);
1441                 dev_uc_sync_multiple(slave_dev, bond_dev);
1442
1443                 netif_addr_unlock_bh(bond_dev);
1444         }
1445
1446         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1447                 /* add lacpdu mc addr to mc list */
1448                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1449
1450                 dev_mc_add(slave_dev, lacpdu_multicast);
1451         }
1452
1453         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1454         if (res) {
1455                 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1456                            slave_dev->name);
1457                 goto err_close;
1458         }
1459
1460         prev_slave = bond_last_slave(bond);
1461
1462         new_slave->delay = 0;
1463         new_slave->link_failure_count = 0;
1464
1465         bond_update_speed_duplex(new_slave);
1466
1467         new_slave->last_rx = jiffies -
1468                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1469         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1470                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1471
1472         if (bond->params.miimon && !bond->params.use_carrier) {
1473                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1474
1475                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1476                         /* miimon is set but a bonded network driver
1477                          * does not support ETHTOOL/MII and
1478                          * arp_interval is not set.  Note: if
1479                          * use_carrier is enabled, we will never go
1480                          * here (because netif_carrier is always
1481                          * supported); thus, we don't need to change
1482                          * the messages for netif_carrier.
1483                          */
1484                         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",
1485                                     slave_dev->name);
1486                 } else if (link_reporting == -1) {
1487                         /* unable get link status using mii/ethtool */
1488                         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",
1489                                     slave_dev->name);
1490                 }
1491         }
1492
1493         /* check for initial state */
1494         if (bond->params.miimon) {
1495                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1496                         if (bond->params.updelay) {
1497                                 bond_set_slave_link_state(new_slave,
1498                                                           BOND_LINK_BACK);
1499                                 new_slave->delay = bond->params.updelay;
1500                         } else {
1501                                 bond_set_slave_link_state(new_slave,
1502                                                           BOND_LINK_UP);
1503                         }
1504                 } else {
1505                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN);
1506                 }
1507         } else if (bond->params.arp_interval) {
1508                 bond_set_slave_link_state(new_slave,
1509                                           (netif_carrier_ok(slave_dev) ?
1510                                           BOND_LINK_UP : BOND_LINK_DOWN));
1511         } else {
1512                 bond_set_slave_link_state(new_slave, BOND_LINK_UP);
1513         }
1514
1515         if (new_slave->link != BOND_LINK_DOWN)
1516                 new_slave->last_link_up = jiffies;
1517         netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1518                    new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1519                    (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1520
1521         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1522                 /* if there is a primary slave, remember it */
1523                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1524                         rcu_assign_pointer(bond->primary_slave, new_slave);
1525                         bond->force_primary = true;
1526                 }
1527         }
1528
1529         switch (BOND_MODE(bond)) {
1530         case BOND_MODE_ACTIVEBACKUP:
1531                 bond_set_slave_inactive_flags(new_slave,
1532                                               BOND_SLAVE_NOTIFY_NOW);
1533                 break;
1534         case BOND_MODE_8023AD:
1535                 /* in 802.3ad mode, the internal mechanism
1536                  * will activate the slaves in the selected
1537                  * aggregator
1538                  */
1539                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1540                 /* if this is the first slave */
1541                 if (!prev_slave) {
1542                         SLAVE_AD_INFO(new_slave)->id = 1;
1543                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1544                          * can be called only after the mac address of the bond is set
1545                          */
1546                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1547                 } else {
1548                         SLAVE_AD_INFO(new_slave)->id =
1549                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1550                 }
1551
1552                 bond_3ad_bind_slave(new_slave);
1553                 break;
1554         case BOND_MODE_TLB:
1555         case BOND_MODE_ALB:
1556                 bond_set_active_slave(new_slave);
1557                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1558                 break;
1559         default:
1560                 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1561
1562                 /* always active in trunk mode */
1563                 bond_set_active_slave(new_slave);
1564
1565                 /* In trunking mode there is little meaning to curr_active_slave
1566                  * anyway (it holds no special properties of the bond device),
1567                  * so we can change it without calling change_active_interface()
1568                  */
1569                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1570                     new_slave->link == BOND_LINK_UP)
1571                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1572
1573                 break;
1574         } /* switch(bond_mode) */
1575
1576 #ifdef CONFIG_NET_POLL_CONTROLLER
1577         slave_dev->npinfo = bond->dev->npinfo;
1578         if (slave_dev->npinfo) {
1579                 if (slave_enable_netpoll(new_slave)) {
1580                         netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1581                         res = -EBUSY;
1582                         goto err_detach;
1583                 }
1584         }
1585 #endif
1586
1587         if (!(bond_dev->features & NETIF_F_LRO))
1588                 dev_disable_lro(slave_dev);
1589
1590         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1591                                          new_slave);
1592         if (res) {
1593                 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1594                 goto err_detach;
1595         }
1596
1597         res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1598         if (res) {
1599                 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1600                 goto err_unregister;
1601         }
1602
1603         res = bond_sysfs_slave_add(new_slave);
1604         if (res) {
1605                 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1606                 goto err_upper_unlink;
1607         }
1608
1609         bond->slave_cnt++;
1610         bond_compute_features(bond);
1611         bond_set_carrier(bond);
1612
1613         if (bond_uses_primary(bond)) {
1614                 block_netpoll_tx();
1615                 bond_select_active_slave(bond);
1616                 unblock_netpoll_tx();
1617         }
1618
1619         if (bond_mode_uses_xmit_hash(bond))
1620                 bond_update_slave_arr(bond, NULL);
1621
1622         netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1623                     slave_dev->name,
1624                     bond_is_active_slave(new_slave) ? "an active" : "a backup",
1625                     new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1626
1627         /* enslave is successful */
1628         bond_queue_slave_event(new_slave);
1629         return 0;
1630
1631 /* Undo stages on error */
1632 err_upper_unlink:
1633         bond_upper_dev_unlink(bond_dev, slave_dev);
1634
1635 err_unregister:
1636         netdev_rx_handler_unregister(slave_dev);
1637
1638 err_detach:
1639         if (!bond_uses_primary(bond))
1640                 bond_hw_addr_flush(bond_dev, slave_dev);
1641
1642         vlan_vids_del_by_dev(slave_dev, bond_dev);
1643         if (rcu_access_pointer(bond->primary_slave) == new_slave)
1644                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1645         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1646                 block_netpoll_tx();
1647                 bond_change_active_slave(bond, NULL);
1648                 bond_select_active_slave(bond);
1649                 unblock_netpoll_tx();
1650         }
1651         /* either primary_slave or curr_active_slave might've changed */
1652         synchronize_rcu();
1653         slave_disable_netpoll(new_slave);
1654
1655 err_close:
1656         slave_dev->priv_flags &= ~IFF_BONDING;
1657         dev_close(slave_dev);
1658
1659 err_restore_mac:
1660         if (!bond->params.fail_over_mac ||
1661             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1662                 /* XXX TODO - fom follow mode needs to change master's
1663                  * MAC if this slave's MAC is in use by the bond, or at
1664                  * least print a warning.
1665                  */
1666                 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1667                 addr.sa_family = slave_dev->type;
1668                 dev_set_mac_address(slave_dev, &addr);
1669         }
1670
1671 err_restore_mtu:
1672         dev_set_mtu(slave_dev, new_slave->original_mtu);
1673
1674 err_free:
1675         bond_free_slave(new_slave);
1676
1677 err_undo_flags:
1678         /* Enslave of first slave has failed and we need to fix master's mac */
1679         if (!bond_has_slaves(bond) &&
1680             ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
1681                 eth_hw_addr_random(bond_dev);
1682
1683         return res;
1684 }
1685
1686 /* Try to release the slave device <slave> from the bond device <master>
1687  * It is legal to access curr_active_slave without a lock because all the function
1688  * is RTNL-locked. If "all" is true it means that the function is being called
1689  * while destroying a bond interface and all slaves are being released.
1690  *
1691  * The rules for slave state should be:
1692  *   for Active/Backup:
1693  *     Active stays on all backups go down
1694  *   for Bonded connections:
1695  *     The first up interface should be left on and all others downed.
1696  */
1697 static int __bond_release_one(struct net_device *bond_dev,
1698                               struct net_device *slave_dev,
1699                               bool all)
1700 {
1701         struct bonding *bond = netdev_priv(bond_dev);
1702         struct slave *slave, *oldcurrent;
1703         struct sockaddr addr;
1704         int old_flags = bond_dev->flags;
1705         netdev_features_t old_features = bond_dev->features;
1706
1707         /* slave is not a slave or master is not master of this slave */
1708         if (!(slave_dev->flags & IFF_SLAVE) ||
1709             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1710                 netdev_dbg(bond_dev, "cannot release %s\n",
1711                            slave_dev->name);
1712                 return -EINVAL;
1713         }
1714
1715         block_netpoll_tx();
1716
1717         slave = bond_get_slave_by_dev(bond, slave_dev);
1718         if (!slave) {
1719                 /* not a slave of this bond */
1720                 netdev_info(bond_dev, "%s not enslaved\n",
1721                             slave_dev->name);
1722                 unblock_netpoll_tx();
1723                 return -EINVAL;
1724         }
1725
1726         bond_sysfs_slave_del(slave);
1727
1728         /* recompute stats just before removing the slave */
1729         bond_get_stats(bond->dev, &bond->bond_stats);
1730
1731         bond_upper_dev_unlink(bond_dev, slave_dev);
1732         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1733          * for this slave anymore.
1734          */
1735         netdev_rx_handler_unregister(slave_dev);
1736
1737         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1738                 bond_3ad_unbind_slave(slave);
1739
1740         if (bond_mode_uses_xmit_hash(bond))
1741                 bond_update_slave_arr(bond, slave);
1742
1743         netdev_info(bond_dev, "Releasing %s interface %s\n",
1744                     bond_is_active_slave(slave) ? "active" : "backup",
1745                     slave_dev->name);
1746
1747         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1748
1749         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1750
1751         if (!all && (!bond->params.fail_over_mac ||
1752                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1753                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1754                     bond_has_slaves(bond))
1755                         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",
1756                                     slave_dev->name, slave->perm_hwaddr,
1757                                     bond_dev->name, slave_dev->name);
1758         }
1759
1760         if (rtnl_dereference(bond->primary_slave) == slave)
1761                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1762
1763         if (oldcurrent == slave)
1764                 bond_change_active_slave(bond, NULL);
1765
1766         if (bond_is_lb(bond)) {
1767                 /* Must be called only after the slave has been
1768                  * detached from the list and the curr_active_slave
1769                  * has been cleared (if our_slave == old_current),
1770                  * but before a new active slave is selected.
1771                  */
1772                 bond_alb_deinit_slave(bond, slave);
1773         }
1774
1775         if (all) {
1776                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1777         } else if (oldcurrent == slave) {
1778                 /* Note that we hold RTNL over this sequence, so there
1779                  * is no concern that another slave add/remove event
1780                  * will interfere.
1781                  */
1782                 bond_select_active_slave(bond);
1783         }
1784
1785         if (!bond_has_slaves(bond)) {
1786                 bond_set_carrier(bond);
1787                 eth_hw_addr_random(bond_dev);
1788         }
1789
1790         unblock_netpoll_tx();
1791         synchronize_rcu();
1792         bond->slave_cnt--;
1793
1794         if (!bond_has_slaves(bond)) {
1795                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1796                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1797         }
1798
1799         bond_compute_features(bond);
1800         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1801             (old_features & NETIF_F_VLAN_CHALLENGED))
1802                 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1803                             slave_dev->name, bond_dev->name);
1804
1805         vlan_vids_del_by_dev(slave_dev, bond_dev);
1806
1807         /* If the mode uses primary, then this case was handled above by
1808          * bond_change_active_slave(..., NULL)
1809          */
1810         if (!bond_uses_primary(bond)) {
1811                 /* unset promiscuity level from slave
1812                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1813                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1814                  * value of that flag before that change, as that was the value
1815                  * when this slave was attached, so we cache at the start of the
1816                  * function and use it here. Same goes for ALLMULTI below
1817                  */
1818                 if (old_flags & IFF_PROMISC)
1819                         dev_set_promiscuity(slave_dev, -1);
1820
1821                 /* unset allmulti level from slave */
1822                 if (old_flags & IFF_ALLMULTI)
1823                         dev_set_allmulti(slave_dev, -1);
1824
1825                 bond_hw_addr_flush(bond_dev, slave_dev);
1826         }
1827
1828         slave_disable_netpoll(slave);
1829
1830         /* close slave before restoring its mac address */
1831         dev_close(slave_dev);
1832
1833         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1834             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1835                 /* restore original ("permanent") mac address */
1836                 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1837                 addr.sa_family = slave_dev->type;
1838                 dev_set_mac_address(slave_dev, &addr);
1839         }
1840
1841         dev_set_mtu(slave_dev, slave->original_mtu);
1842
1843         slave_dev->priv_flags &= ~IFF_BONDING;
1844
1845         bond_free_slave(slave);
1846
1847         return 0;
1848 }
1849
1850 /* A wrapper used because of ndo_del_link */
1851 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1852 {
1853         return __bond_release_one(bond_dev, slave_dev, false);
1854 }
1855
1856 /* First release a slave and then destroy the bond if no more slaves are left.
1857  * Must be under rtnl_lock when this function is called.
1858  */
1859 static int  bond_release_and_destroy(struct net_device *bond_dev,
1860                                      struct net_device *slave_dev)
1861 {
1862         struct bonding *bond = netdev_priv(bond_dev);
1863         int ret;
1864
1865         ret = bond_release(bond_dev, slave_dev);
1866         if (ret == 0 && !bond_has_slaves(bond)) {
1867                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1868                 netdev_info(bond_dev, "Destroying bond %s\n",
1869                             bond_dev->name);
1870                 unregister_netdevice(bond_dev);
1871         }
1872         return ret;
1873 }
1874
1875 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1876 {
1877         struct bonding *bond = netdev_priv(bond_dev);
1878         bond_fill_ifbond(bond, info);
1879         return 0;
1880 }
1881
1882 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1883 {
1884         struct bonding *bond = netdev_priv(bond_dev);
1885         struct list_head *iter;
1886         int i = 0, res = -ENODEV;
1887         struct slave *slave;
1888
1889         bond_for_each_slave(bond, slave, iter) {
1890                 if (i++ == (int)info->slave_id) {
1891                         res = 0;
1892                         bond_fill_ifslave(slave, info);
1893                         break;
1894                 }
1895         }
1896
1897         return res;
1898 }
1899
1900 /*-------------------------------- Monitoring -------------------------------*/
1901
1902 /* called with rcu_read_lock() */
1903 static int bond_miimon_inspect(struct bonding *bond)
1904 {
1905         int link_state, commit = 0;
1906         struct list_head *iter;
1907         struct slave *slave;
1908         bool ignore_updelay;
1909
1910         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
1911
1912         bond_for_each_slave_rcu(bond, slave, iter) {
1913                 slave->new_link = BOND_LINK_NOCHANGE;
1914
1915                 link_state = bond_check_dev_link(bond, slave->dev, 0);
1916
1917                 switch (slave->link) {
1918                 case BOND_LINK_UP:
1919                         if (link_state)
1920                                 continue;
1921
1922                         bond_set_slave_link_state(slave, BOND_LINK_FAIL);
1923                         slave->delay = bond->params.downdelay;
1924                         if (slave->delay) {
1925                                 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
1926                                             (BOND_MODE(bond) ==
1927                                              BOND_MODE_ACTIVEBACKUP) ?
1928                                              (bond_is_active_slave(slave) ?
1929                                               "active " : "backup ") : "",
1930                                             slave->dev->name,
1931                                             bond->params.downdelay * bond->params.miimon);
1932                         }
1933                         /*FALLTHRU*/
1934                 case BOND_LINK_FAIL:
1935                         if (link_state) {
1936                                 /* recovered before downdelay expired */
1937                                 bond_set_slave_link_state(slave, BOND_LINK_UP);
1938                                 slave->last_link_up = jiffies;
1939                                 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
1940                                             (bond->params.downdelay - slave->delay) *
1941                                             bond->params.miimon,
1942                                             slave->dev->name);
1943                                 continue;
1944                         }
1945
1946                         if (slave->delay <= 0) {
1947                                 slave->new_link = BOND_LINK_DOWN;
1948                                 commit++;
1949                                 continue;
1950                         }
1951
1952                         slave->delay--;
1953                         break;
1954
1955                 case BOND_LINK_DOWN:
1956                         if (!link_state)
1957                                 continue;
1958
1959                         bond_set_slave_link_state(slave, BOND_LINK_BACK);
1960                         slave->delay = bond->params.updelay;
1961
1962                         if (slave->delay) {
1963                                 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
1964                                             slave->dev->name,
1965                                             ignore_updelay ? 0 :
1966                                             bond->params.updelay *
1967                                             bond->params.miimon);
1968                         }
1969                         /*FALLTHRU*/
1970                 case BOND_LINK_BACK:
1971                         if (!link_state) {
1972                                 bond_set_slave_link_state(slave,
1973                                                           BOND_LINK_DOWN);
1974                                 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
1975                                             (bond->params.updelay - slave->delay) *
1976                                             bond->params.miimon,
1977                                             slave->dev->name);
1978
1979                                 continue;
1980                         }
1981
1982                         if (ignore_updelay)
1983                                 slave->delay = 0;
1984
1985                         if (slave->delay <= 0) {
1986                                 slave->new_link = BOND_LINK_UP;
1987                                 commit++;
1988                                 ignore_updelay = false;
1989                                 continue;
1990                         }
1991
1992                         slave->delay--;
1993                         break;
1994                 }
1995         }
1996
1997         return commit;
1998 }
1999
2000 static void bond_miimon_commit(struct bonding *bond)
2001 {
2002         struct list_head *iter;
2003         struct slave *slave, *primary;
2004
2005         bond_for_each_slave(bond, slave, iter) {
2006                 switch (slave->new_link) {
2007                 case BOND_LINK_NOCHANGE:
2008                         continue;
2009
2010                 case BOND_LINK_UP:
2011                         bond_set_slave_link_state(slave, BOND_LINK_UP);
2012                         slave->last_link_up = jiffies;
2013
2014                         primary = rtnl_dereference(bond->primary_slave);
2015                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2016                                 /* prevent it from being the active one */
2017                                 bond_set_backup_slave(slave);
2018                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2019                                 /* make it immediately active */
2020                                 bond_set_active_slave(slave);
2021                         } else if (slave != primary) {
2022                                 /* prevent it from being the active one */
2023                                 bond_set_backup_slave(slave);
2024                         }
2025
2026                         netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2027                                     slave->dev->name,
2028                                     slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2029                                     slave->duplex ? "full" : "half");
2030
2031                         /* notify ad that the link status has changed */
2032                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2033                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2034
2035                         if (bond_is_lb(bond))
2036                                 bond_alb_handle_link_change(bond, slave,
2037                                                             BOND_LINK_UP);
2038
2039                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2040                                 bond_update_slave_arr(bond, NULL);
2041
2042                         if (!bond->curr_active_slave || slave == primary)
2043                                 goto do_failover;
2044
2045                         continue;
2046
2047                 case BOND_LINK_DOWN:
2048                         if (slave->link_failure_count < UINT_MAX)
2049                                 slave->link_failure_count++;
2050
2051                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2052
2053                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2054                             BOND_MODE(bond) == BOND_MODE_8023AD)
2055                                 bond_set_slave_inactive_flags(slave,
2056                                                               BOND_SLAVE_NOTIFY_NOW);
2057
2058                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2059                                     slave->dev->name);
2060
2061                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2062                                 bond_3ad_handle_link_change(slave,
2063                                                             BOND_LINK_DOWN);
2064
2065                         if (bond_is_lb(bond))
2066                                 bond_alb_handle_link_change(bond, slave,
2067                                                             BOND_LINK_DOWN);
2068
2069                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2070                                 bond_update_slave_arr(bond, NULL);
2071
2072                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2073                                 goto do_failover;
2074
2075                         continue;
2076
2077                 default:
2078                         netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2079                                    slave->new_link, slave->dev->name);
2080                         slave->new_link = BOND_LINK_NOCHANGE;
2081
2082                         continue;
2083                 }
2084
2085 do_failover:
2086                 block_netpoll_tx();
2087                 bond_select_active_slave(bond);
2088                 unblock_netpoll_tx();
2089         }
2090
2091         bond_set_carrier(bond);
2092 }
2093
2094 /* bond_mii_monitor
2095  *
2096  * Really a wrapper that splits the mii monitor into two phases: an
2097  * inspection, then (if inspection indicates something needs to be done)
2098  * an acquisition of appropriate locks followed by a commit phase to
2099  * implement whatever link state changes are indicated.
2100  */
2101 static void bond_mii_monitor(struct work_struct *work)
2102 {
2103         struct bonding *bond = container_of(work, struct bonding,
2104                                             mii_work.work);
2105         bool should_notify_peers = false;
2106         unsigned long delay;
2107
2108         delay = msecs_to_jiffies(bond->params.miimon);
2109
2110         if (!bond_has_slaves(bond))
2111                 goto re_arm;
2112
2113         rcu_read_lock();
2114
2115         should_notify_peers = bond_should_notify_peers(bond);
2116
2117         if (bond_miimon_inspect(bond)) {
2118                 rcu_read_unlock();
2119
2120                 /* Race avoidance with bond_close cancel of workqueue */
2121                 if (!rtnl_trylock()) {
2122                         delay = 1;
2123                         should_notify_peers = false;
2124                         goto re_arm;
2125                 }
2126
2127                 bond_miimon_commit(bond);
2128
2129                 rtnl_unlock();  /* might sleep, hold no other locks */
2130         } else
2131                 rcu_read_unlock();
2132
2133 re_arm:
2134         if (bond->params.miimon)
2135                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2136
2137         if (should_notify_peers) {
2138                 if (!rtnl_trylock())
2139                         return;
2140                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2141                 rtnl_unlock();
2142         }
2143 }
2144
2145 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2146 {
2147         struct net_device *upper;
2148         struct list_head *iter;
2149         bool ret = false;
2150
2151         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2152                 return true;
2153
2154         rcu_read_lock();
2155         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2156                 if (ip == bond_confirm_addr(upper, 0, ip)) {
2157                         ret = true;
2158                         break;
2159                 }
2160         }
2161         rcu_read_unlock();
2162
2163         return ret;
2164 }
2165
2166 /* We go to the (large) trouble of VLAN tagging ARP frames because
2167  * switches in VLAN mode (especially if ports are configured as
2168  * "native" to a VLAN) might not pass non-tagged frames.
2169  */
2170 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2171                           __be32 dest_ip, __be32 src_ip,
2172                           struct bond_vlan_tag *tags)
2173 {
2174         struct sk_buff *skb;
2175         struct bond_vlan_tag *outer_tag = tags;
2176
2177         netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2178                    arp_op, slave_dev->name, &dest_ip, &src_ip);
2179
2180         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2181                          NULL, slave_dev->dev_addr, NULL);
2182
2183         if (!skb) {
2184                 net_err_ratelimited("ARP packet allocation failed\n");
2185                 return;
2186         }
2187
2188         if (!tags || tags->vlan_proto == VLAN_N_VID)
2189                 goto xmit;
2190
2191         tags++;
2192
2193         /* Go through all the tags backwards and add them to the packet */
2194         while (tags->vlan_proto != VLAN_N_VID) {
2195                 if (!tags->vlan_id) {
2196                         tags++;
2197                         continue;
2198                 }
2199
2200                 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2201                            ntohs(outer_tag->vlan_proto), tags->vlan_id);
2202                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2203                                                 tags->vlan_id);
2204                 if (!skb) {
2205                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2206                         return;
2207                 }
2208
2209                 tags++;
2210         }
2211         /* Set the outer tag */
2212         if (outer_tag->vlan_id) {
2213                 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2214                            ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2215                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2216                                        outer_tag->vlan_id);
2217         }
2218
2219 xmit:
2220         arp_xmit(skb);
2221 }
2222
2223 /* Validate the device path between the @start_dev and the @end_dev.
2224  * The path is valid if the @end_dev is reachable through device
2225  * stacking.
2226  * When the path is validated, collect any vlan information in the
2227  * path.
2228  */
2229 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2230                                               struct net_device *end_dev,
2231                                               int level)
2232 {
2233         struct bond_vlan_tag *tags;
2234         struct net_device *upper;
2235         struct list_head  *iter;
2236
2237         if (start_dev == end_dev) {
2238                 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2239                 if (!tags)
2240                         return ERR_PTR(-ENOMEM);
2241                 tags[level].vlan_proto = VLAN_N_VID;
2242                 return tags;
2243         }
2244
2245         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2246                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2247                 if (IS_ERR_OR_NULL(tags)) {
2248                         if (IS_ERR(tags))
2249                                 return tags;
2250                         continue;
2251                 }
2252                 if (is_vlan_dev(upper)) {
2253                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2254                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2255                 }
2256
2257                 return tags;
2258         }
2259
2260         return NULL;
2261 }
2262
2263 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2264 {
2265         struct rtable *rt;
2266         struct bond_vlan_tag *tags;
2267         __be32 *targets = bond->params.arp_targets, addr;
2268         int i;
2269
2270         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2271                 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2272                 tags = NULL;
2273
2274                 /* Find out through which dev should the packet go */
2275                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2276                                      RTO_ONLINK, 0);
2277                 if (IS_ERR(rt)) {
2278                         /* there's no route to target - try to send arp
2279                          * probe to generate any traffic (arp_validate=0)
2280                          */
2281                         if (bond->params.arp_validate)
2282                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2283                                                      bond->dev->name,
2284                                                      &targets[i]);
2285                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2286                                       0, tags);
2287                         continue;
2288                 }
2289
2290                 /* bond device itself */
2291                 if (rt->dst.dev == bond->dev)
2292                         goto found;
2293
2294                 rcu_read_lock();
2295                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2296                 rcu_read_unlock();
2297
2298                 if (!IS_ERR_OR_NULL(tags))
2299                         goto found;
2300
2301                 /* Not our device - skip */
2302                 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2303                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2304
2305                 ip_rt_put(rt);
2306                 continue;
2307
2308 found:
2309                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2310                 ip_rt_put(rt);
2311                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2312                               addr, tags);
2313                 kfree(tags);
2314         }
2315 }
2316
2317 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2318 {
2319         int i;
2320
2321         if (!sip || !bond_has_this_ip(bond, tip)) {
2322                 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2323                            &sip, &tip);
2324                 return;
2325         }
2326
2327         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2328         if (i == -1) {
2329                 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2330                            &sip);
2331                 return;
2332         }
2333         slave->last_rx = jiffies;
2334         slave->target_last_arp_rx[i] = jiffies;
2335 }
2336
2337 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2338                  struct slave *slave)
2339 {
2340         struct arphdr *arp = (struct arphdr *)skb->data;
2341         struct slave *curr_active_slave;
2342         unsigned char *arp_ptr;
2343         __be32 sip, tip;
2344         int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2345
2346         if (!slave_do_arp_validate(bond, slave)) {
2347                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2348                     !slave_do_arp_validate_only(bond))
2349                         slave->last_rx = jiffies;
2350                 return RX_HANDLER_ANOTHER;
2351         } else if (!is_arp) {
2352                 return RX_HANDLER_ANOTHER;
2353         }
2354
2355         alen = arp_hdr_len(bond->dev);
2356
2357         netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2358                    skb->dev->name);
2359
2360         if (alen > skb_headlen(skb)) {
2361                 arp = kmalloc(alen, GFP_ATOMIC);
2362                 if (!arp)
2363                         goto out_unlock;
2364                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2365                         goto out_unlock;
2366         }
2367
2368         if (arp->ar_hln != bond->dev->addr_len ||
2369             skb->pkt_type == PACKET_OTHERHOST ||
2370             skb->pkt_type == PACKET_LOOPBACK ||
2371             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2372             arp->ar_pro != htons(ETH_P_IP) ||
2373             arp->ar_pln != 4)
2374                 goto out_unlock;
2375
2376         arp_ptr = (unsigned char *)(arp + 1);
2377         arp_ptr += bond->dev->addr_len;
2378         memcpy(&sip, arp_ptr, 4);
2379         arp_ptr += 4 + bond->dev->addr_len;
2380         memcpy(&tip, arp_ptr, 4);
2381
2382         netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2383                    slave->dev->name, bond_slave_state(slave),
2384                      bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2385                      &sip, &tip);
2386
2387         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2388
2389         /* Backup slaves won't see the ARP reply, but do come through
2390          * here for each ARP probe (so we swap the sip/tip to validate
2391          * the probe).  In a "redundant switch, common router" type of
2392          * configuration, the ARP probe will (hopefully) travel from
2393          * the active, through one switch, the router, then the other
2394          * switch before reaching the backup.
2395          *
2396          * We 'trust' the arp requests if there is an active slave and
2397          * it received valid arp reply(s) after it became active. This
2398          * is done to avoid endless looping when we can't reach the
2399          * arp_ip_target and fool ourselves with our own arp requests.
2400          */
2401
2402         if (bond_is_active_slave(slave))
2403                 bond_validate_arp(bond, slave, sip, tip);
2404         else if (curr_active_slave &&
2405                  time_after(slave_last_rx(bond, curr_active_slave),
2406                             curr_active_slave->last_link_up))
2407                 bond_validate_arp(bond, slave, tip, sip);
2408
2409 out_unlock:
2410         if (arp != (struct arphdr *)skb->data)
2411                 kfree(arp);
2412         return RX_HANDLER_ANOTHER;
2413 }
2414
2415 /* function to verify if we're in the arp_interval timeslice, returns true if
2416  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2417  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2418  */
2419 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2420                                   int mod)
2421 {
2422         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2423
2424         return time_in_range(jiffies,
2425                              last_act - delta_in_ticks,
2426                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2427 }
2428
2429 /* This function is called regularly to monitor each slave's link
2430  * ensuring that traffic is being sent and received when arp monitoring
2431  * is used in load-balancing mode. if the adapter has been dormant, then an
2432  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2433  * arp monitoring in active backup mode.
2434  */
2435 static void bond_loadbalance_arp_mon(struct work_struct *work)
2436 {
2437         struct bonding *bond = container_of(work, struct bonding,
2438                                             arp_work.work);
2439         struct slave *slave, *oldcurrent;
2440         struct list_head *iter;
2441         int do_failover = 0, slave_state_changed = 0;
2442
2443         if (!bond_has_slaves(bond))
2444                 goto re_arm;
2445
2446         rcu_read_lock();
2447
2448         oldcurrent = rcu_dereference(bond->curr_active_slave);
2449         /* see if any of the previous devices are up now (i.e. they have
2450          * xmt and rcv traffic). the curr_active_slave does not come into
2451          * the picture unless it is null. also, slave->last_link_up is not
2452          * needed here because we send an arp on each slave and give a slave
2453          * as long as it needs to get the tx/rx within the delta.
2454          * TODO: what about up/down delay in arp mode? it wasn't here before
2455          *       so it can wait
2456          */
2457         bond_for_each_slave_rcu(bond, slave, iter) {
2458                 unsigned long trans_start = dev_trans_start(slave->dev);
2459
2460                 if (slave->link != BOND_LINK_UP) {
2461                         if (bond_time_in_interval(bond, trans_start, 1) &&
2462                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2463
2464                                 slave->link  = BOND_LINK_UP;
2465                                 slave_state_changed = 1;
2466
2467                                 /* primary_slave has no meaning in round-robin
2468                                  * mode. the window of a slave being up and
2469                                  * curr_active_slave being null after enslaving
2470                                  * is closed.
2471                                  */
2472                                 if (!oldcurrent) {
2473                                         netdev_info(bond->dev, "link status definitely up for interface %s\n",
2474                                                     slave->dev->name);
2475                                         do_failover = 1;
2476                                 } else {
2477                                         netdev_info(bond->dev, "interface %s is now up\n",
2478                                                     slave->dev->name);
2479                                 }
2480                         }
2481                 } else {
2482                         /* slave->link == BOND_LINK_UP */
2483
2484                         /* not all switches will respond to an arp request
2485                          * when the source ip is 0, so don't take the link down
2486                          * if we don't know our ip yet
2487                          */
2488                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2489                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2490
2491                                 slave->link  = BOND_LINK_DOWN;
2492                                 slave_state_changed = 1;
2493
2494                                 if (slave->link_failure_count < UINT_MAX)
2495                                         slave->link_failure_count++;
2496
2497                                 netdev_info(bond->dev, "interface %s is now down\n",
2498                                             slave->dev->name);
2499
2500                                 if (slave == oldcurrent)
2501                                         do_failover = 1;
2502                         }
2503                 }
2504
2505                 /* note: if switch is in round-robin mode, all links
2506                  * must tx arp to ensure all links rx an arp - otherwise
2507                  * links may oscillate or not come up at all; if switch is
2508                  * in something like xor mode, there is nothing we can
2509                  * do - all replies will be rx'ed on same link causing slaves
2510                  * to be unstable during low/no traffic periods
2511                  */
2512                 if (bond_slave_is_up(slave))
2513                         bond_arp_send_all(bond, slave);
2514         }
2515
2516         rcu_read_unlock();
2517
2518         if (do_failover || slave_state_changed) {
2519                 if (!rtnl_trylock())
2520                         goto re_arm;
2521
2522                 if (slave_state_changed) {
2523                         bond_slave_state_change(bond);
2524                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2525                                 bond_update_slave_arr(bond, NULL);
2526                 }
2527                 if (do_failover) {
2528                         block_netpoll_tx();
2529                         bond_select_active_slave(bond);
2530                         unblock_netpoll_tx();
2531                 }
2532                 rtnl_unlock();
2533         }
2534
2535 re_arm:
2536         if (bond->params.arp_interval)
2537                 queue_delayed_work(bond->wq, &bond->arp_work,
2538                                    msecs_to_jiffies(bond->params.arp_interval));
2539 }
2540
2541 /* Called to inspect slaves for active-backup mode ARP monitor link state
2542  * changes.  Sets new_link in slaves to specify what action should take
2543  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2544  * to link states must be committed.
2545  *
2546  * Called with rcu_read_lock held.
2547  */
2548 static int bond_ab_arp_inspect(struct bonding *bond)
2549 {
2550         unsigned long trans_start, last_rx;
2551         struct list_head *iter;
2552         struct slave *slave;
2553         int commit = 0;
2554
2555         bond_for_each_slave_rcu(bond, slave, iter) {
2556                 slave->new_link = BOND_LINK_NOCHANGE;
2557                 last_rx = slave_last_rx(bond, slave);
2558
2559                 if (slave->link != BOND_LINK_UP) {
2560                         if (bond_time_in_interval(bond, last_rx, 1)) {
2561                                 slave->new_link = BOND_LINK_UP;
2562                                 commit++;
2563                         }
2564                         continue;
2565                 }
2566
2567                 /* Give slaves 2*delta after being enslaved or made
2568                  * active.  This avoids bouncing, as the last receive
2569                  * times need a full ARP monitor cycle to be updated.
2570                  */
2571                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2572                         continue;
2573
2574                 /* Backup slave is down if:
2575                  * - No current_arp_slave AND
2576                  * - more than 3*delta since last receive AND
2577                  * - the bond has an IP address
2578                  *
2579                  * Note: a non-null current_arp_slave indicates
2580                  * the curr_active_slave went down and we are
2581                  * searching for a new one; under this condition
2582                  * we only take the curr_active_slave down - this
2583                  * gives each slave a chance to tx/rx traffic
2584                  * before being taken out
2585                  */
2586                 if (!bond_is_active_slave(slave) &&
2587                     !rcu_access_pointer(bond->current_arp_slave) &&
2588                     !bond_time_in_interval(bond, last_rx, 3)) {
2589                         slave->new_link = BOND_LINK_DOWN;
2590                         commit++;
2591                 }
2592
2593                 /* Active slave is down if:
2594                  * - more than 2*delta since transmitting OR
2595                  * - (more than 2*delta since receive AND
2596                  *    the bond has an IP address)
2597                  */
2598                 trans_start = dev_trans_start(slave->dev);
2599                 if (bond_is_active_slave(slave) &&
2600                     (!bond_time_in_interval(bond, trans_start, 2) ||
2601                      !bond_time_in_interval(bond, last_rx, 2))) {
2602                         slave->new_link = BOND_LINK_DOWN;
2603                         commit++;
2604                 }
2605         }
2606
2607         return commit;
2608 }
2609
2610 /* Called to commit link state changes noted by inspection step of
2611  * active-backup mode ARP monitor.
2612  *
2613  * Called with RTNL hold.
2614  */
2615 static void bond_ab_arp_commit(struct bonding *bond)
2616 {
2617         unsigned long trans_start;
2618         struct list_head *iter;
2619         struct slave *slave;
2620
2621         bond_for_each_slave(bond, slave, iter) {
2622                 switch (slave->new_link) {
2623                 case BOND_LINK_NOCHANGE:
2624                         continue;
2625
2626                 case BOND_LINK_UP:
2627                         trans_start = dev_trans_start(slave->dev);
2628                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
2629                             (!rtnl_dereference(bond->curr_active_slave) &&
2630                              bond_time_in_interval(bond, trans_start, 1))) {
2631                                 struct slave *current_arp_slave;
2632
2633                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2634                                 bond_set_slave_link_state(slave, BOND_LINK_UP);
2635                                 if (current_arp_slave) {
2636                                         bond_set_slave_inactive_flags(
2637                                                 current_arp_slave,
2638                                                 BOND_SLAVE_NOTIFY_NOW);
2639                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2640                                 }
2641
2642                                 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2643                                             slave->dev->name);
2644
2645                                 if (!rtnl_dereference(bond->curr_active_slave) ||
2646                                     slave == rtnl_dereference(bond->primary_slave))
2647                                         goto do_failover;
2648
2649                         }
2650
2651                         continue;
2652
2653                 case BOND_LINK_DOWN:
2654                         if (slave->link_failure_count < UINT_MAX)
2655                                 slave->link_failure_count++;
2656
2657                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2658                         bond_set_slave_inactive_flags(slave,
2659                                                       BOND_SLAVE_NOTIFY_NOW);
2660
2661                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2662                                     slave->dev->name);
2663
2664                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
2665                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2666                                 goto do_failover;
2667                         }
2668
2669                         continue;
2670
2671                 default:
2672                         netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2673                                    slave->new_link, slave->dev->name);
2674                         continue;
2675                 }
2676
2677 do_failover:
2678                 block_netpoll_tx();
2679                 bond_select_active_slave(bond);
2680                 unblock_netpoll_tx();
2681         }
2682
2683         bond_set_carrier(bond);
2684 }
2685
2686 /* Send ARP probes for active-backup mode ARP monitor.
2687  *
2688  * Called with rcu_read_lock held.
2689  */
2690 static bool bond_ab_arp_probe(struct bonding *bond)
2691 {
2692         struct slave *slave, *before = NULL, *new_slave = NULL,
2693                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2694                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2695         struct list_head *iter;
2696         bool found = false;
2697         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2698
2699         if (curr_arp_slave && curr_active_slave)
2700                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2701                             curr_arp_slave->dev->name,
2702                             curr_active_slave->dev->name);
2703
2704         if (curr_active_slave) {
2705                 bond_arp_send_all(bond, curr_active_slave);
2706                 return should_notify_rtnl;
2707         }
2708
2709         /* if we don't have a curr_active_slave, search for the next available
2710          * backup slave from the current_arp_slave and make it the candidate
2711          * for becoming the curr_active_slave
2712          */
2713
2714         if (!curr_arp_slave) {
2715                 curr_arp_slave = bond_first_slave_rcu(bond);
2716                 if (!curr_arp_slave)
2717                         return should_notify_rtnl;
2718         }
2719
2720         bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2721
2722         bond_for_each_slave_rcu(bond, slave, iter) {
2723                 if (!found && !before && bond_slave_is_up(slave))
2724                         before = slave;
2725
2726                 if (found && !new_slave && bond_slave_is_up(slave))
2727                         new_slave = slave;
2728                 /* if the link state is up at this point, we
2729                  * mark it down - this can happen if we have
2730                  * simultaneous link failures and
2731                  * reselect_active_interface doesn't make this
2732                  * one the current slave so it is still marked
2733                  * up when it is actually down
2734                  */
2735                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2736                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2737                         if (slave->link_failure_count < UINT_MAX)
2738                                 slave->link_failure_count++;
2739
2740                         bond_set_slave_inactive_flags(slave,
2741                                                       BOND_SLAVE_NOTIFY_LATER);
2742
2743                         netdev_info(bond->dev, "backup interface %s is now down\n",
2744                                     slave->dev->name);
2745                 }
2746                 if (slave == curr_arp_slave)
2747                         found = true;
2748         }
2749
2750         if (!new_slave && before)
2751                 new_slave = before;
2752
2753         if (!new_slave)
2754                 goto check_state;
2755
2756         bond_set_slave_link_state(new_slave, BOND_LINK_BACK);
2757         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2758         bond_arp_send_all(bond, new_slave);
2759         new_slave->last_link_up = jiffies;
2760         rcu_assign_pointer(bond->current_arp_slave, new_slave);
2761
2762 check_state:
2763         bond_for_each_slave_rcu(bond, slave, iter) {
2764                 if (slave->should_notify) {
2765                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2766                         break;
2767                 }
2768         }
2769         return should_notify_rtnl;
2770 }
2771
2772 static void bond_activebackup_arp_mon(struct work_struct *work)
2773 {
2774         struct bonding *bond = container_of(work, struct bonding,
2775                                             arp_work.work);
2776         bool should_notify_peers = false;
2777         bool should_notify_rtnl = false;
2778         int delta_in_ticks;
2779
2780         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2781
2782         if (!bond_has_slaves(bond))
2783                 goto re_arm;
2784
2785         rcu_read_lock();
2786
2787         should_notify_peers = bond_should_notify_peers(bond);
2788
2789         if (bond_ab_arp_inspect(bond)) {
2790                 rcu_read_unlock();
2791
2792                 /* Race avoidance with bond_close flush of workqueue */
2793                 if (!rtnl_trylock()) {
2794                         delta_in_ticks = 1;
2795                         should_notify_peers = false;
2796                         goto re_arm;
2797                 }
2798
2799                 bond_ab_arp_commit(bond);
2800
2801                 rtnl_unlock();
2802                 rcu_read_lock();
2803         }
2804
2805         should_notify_rtnl = bond_ab_arp_probe(bond);
2806         rcu_read_unlock();
2807
2808 re_arm:
2809         if (bond->params.arp_interval)
2810                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2811
2812         if (should_notify_peers || should_notify_rtnl) {
2813                 if (!rtnl_trylock())
2814                         return;
2815
2816                 if (should_notify_peers)
2817                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2818                                                  bond->dev);
2819                 if (should_notify_rtnl)
2820                         bond_slave_state_notify(bond);
2821
2822                 rtnl_unlock();
2823         }
2824 }
2825
2826 /*-------------------------- netdev event handling --------------------------*/
2827
2828 /* Change device name */
2829 static int bond_event_changename(struct bonding *bond)
2830 {
2831         bond_remove_proc_entry(bond);
2832         bond_create_proc_entry(bond);
2833
2834         bond_debug_reregister(bond);
2835
2836         return NOTIFY_DONE;
2837 }
2838
2839 static int bond_master_netdev_event(unsigned long event,
2840                                     struct net_device *bond_dev)
2841 {
2842         struct bonding *event_bond = netdev_priv(bond_dev);
2843
2844         switch (event) {
2845         case NETDEV_CHANGENAME:
2846                 return bond_event_changename(event_bond);
2847         case NETDEV_UNREGISTER:
2848                 bond_remove_proc_entry(event_bond);
2849                 break;
2850         case NETDEV_REGISTER:
2851                 bond_create_proc_entry(event_bond);
2852                 break;
2853         case NETDEV_NOTIFY_PEERS:
2854                 if (event_bond->send_peer_notif)
2855                         event_bond->send_peer_notif--;
2856                 break;
2857         default:
2858                 break;
2859         }
2860
2861         return NOTIFY_DONE;
2862 }
2863
2864 static int bond_slave_netdev_event(unsigned long event,
2865                                    struct net_device *slave_dev)
2866 {
2867         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
2868         struct bonding *bond;
2869         struct net_device *bond_dev;
2870         u32 old_speed;
2871         u8 old_duplex;
2872
2873         /* A netdev event can be generated while enslaving a device
2874          * before netdev_rx_handler_register is called in which case
2875          * slave will be NULL
2876          */
2877         if (!slave)
2878                 return NOTIFY_DONE;
2879         bond_dev = slave->bond->dev;
2880         bond = slave->bond;
2881         primary = rtnl_dereference(bond->primary_slave);
2882
2883         switch (event) {
2884         case NETDEV_UNREGISTER:
2885                 if (bond_dev->type != ARPHRD_ETHER)
2886                         bond_release_and_destroy(bond_dev, slave_dev);
2887                 else
2888                         bond_release(bond_dev, slave_dev);
2889                 break;
2890         case NETDEV_UP:
2891         case NETDEV_CHANGE:
2892                 old_speed = slave->speed;
2893                 old_duplex = slave->duplex;
2894
2895                 bond_update_speed_duplex(slave);
2896
2897                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2898                         if (old_speed != slave->speed)
2899                                 bond_3ad_adapter_speed_changed(slave);
2900                         if (old_duplex != slave->duplex)
2901                                 bond_3ad_adapter_duplex_changed(slave);
2902                 }
2903                 /* Refresh slave-array if applicable!
2904                  * If the setup does not use miimon or arpmon (mode-specific!),
2905                  * then these events will not cause the slave-array to be
2906                  * refreshed. This will cause xmit to use a slave that is not
2907                  * usable. Avoid such situation by refeshing the array at these
2908                  * events. If these (miimon/arpmon) parameters are configured
2909                  * then array gets refreshed twice and that should be fine!
2910                  */
2911                 if (bond_mode_uses_xmit_hash(bond))
2912                         bond_update_slave_arr(bond, NULL);
2913                 break;
2914         case NETDEV_DOWN:
2915                 if (bond_mode_uses_xmit_hash(bond))
2916                         bond_update_slave_arr(bond, NULL);
2917                 break;
2918         case NETDEV_CHANGEMTU:
2919                 /* TODO: Should slaves be allowed to
2920                  * independently alter their MTU?  For
2921                  * an active-backup bond, slaves need
2922                  * not be the same type of device, so
2923                  * MTUs may vary.  For other modes,
2924                  * slaves arguably should have the
2925                  * same MTUs. To do this, we'd need to
2926                  * take over the slave's change_mtu
2927                  * function for the duration of their
2928                  * servitude.
2929                  */
2930                 break;
2931         case NETDEV_CHANGENAME:
2932                 /* we don't care if we don't have primary set */
2933                 if (!bond_uses_primary(bond) ||
2934                     !bond->params.primary[0])
2935                         break;
2936
2937                 if (slave == primary) {
2938                         /* slave's name changed - he's no longer primary */
2939                         RCU_INIT_POINTER(bond->primary_slave, NULL);
2940                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2941                         /* we have a new primary slave */
2942                         rcu_assign_pointer(bond->primary_slave, slave);
2943                 } else { /* we didn't change primary - exit */
2944                         break;
2945                 }
2946
2947                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
2948                             primary ? slave_dev->name : "none");
2949
2950                 block_netpoll_tx();
2951                 bond_select_active_slave(bond);
2952                 unblock_netpoll_tx();
2953                 break;
2954         case NETDEV_FEAT_CHANGE:
2955                 bond_compute_features(bond);
2956                 break;
2957         case NETDEV_RESEND_IGMP:
2958                 /* Propagate to master device */
2959                 call_netdevice_notifiers(event, slave->bond->dev);
2960                 break;
2961         default:
2962                 break;
2963         }
2964
2965         return NOTIFY_DONE;
2966 }
2967
2968 /* bond_netdev_event: handle netdev notifier chain events.
2969  *
2970  * This function receives events for the netdev chain.  The caller (an
2971  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2972  * locks for us to safely manipulate the slave devices (RTNL lock,
2973  * dev_probe_lock).
2974  */
2975 static int bond_netdev_event(struct notifier_block *this,
2976                              unsigned long event, void *ptr)
2977 {
2978         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2979
2980         netdev_dbg(event_dev, "event: %lx\n", event);
2981
2982         if (!(event_dev->priv_flags & IFF_BONDING))
2983                 return NOTIFY_DONE;
2984
2985         if (event_dev->flags & IFF_MASTER) {
2986                 netdev_dbg(event_dev, "IFF_MASTER\n");
2987                 return bond_master_netdev_event(event, event_dev);
2988         }
2989
2990         if (event_dev->flags & IFF_SLAVE) {
2991                 netdev_dbg(event_dev, "IFF_SLAVE\n");
2992                 return bond_slave_netdev_event(event, event_dev);
2993         }
2994
2995         return NOTIFY_DONE;
2996 }
2997
2998 static struct notifier_block bond_netdev_notifier = {
2999         .notifier_call = bond_netdev_event,
3000 };
3001
3002 /*---------------------------- Hashing Policies -----------------------------*/
3003
3004 /* L2 hash helper */
3005 static inline u32 bond_eth_hash(struct sk_buff *skb)
3006 {
3007         struct ethhdr *ep, hdr_tmp;
3008
3009         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3010         if (ep)
3011                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3012         return 0;
3013 }
3014
3015 /* Extract the appropriate headers based on bond's xmit policy */
3016 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3017                               struct flow_keys *fk)
3018 {
3019         const struct ipv6hdr *iph6;
3020         const struct iphdr *iph;
3021         int noff, proto = -1;
3022
3023         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3024                 return skb_flow_dissect(skb, fk);
3025
3026         fk->ports = 0;
3027         noff = skb_network_offset(skb);
3028         if (skb->protocol == htons(ETH_P_IP)) {
3029                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3030                         return false;
3031                 iph = ip_hdr(skb);
3032                 fk->src = iph->saddr;
3033                 fk->dst = iph->daddr;
3034                 noff += iph->ihl << 2;
3035                 if (!ip_is_fragment(iph))
3036                         proto = iph->protocol;
3037         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3038                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3039                         return false;
3040                 iph6 = ipv6_hdr(skb);
3041                 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
3042                 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
3043                 noff += sizeof(*iph6);
3044                 proto = iph6->nexthdr;
3045         } else {
3046                 return false;
3047         }
3048         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3049                 fk->ports = skb_flow_get_ports(skb, noff, proto);
3050
3051         return true;
3052 }
3053
3054 /**
3055  * bond_xmit_hash - generate a hash value based on the xmit policy
3056  * @bond: bonding device
3057  * @skb: buffer to use for headers
3058  *
3059  * This function will extract the necessary headers from the skb buffer and use
3060  * them to generate a hash based on the xmit_policy set in the bonding device
3061  */
3062 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3063 {
3064         struct flow_keys flow;
3065         u32 hash;
3066
3067         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3068             !bond_flow_dissect(bond, skb, &flow))
3069                 return bond_eth_hash(skb);
3070
3071         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3072             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3073                 hash = bond_eth_hash(skb);
3074         else
3075                 hash = (__force u32)flow.ports;
3076         hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
3077         hash ^= (hash >> 16);
3078         hash ^= (hash >> 8);
3079
3080         return hash;
3081 }
3082
3083 /*-------------------------- Device entry points ----------------------------*/
3084
3085 static void bond_work_init_all(struct bonding *bond)
3086 {
3087         INIT_DELAYED_WORK(&bond->mcast_work,
3088                           bond_resend_igmp_join_requests_delayed);
3089         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3090         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3091         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3092                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3093         else
3094                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3095         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3096         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3097 }
3098
3099 static void bond_work_cancel_all(struct bonding *bond)
3100 {
3101         cancel_delayed_work_sync(&bond->mii_work);
3102         cancel_delayed_work_sync(&bond->arp_work);
3103         cancel_delayed_work_sync(&bond->alb_work);
3104         cancel_delayed_work_sync(&bond->ad_work);
3105         cancel_delayed_work_sync(&bond->mcast_work);
3106         cancel_delayed_work_sync(&bond->slave_arr_work);
3107 }
3108
3109 static int bond_open(struct net_device *bond_dev)
3110 {
3111         struct bonding *bond = netdev_priv(bond_dev);
3112         struct list_head *iter;
3113         struct slave *slave;
3114
3115         /* reset slave->backup and slave->inactive */
3116         if (bond_has_slaves(bond)) {
3117                 bond_for_each_slave(bond, slave, iter) {
3118                         if (bond_uses_primary(bond) &&
3119                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3120                                 bond_set_slave_inactive_flags(slave,
3121                                                               BOND_SLAVE_NOTIFY_NOW);
3122                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3123                                 bond_set_slave_active_flags(slave,
3124                                                             BOND_SLAVE_NOTIFY_NOW);
3125                         }
3126                 }
3127         }
3128
3129         bond_work_init_all(bond);
3130
3131         if (bond_is_lb(bond)) {
3132                 /* bond_alb_initialize must be called before the timer
3133                  * is started.
3134                  */
3135                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3136                         return -ENOMEM;
3137                 if (bond->params.tlb_dynamic_lb)
3138                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3139         }
3140
3141         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3142                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3143
3144         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3145                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3146                 bond->recv_probe = bond_arp_rcv;
3147         }
3148
3149         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3150                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3151                 /* register to receive LACPDUs */
3152                 bond->recv_probe = bond_3ad_lacpdu_recv;
3153                 bond_3ad_initiate_agg_selection(bond, 1);
3154         }
3155
3156         if (bond_mode_uses_xmit_hash(bond))
3157                 bond_update_slave_arr(bond, NULL);
3158
3159         return 0;
3160 }
3161
3162 static int bond_close(struct net_device *bond_dev)
3163 {
3164         struct bonding *bond = netdev_priv(bond_dev);
3165
3166         bond_work_cancel_all(bond);
3167         bond->send_peer_notif = 0;
3168         if (bond_is_lb(bond))
3169                 bond_alb_deinitialize(bond);
3170         bond->recv_probe = NULL;
3171
3172         return 0;
3173 }
3174
3175 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3176                                                 struct rtnl_link_stats64 *stats)
3177 {
3178         struct bonding *bond = netdev_priv(bond_dev);
3179         struct rtnl_link_stats64 temp;
3180         struct list_head *iter;
3181         struct slave *slave;
3182
3183         memcpy(stats, &bond->bond_stats, sizeof(*stats));
3184
3185         bond_for_each_slave(bond, slave, iter) {
3186                 const struct rtnl_link_stats64 *sstats =
3187                         dev_get_stats(slave->dev, &temp);
3188                 struct rtnl_link_stats64 *pstats = &slave->slave_stats;
3189
3190                 stats->rx_packets +=  sstats->rx_packets - pstats->rx_packets;
3191                 stats->rx_bytes += sstats->rx_bytes - pstats->rx_bytes;
3192                 stats->rx_errors += sstats->rx_errors - pstats->rx_errors;
3193                 stats->rx_dropped += sstats->rx_dropped - pstats->rx_dropped;
3194
3195                 stats->tx_packets += sstats->tx_packets - pstats->tx_packets;;
3196                 stats->tx_bytes += sstats->tx_bytes - pstats->tx_bytes;
3197                 stats->tx_errors += sstats->tx_errors - pstats->tx_errors;
3198                 stats->tx_dropped += sstats->tx_dropped - pstats->tx_dropped;
3199
3200                 stats->multicast += sstats->multicast - pstats->multicast;
3201                 stats->collisions += sstats->collisions - pstats->collisions;
3202
3203                 stats->rx_length_errors += sstats->rx_length_errors - pstats->rx_length_errors;
3204                 stats->rx_over_errors += sstats->rx_over_errors - pstats->rx_over_errors;
3205                 stats->rx_crc_errors += sstats->rx_crc_errors - pstats->rx_crc_errors;
3206                 stats->rx_frame_errors += sstats->rx_frame_errors - pstats->rx_frame_errors;
3207                 stats->rx_fifo_errors += sstats->rx_fifo_errors - pstats->rx_fifo_errors;
3208                 stats->rx_missed_errors += sstats->rx_missed_errors - pstats->rx_missed_errors;
3209
3210                 stats->tx_aborted_errors += sstats->tx_aborted_errors - pstats->tx_aborted_errors;
3211                 stats->tx_carrier_errors += sstats->tx_carrier_errors - pstats->tx_carrier_errors;
3212                 stats->tx_fifo_errors += sstats->tx_fifo_errors - pstats->tx_fifo_errors;
3213                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors - pstats->tx_heartbeat_errors;
3214                 stats->tx_window_errors += sstats->tx_window_errors - pstats->tx_window_errors;
3215
3216                 /* save off the slave stats for the next run */
3217                 memcpy(pstats, sstats, sizeof(*sstats));
3218         }
3219         memcpy(&bond->bond_stats, stats, sizeof(*stats));
3220
3221         return stats;
3222 }
3223
3224 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3225 {
3226         struct bonding *bond = netdev_priv(bond_dev);
3227         struct net_device *slave_dev = NULL;
3228         struct ifbond k_binfo;
3229         struct ifbond __user *u_binfo = NULL;
3230         struct ifslave k_sinfo;
3231         struct ifslave __user *u_sinfo = NULL;
3232         struct mii_ioctl_data *mii = NULL;
3233         struct bond_opt_value newval;
3234         struct net *net;
3235         int res = 0;
3236
3237         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3238
3239         switch (cmd) {
3240         case SIOCGMIIPHY:
3241                 mii = if_mii(ifr);
3242                 if (!mii)
3243                         return -EINVAL;
3244
3245                 mii->phy_id = 0;
3246                 /* Fall Through */
3247         case SIOCGMIIREG:
3248                 /* We do this again just in case we were called by SIOCGMIIREG
3249                  * instead of SIOCGMIIPHY.
3250                  */
3251                 mii = if_mii(ifr);
3252                 if (!mii)
3253                         return -EINVAL;
3254
3255                 if (mii->reg_num == 1) {
3256                         mii->val_out = 0;
3257                         if (netif_carrier_ok(bond->dev))
3258                                 mii->val_out = BMSR_LSTATUS;
3259                 }
3260
3261                 return 0;
3262         case BOND_INFO_QUERY_OLD:
3263         case SIOCBONDINFOQUERY:
3264                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3265
3266                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3267                         return -EFAULT;
3268
3269                 res = bond_info_query(bond_dev, &k_binfo);
3270                 if (res == 0 &&
3271                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3272                         return -EFAULT;
3273
3274                 return res;
3275         case BOND_SLAVE_INFO_QUERY_OLD:
3276         case SIOCBONDSLAVEINFOQUERY:
3277                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3278
3279                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3280                         return -EFAULT;
3281
3282                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3283                 if (res == 0 &&
3284                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3285                         return -EFAULT;
3286
3287                 return res;
3288         default:
3289                 break;
3290         }
3291
3292         net = dev_net(bond_dev);
3293
3294         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3295                 return -EPERM;
3296
3297         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3298
3299         netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3300
3301         if (!slave_dev)
3302                 return -ENODEV;
3303
3304         netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3305         switch (cmd) {
3306         case BOND_ENSLAVE_OLD:
3307         case SIOCBONDENSLAVE:
3308                 res = bond_enslave(bond_dev, slave_dev);
3309                 break;
3310         case BOND_RELEASE_OLD:
3311         case SIOCBONDRELEASE:
3312                 res = bond_release(bond_dev, slave_dev);
3313                 break;
3314         case BOND_SETHWADDR_OLD:
3315         case SIOCBONDSETHWADDR:
3316                 bond_set_dev_addr(bond_dev, slave_dev);
3317                 res = 0;
3318                 break;
3319         case BOND_CHANGE_ACTIVE_OLD:
3320         case SIOCBONDCHANGEACTIVE:
3321                 bond_opt_initstr(&newval, slave_dev->name);
3322                 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3323                 break;
3324         default:
3325                 res = -EOPNOTSUPP;
3326         }
3327
3328         return res;
3329 }
3330
3331 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3332 {
3333         struct bonding *bond = netdev_priv(bond_dev);
3334
3335         if (change & IFF_PROMISC)
3336                 bond_set_promiscuity(bond,
3337                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3338
3339         if (change & IFF_ALLMULTI)
3340                 bond_set_allmulti(bond,
3341                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3342 }
3343
3344 static void bond_set_rx_mode(struct net_device *bond_dev)
3345 {
3346         struct bonding *bond = netdev_priv(bond_dev);
3347         struct list_head *iter;
3348         struct slave *slave;
3349
3350         rcu_read_lock();
3351         if (bond_uses_primary(bond)) {
3352                 slave = rcu_dereference(bond->curr_active_slave);
3353                 if (slave) {
3354                         dev_uc_sync(slave->dev, bond_dev);
3355                         dev_mc_sync(slave->dev, bond_dev);
3356                 }
3357         } else {
3358                 bond_for_each_slave_rcu(bond, slave, iter) {
3359                         dev_uc_sync_multiple(slave->dev, bond_dev);
3360                         dev_mc_sync_multiple(slave->dev, bond_dev);
3361                 }
3362         }
3363         rcu_read_unlock();
3364 }
3365
3366 static int bond_neigh_init(struct neighbour *n)
3367 {
3368         struct bonding *bond = netdev_priv(n->dev);
3369         const struct net_device_ops *slave_ops;
3370         struct neigh_parms parms;
3371         struct slave *slave;
3372         int ret;
3373
3374         slave = bond_first_slave(bond);
3375         if (!slave)
3376                 return 0;
3377         slave_ops = slave->dev->netdev_ops;
3378         if (!slave_ops->ndo_neigh_setup)
3379                 return 0;
3380
3381         parms.neigh_setup = NULL;
3382         parms.neigh_cleanup = NULL;
3383         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3384         if (ret)
3385                 return ret;
3386
3387         /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3388          * after the last slave has been detached.  Assumes that all slaves
3389          * utilize the same neigh_cleanup (true at this writing as only user
3390          * is ipoib).
3391          */
3392         n->parms->neigh_cleanup = parms.neigh_cleanup;
3393
3394         if (!parms.neigh_setup)
3395                 return 0;
3396
3397         return parms.neigh_setup(n);
3398 }
3399
3400 /* The bonding ndo_neigh_setup is called at init time beofre any
3401  * slave exists. So we must declare proxy setup function which will
3402  * be used at run time to resolve the actual slave neigh param setup.
3403  *
3404  * It's also called by master devices (such as vlans) to setup their
3405  * underlying devices. In that case - do nothing, we're already set up from
3406  * our init.
3407  */
3408 static int bond_neigh_setup(struct net_device *dev,
3409                             struct neigh_parms *parms)
3410 {
3411         /* modify only our neigh_parms */
3412         if (parms->dev == dev)
3413                 parms->neigh_setup = bond_neigh_init;
3414
3415         return 0;
3416 }
3417
3418 /* Change the MTU of all of a master's slaves to match the master */
3419 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3420 {
3421         struct bonding *bond = netdev_priv(bond_dev);
3422         struct slave *slave, *rollback_slave;
3423         struct list_head *iter;
3424         int res = 0;
3425
3426         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3427
3428         bond_for_each_slave(bond, slave, iter) {
3429                 netdev_dbg(bond_dev, "s %p c_m %p\n",
3430                            slave, slave->dev->netdev_ops->ndo_change_mtu);
3431
3432                 res = dev_set_mtu(slave->dev, new_mtu);
3433
3434                 if (res) {
3435                         /* If we failed to set the slave's mtu to the new value
3436                          * we must abort the operation even in ACTIVE_BACKUP
3437                          * mode, because if we allow the backup slaves to have
3438                          * different mtu values than the active slave we'll
3439                          * need to change their mtu when doing a failover. That
3440                          * means changing their mtu from timer context, which
3441                          * is probably not a good idea.
3442                          */
3443                         netdev_dbg(bond_dev, "err %d %s\n", res,
3444                                    slave->dev->name);
3445                         goto unwind;
3446                 }
3447         }
3448
3449         bond_dev->mtu = new_mtu;
3450
3451         return 0;
3452
3453 unwind:
3454         /* unwind from head to the slave that failed */
3455         bond_for_each_slave(bond, rollback_slave, iter) {
3456                 int tmp_res;
3457
3458                 if (rollback_slave == slave)
3459                         break;
3460
3461                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3462                 if (tmp_res) {
3463                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3464                                    tmp_res, rollback_slave->dev->name);
3465                 }
3466         }
3467
3468         return res;
3469 }
3470
3471 /* Change HW address
3472  *
3473  * Note that many devices must be down to change the HW address, and
3474  * downing the master releases all slaves.  We can make bonds full of
3475  * bonding devices to test this, however.
3476  */
3477 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3478 {
3479         struct bonding *bond = netdev_priv(bond_dev);
3480         struct slave *slave, *rollback_slave;
3481         struct sockaddr *sa = addr, tmp_sa;
3482         struct list_head *iter;
3483         int res = 0;
3484
3485         if (BOND_MODE(bond) == BOND_MODE_ALB)
3486                 return bond_alb_set_mac_address(bond_dev, addr);
3487
3488
3489         netdev_dbg(bond_dev, "bond=%p\n", bond);
3490
3491         /* If fail_over_mac is enabled, do nothing and return success.
3492          * Returning an error causes ifenslave to fail.
3493          */
3494         if (bond->params.fail_over_mac &&
3495             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3496                 return 0;
3497
3498         if (!is_valid_ether_addr(sa->sa_data))
3499                 return -EADDRNOTAVAIL;
3500
3501         bond_for_each_slave(bond, slave, iter) {
3502                 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3503                 res = dev_set_mac_address(slave->dev, addr);
3504                 if (res) {
3505                         /* TODO: consider downing the slave
3506                          * and retry ?
3507                          * User should expect communications
3508                          * breakage anyway until ARP finish
3509                          * updating, so...
3510                          */
3511                         netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3512                         goto unwind;
3513                 }
3514         }
3515
3516         /* success */
3517         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3518         return 0;
3519
3520 unwind:
3521         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3522         tmp_sa.sa_family = bond_dev->type;
3523
3524         /* unwind from head to the slave that failed */
3525         bond_for_each_slave(bond, rollback_slave, iter) {
3526                 int tmp_res;
3527
3528                 if (rollback_slave == slave)
3529                         break;
3530
3531                 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3532                 if (tmp_res) {
3533                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3534                                    tmp_res, rollback_slave->dev->name);
3535                 }
3536         }
3537
3538         return res;
3539 }
3540
3541 /**
3542  * bond_xmit_slave_id - transmit skb through slave with slave_id
3543  * @bond: bonding device that is transmitting
3544  * @skb: buffer to transmit
3545  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3546  *
3547  * This function tries to transmit through slave with slave_id but in case
3548  * it fails, it tries to find the first available slave for transmission.
3549  * The skb is consumed in all cases, thus the function is void.
3550  */
3551 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3552 {
3553         struct list_head *iter;
3554         struct slave *slave;
3555         int i = slave_id;
3556
3557         /* Here we start from the slave with slave_id */
3558         bond_for_each_slave_rcu(bond, slave, iter) {
3559                 if (--i < 0) {
3560                         if (bond_slave_can_tx(slave)) {
3561                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3562                                 return;
3563                         }
3564                 }
3565         }
3566
3567         /* Here we start from the first slave up to slave_id */
3568         i = slave_id;
3569         bond_for_each_slave_rcu(bond, slave, iter) {
3570                 if (--i < 0)
3571                         break;
3572                 if (bond_slave_can_tx(slave)) {
3573                         bond_dev_queue_xmit(bond, skb, slave->dev);
3574                         return;
3575                 }
3576         }
3577         /* no slave that can tx has been found */
3578         bond_tx_drop(bond->dev, skb);
3579 }
3580
3581 /**
3582  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3583  * @bond: bonding device to use
3584  *
3585  * Based on the value of the bonding device's packets_per_slave parameter
3586  * this function generates a slave id, which is usually used as the next
3587  * slave to transmit through.
3588  */
3589 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3590 {
3591         u32 slave_id;
3592         struct reciprocal_value reciprocal_packets_per_slave;
3593         int packets_per_slave = bond->params.packets_per_slave;
3594
3595         switch (packets_per_slave) {
3596         case 0:
3597                 slave_id = prandom_u32();
3598                 break;
3599         case 1:
3600                 slave_id = bond->rr_tx_counter;
3601                 break;
3602         default:
3603                 reciprocal_packets_per_slave =
3604                         bond->params.reciprocal_packets_per_slave;
3605                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3606                                              reciprocal_packets_per_slave);
3607                 break;
3608         }
3609         bond->rr_tx_counter++;
3610
3611         return slave_id;
3612 }
3613
3614 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3615 {
3616         struct bonding *bond = netdev_priv(bond_dev);
3617         struct iphdr *iph = ip_hdr(skb);
3618         struct slave *slave;
3619         u32 slave_id;
3620
3621         /* Start with the curr_active_slave that joined the bond as the
3622          * default for sending IGMP traffic.  For failover purposes one
3623          * needs to maintain some consistency for the interface that will
3624          * send the join/membership reports.  The curr_active_slave found
3625          * will send all of this type of traffic.
3626          */
3627         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3628                 slave = rcu_dereference(bond->curr_active_slave);
3629                 if (slave)
3630                         bond_dev_queue_xmit(bond, skb, slave->dev);
3631                 else
3632                         bond_xmit_slave_id(bond, skb, 0);
3633         } else {
3634                 int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3635
3636                 if (likely(slave_cnt)) {
3637                         slave_id = bond_rr_gen_slave_id(bond);
3638                         bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3639                 } else {
3640                         bond_tx_drop(bond_dev, skb);
3641                 }
3642         }
3643
3644         return NETDEV_TX_OK;
3645 }
3646
3647 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3648  * the bond has a usable interface.
3649  */
3650 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3651 {
3652         struct bonding *bond = netdev_priv(bond_dev);
3653         struct slave *slave;
3654
3655         slave = rcu_dereference(bond->curr_active_slave);
3656         if (slave)
3657                 bond_dev_queue_xmit(bond, skb, slave->dev);
3658         else
3659                 bond_tx_drop(bond_dev, skb);
3660
3661         return NETDEV_TX_OK;
3662 }
3663
3664 /* Use this to update slave_array when (a) it's not appropriate to update
3665  * slave_array right away (note that update_slave_array() may sleep)
3666  * and / or (b) RTNL is not held.
3667  */
3668 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3669 {
3670         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3671 }
3672
3673 /* Slave array work handler. Holds only RTNL */
3674 static void bond_slave_arr_handler(struct work_struct *work)
3675 {
3676         struct bonding *bond = container_of(work, struct bonding,
3677                                             slave_arr_work.work);
3678         int ret;
3679
3680         if (!rtnl_trylock())
3681                 goto err;
3682
3683         ret = bond_update_slave_arr(bond, NULL);
3684         rtnl_unlock();
3685         if (ret) {
3686                 pr_warn_ratelimited("Failed to update slave array from WT\n");
3687                 goto err;
3688         }
3689         return;
3690
3691 err:
3692         bond_slave_arr_work_rearm(bond, 1);
3693 }
3694
3695 /* Build the usable slaves array in control path for modes that use xmit-hash
3696  * to determine the slave interface -
3697  * (a) BOND_MODE_8023AD
3698  * (b) BOND_MODE_XOR
3699  * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3700  *
3701  * The caller is expected to hold RTNL only and NO other lock!
3702  */
3703 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3704 {
3705         struct slave *slave;
3706         struct list_head *iter;
3707         struct bond_up_slave *new_arr, *old_arr;
3708         int slaves_in_agg;
3709         int agg_id = 0;
3710         int ret = 0;
3711
3712 #ifdef CONFIG_LOCKDEP
3713         WARN_ON(lockdep_is_held(&bond->mode_lock));
3714 #endif
3715
3716         new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3717                           GFP_KERNEL);
3718         if (!new_arr) {
3719                 ret = -ENOMEM;
3720                 pr_err("Failed to build slave-array.\n");
3721                 goto out;
3722         }
3723         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3724                 struct ad_info ad_info;
3725
3726                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3727                         pr_debug("bond_3ad_get_active_agg_info failed\n");
3728                         kfree_rcu(new_arr, rcu);
3729                         /* No active aggragator means it's not safe to use
3730                          * the previous array.
3731                          */
3732                         old_arr = rtnl_dereference(bond->slave_arr);
3733                         if (old_arr) {
3734                                 RCU_INIT_POINTER(bond->slave_arr, NULL);
3735                                 kfree_rcu(old_arr, rcu);
3736                         }
3737                         goto out;
3738                 }
3739                 slaves_in_agg = ad_info.ports;
3740                 agg_id = ad_info.aggregator_id;
3741         }
3742         bond_for_each_slave(bond, slave, iter) {
3743                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3744                         struct aggregator *agg;
3745
3746                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
3747                         if (!agg || agg->aggregator_identifier != agg_id)
3748                                 continue;
3749                 }
3750                 if (!bond_slave_can_tx(slave))
3751                         continue;
3752                 if (skipslave == slave)
3753                         continue;
3754                 new_arr->arr[new_arr->count++] = slave;
3755         }
3756
3757         old_arr = rtnl_dereference(bond->slave_arr);
3758         rcu_assign_pointer(bond->slave_arr, new_arr);
3759         if (old_arr)
3760                 kfree_rcu(old_arr, rcu);
3761 out:
3762         if (ret != 0 && skipslave) {
3763                 int idx;
3764
3765                 /* Rare situation where caller has asked to skip a specific
3766                  * slave but allocation failed (most likely!). BTW this is
3767                  * only possible when the call is initiated from
3768                  * __bond_release_one(). In this situation; overwrite the
3769                  * skipslave entry in the array with the last entry from the
3770                  * array to avoid a situation where the xmit path may choose
3771                  * this to-be-skipped slave to send a packet out.
3772                  */
3773                 old_arr = rtnl_dereference(bond->slave_arr);
3774                 for (idx = 0; idx < old_arr->count; idx++) {
3775                         if (skipslave == old_arr->arr[idx]) {
3776                                 old_arr->arr[idx] =
3777                                     old_arr->arr[old_arr->count-1];
3778                                 old_arr->count--;
3779                                 break;
3780                         }
3781                 }
3782         }
3783         return ret;
3784 }
3785
3786 /* Use this Xmit function for 3AD as well as XOR modes. The current
3787  * usable slave array is formed in the control path. The xmit function
3788  * just calculates hash and sends the packet out.
3789  */
3790 static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3791 {
3792         struct bonding *bond = netdev_priv(dev);
3793         struct slave *slave;
3794         struct bond_up_slave *slaves;
3795         unsigned int count;
3796
3797         slaves = rcu_dereference(bond->slave_arr);
3798         count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3799         if (likely(count)) {
3800                 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3801                 bond_dev_queue_xmit(bond, skb, slave->dev);
3802         } else {
3803                 bond_tx_drop(dev, skb);
3804         }
3805
3806         return NETDEV_TX_OK;
3807 }
3808
3809 /* in broadcast mode, we send everything to all usable interfaces. */
3810 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3811 {
3812         struct bonding *bond = netdev_priv(bond_dev);
3813         struct slave *slave = NULL;
3814         struct list_head *iter;
3815
3816         bond_for_each_slave_rcu(bond, slave, iter) {
3817                 if (bond_is_last_slave(bond, slave))
3818                         break;
3819                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3820                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3821
3822                         if (!skb2) {
3823                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3824                                                     bond_dev->name, __func__);
3825                                 continue;
3826                         }
3827                         bond_dev_queue_xmit(bond, skb2, slave->dev);
3828                 }
3829         }
3830         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3831                 bond_dev_queue_xmit(bond, skb, slave->dev);
3832         else
3833                 bond_tx_drop(bond_dev, skb);
3834
3835         return NETDEV_TX_OK;
3836 }
3837
3838 /*------------------------- Device initialization ---------------------------*/
3839
3840 /* Lookup the slave that corresponds to a qid */
3841 static inline int bond_slave_override(struct bonding *bond,
3842                                       struct sk_buff *skb)
3843 {
3844         struct slave *slave = NULL;
3845         struct list_head *iter;
3846
3847         if (!skb->queue_mapping)
3848                 return 1;
3849
3850         /* Find out if any slaves have the same mapping as this skb. */
3851         bond_for_each_slave_rcu(bond, slave, iter) {
3852                 if (slave->queue_id == skb->queue_mapping) {
3853                         if (bond_slave_is_up(slave) &&
3854                             slave->link == BOND_LINK_UP) {
3855                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3856                                 return 0;
3857                         }
3858                         /* If the slave isn't UP, use default transmit policy. */
3859                         break;
3860                 }
3861         }
3862
3863         return 1;
3864 }
3865
3866
3867 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3868                              void *accel_priv, select_queue_fallback_t fallback)
3869 {
3870         /* This helper function exists to help dev_pick_tx get the correct
3871          * destination queue.  Using a helper function skips a call to
3872          * skb_tx_hash and will put the skbs in the queue we expect on their
3873          * way down to the bonding driver.
3874          */
3875         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3876
3877         /* Save the original txq to restore before passing to the driver */
3878         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3879
3880         if (unlikely(txq >= dev->real_num_tx_queues)) {
3881                 do {
3882                         txq -= dev->real_num_tx_queues;
3883                 } while (txq >= dev->real_num_tx_queues);
3884         }
3885         return txq;
3886 }
3887
3888 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3889 {
3890         struct bonding *bond = netdev_priv(dev);
3891
3892         if (bond_should_override_tx_queue(bond) &&
3893             !bond_slave_override(bond, skb))
3894                 return NETDEV_TX_OK;
3895
3896         switch (BOND_MODE(bond)) {
3897         case BOND_MODE_ROUNDROBIN:
3898                 return bond_xmit_roundrobin(skb, dev);
3899         case BOND_MODE_ACTIVEBACKUP:
3900                 return bond_xmit_activebackup(skb, dev);
3901         case BOND_MODE_8023AD:
3902         case BOND_MODE_XOR:
3903                 return bond_3ad_xor_xmit(skb, dev);
3904         case BOND_MODE_BROADCAST:
3905                 return bond_xmit_broadcast(skb, dev);
3906         case BOND_MODE_ALB:
3907                 return bond_alb_xmit(skb, dev);
3908         case BOND_MODE_TLB:
3909                 return bond_tlb_xmit(skb, dev);
3910         default:
3911                 /* Should never happen, mode already checked */
3912                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
3913                 WARN_ON_ONCE(1);
3914                 bond_tx_drop(dev, skb);
3915                 return NETDEV_TX_OK;
3916         }
3917 }
3918
3919 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3920 {
3921         struct bonding *bond = netdev_priv(dev);
3922         netdev_tx_t ret = NETDEV_TX_OK;
3923
3924         /* If we risk deadlock from transmitting this in the
3925          * netpoll path, tell netpoll to queue the frame for later tx
3926          */
3927         if (unlikely(is_netpoll_tx_blocked(dev)))
3928                 return NETDEV_TX_BUSY;
3929
3930         rcu_read_lock();
3931         if (bond_has_slaves(bond))
3932                 ret = __bond_start_xmit(skb, dev);
3933         else
3934                 bond_tx_drop(dev, skb);
3935         rcu_read_unlock();
3936
3937         return ret;
3938 }
3939
3940 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3941                                      struct ethtool_cmd *ecmd)
3942 {
3943         struct bonding *bond = netdev_priv(bond_dev);
3944         unsigned long speed = 0;
3945         struct list_head *iter;
3946         struct slave *slave;
3947
3948         ecmd->duplex = DUPLEX_UNKNOWN;
3949         ecmd->port = PORT_OTHER;
3950
3951         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
3952          * do not need to check mode.  Though link speed might not represent
3953          * the true receive or transmit bandwidth (not all modes are symmetric)
3954          * this is an accurate maximum.
3955          */
3956         bond_for_each_slave(bond, slave, iter) {
3957                 if (bond_slave_can_tx(slave)) {
3958                         if (slave->speed != SPEED_UNKNOWN)
3959                                 speed += slave->speed;
3960                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
3961                             slave->duplex != DUPLEX_UNKNOWN)
3962                                 ecmd->duplex = slave->duplex;
3963                 }
3964         }
3965         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3966
3967         return 0;
3968 }
3969
3970 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3971                                      struct ethtool_drvinfo *drvinfo)
3972 {
3973         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3974         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3975         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3976                  BOND_ABI_VERSION);
3977 }
3978
3979 static const struct ethtool_ops bond_ethtool_ops = {
3980         .get_drvinfo            = bond_ethtool_get_drvinfo,
3981         .get_settings           = bond_ethtool_get_settings,
3982         .get_link               = ethtool_op_get_link,
3983 };
3984
3985 static const struct net_device_ops bond_netdev_ops = {
3986         .ndo_init               = bond_init,
3987         .ndo_uninit             = bond_uninit,
3988         .ndo_open               = bond_open,
3989         .ndo_stop               = bond_close,
3990         .ndo_start_xmit         = bond_start_xmit,
3991         .ndo_select_queue       = bond_select_queue,
3992         .ndo_get_stats64        = bond_get_stats,
3993         .ndo_do_ioctl           = bond_do_ioctl,
3994         .ndo_change_rx_flags    = bond_change_rx_flags,
3995         .ndo_set_rx_mode        = bond_set_rx_mode,
3996         .ndo_change_mtu         = bond_change_mtu,
3997         .ndo_set_mac_address    = bond_set_mac_address,
3998         .ndo_neigh_setup        = bond_neigh_setup,
3999         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4000         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4001 #ifdef CONFIG_NET_POLL_CONTROLLER
4002         .ndo_netpoll_setup      = bond_netpoll_setup,
4003         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4004         .ndo_poll_controller    = bond_poll_controller,
4005 #endif
4006         .ndo_add_slave          = bond_enslave,
4007         .ndo_del_slave          = bond_release,
4008         .ndo_fix_features       = bond_fix_features,
4009         .ndo_bridge_setlink     = ndo_dflt_netdev_switch_port_bridge_setlink,
4010         .ndo_bridge_dellink     = ndo_dflt_netdev_switch_port_bridge_dellink,
4011 };
4012
4013 static const struct device_type bond_type = {
4014         .name = "bond",
4015 };
4016
4017 static void bond_destructor(struct net_device *bond_dev)
4018 {
4019         struct bonding *bond = netdev_priv(bond_dev);
4020         if (bond->wq)
4021                 destroy_workqueue(bond->wq);
4022         free_netdev(bond_dev);
4023 }
4024
4025 void bond_setup(struct net_device *bond_dev)
4026 {
4027         struct bonding *bond = netdev_priv(bond_dev);
4028
4029         spin_lock_init(&bond->mode_lock);
4030         bond->params = bonding_defaults;
4031
4032         /* Initialize pointers */
4033         bond->dev = bond_dev;
4034
4035         /* Initialize the device entry points */
4036         ether_setup(bond_dev);
4037         bond_dev->netdev_ops = &bond_netdev_ops;
4038         bond_dev->ethtool_ops = &bond_ethtool_ops;
4039
4040         bond_dev->destructor = bond_destructor;
4041
4042         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4043
4044         /* Initialize the device options */
4045         bond_dev->tx_queue_len = 0;
4046         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4047         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT;
4048         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4049
4050         /* don't acquire bond device's netif_tx_lock when transmitting */
4051         bond_dev->features |= NETIF_F_LLTX;
4052
4053         /* By default, we declare the bond to be fully
4054          * VLAN hardware accelerated capable. Special
4055          * care is taken in the various xmit functions
4056          * when there are slaves that are not hw accel
4057          * capable
4058          */
4059
4060         /* Don't allow bond devices to change network namespaces. */
4061         bond_dev->features |= NETIF_F_NETNS_LOCAL;
4062
4063         bond_dev->hw_features = BOND_VLAN_FEATURES |
4064                                 NETIF_F_HW_VLAN_CTAG_TX |
4065                                 NETIF_F_HW_VLAN_CTAG_RX |
4066                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4067
4068         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4069         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4070         bond_dev->features |= bond_dev->hw_features;
4071 }
4072
4073 /* Destroy a bonding device.
4074  * Must be under rtnl_lock when this function is called.
4075  */
4076 static void bond_uninit(struct net_device *bond_dev)
4077 {
4078         struct bonding *bond = netdev_priv(bond_dev);
4079         struct list_head *iter;
4080         struct slave *slave;
4081         struct bond_up_slave *arr;
4082
4083         bond_netpoll_cleanup(bond_dev);
4084
4085         /* Release the bonded slaves */
4086         bond_for_each_slave(bond, slave, iter)
4087                 __bond_release_one(bond_dev, slave->dev, true);
4088         netdev_info(bond_dev, "Released all slaves\n");
4089
4090         arr = rtnl_dereference(bond->slave_arr);
4091         if (arr) {
4092                 RCU_INIT_POINTER(bond->slave_arr, NULL);
4093                 kfree_rcu(arr, rcu);
4094         }
4095
4096         list_del(&bond->bond_list);
4097
4098         bond_debug_unregister(bond);
4099 }
4100
4101 /*------------------------- Module initialization ---------------------------*/
4102
4103 static int bond_check_params(struct bond_params *params)
4104 {
4105         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4106         struct bond_opt_value newval;
4107         const struct bond_opt_value *valptr;
4108         int arp_all_targets_value;
4109
4110         /* Convert string parameters. */
4111         if (mode) {
4112                 bond_opt_initstr(&newval, mode);
4113                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4114                 if (!valptr) {
4115                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4116                         return -EINVAL;
4117                 }
4118                 bond_mode = valptr->value;
4119         }
4120
4121         if (xmit_hash_policy) {
4122                 if ((bond_mode != BOND_MODE_XOR) &&
4123                     (bond_mode != BOND_MODE_8023AD) &&
4124                     (bond_mode != BOND_MODE_TLB)) {
4125                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4126                                 bond_mode_name(bond_mode));
4127                 } else {
4128                         bond_opt_initstr(&newval, xmit_hash_policy);
4129                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4130                                                 &newval);
4131                         if (!valptr) {
4132                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4133                                        xmit_hash_policy);
4134                                 return -EINVAL;
4135                         }
4136                         xmit_hashtype = valptr->value;
4137                 }
4138         }
4139
4140         if (lacp_rate) {
4141                 if (bond_mode != BOND_MODE_8023AD) {
4142                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4143                                 bond_mode_name(bond_mode));
4144                 } else {
4145                         bond_opt_initstr(&newval, lacp_rate);
4146                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4147                                                 &newval);
4148                         if (!valptr) {
4149                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4150                                        lacp_rate);
4151                                 return -EINVAL;
4152                         }
4153                         lacp_fast = valptr->value;
4154                 }
4155         }
4156
4157         if (ad_select) {
4158                 bond_opt_initstr(&newval, ad_select);
4159                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4160                                         &newval);
4161                 if (!valptr) {
4162                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4163                         return -EINVAL;
4164                 }
4165                 params->ad_select = valptr->value;
4166                 if (bond_mode != BOND_MODE_8023AD)
4167                         pr_warn("ad_select param only affects 802.3ad mode\n");
4168         } else {
4169                 params->ad_select = BOND_AD_STABLE;
4170         }
4171
4172         if (max_bonds < 0) {
4173                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4174                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4175                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4176         }
4177
4178         if (miimon < 0) {
4179                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4180                         miimon, INT_MAX);
4181                 miimon = 0;
4182         }
4183
4184         if (updelay < 0) {
4185                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4186                         updelay, INT_MAX);
4187                 updelay = 0;
4188         }
4189
4190         if (downdelay < 0) {
4191                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4192                         downdelay, INT_MAX);
4193                 downdelay = 0;
4194         }
4195
4196         if ((use_carrier != 0) && (use_carrier != 1)) {
4197                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4198                         use_carrier);
4199                 use_carrier = 1;
4200         }
4201
4202         if (num_peer_notif < 0 || num_peer_notif > 255) {
4203                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4204                         num_peer_notif);
4205                 num_peer_notif = 1;
4206         }
4207
4208         /* reset values for 802.3ad/TLB/ALB */
4209         if (!bond_mode_uses_arp(bond_mode)) {
4210                 if (!miimon) {
4211                         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");
4212                         pr_warn("Forcing miimon to 100msec\n");
4213                         miimon = BOND_DEFAULT_MIIMON;
4214                 }
4215         }
4216
4217         if (tx_queues < 1 || tx_queues > 255) {
4218                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4219                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4220                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4221         }
4222
4223         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4224                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4225                         all_slaves_active);
4226                 all_slaves_active = 0;
4227         }
4228
4229         if (resend_igmp < 0 || resend_igmp > 255) {
4230                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4231                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4232                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4233         }
4234
4235         bond_opt_initval(&newval, packets_per_slave);
4236         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4237                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4238                         packets_per_slave, USHRT_MAX);
4239                 packets_per_slave = 1;
4240         }
4241
4242         if (bond_mode == BOND_MODE_ALB) {
4243                 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",
4244                           updelay);
4245         }
4246
4247         if (!miimon) {
4248                 if (updelay || downdelay) {
4249                         /* just warn the user the up/down delay will have
4250                          * no effect since miimon is zero...
4251                          */
4252                         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",
4253                                 updelay, downdelay);
4254                 }
4255         } else {
4256                 /* don't allow arp monitoring */
4257                 if (arp_interval) {
4258                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4259                                 miimon, arp_interval);
4260                         arp_interval = 0;
4261                 }
4262
4263                 if ((updelay % miimon) != 0) {
4264                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4265                                 updelay, miimon, (updelay / miimon) * miimon);
4266                 }
4267
4268                 updelay /= miimon;
4269
4270                 if ((downdelay % miimon) != 0) {
4271                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4272                                 downdelay, miimon,
4273                                 (downdelay / miimon) * miimon);
4274                 }
4275
4276                 downdelay /= miimon;
4277         }
4278
4279         if (arp_interval < 0) {
4280                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4281                         arp_interval, INT_MAX);
4282                 arp_interval = 0;
4283         }
4284
4285         for (arp_ip_count = 0, i = 0;
4286              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4287                 __be32 ip;
4288
4289                 /* not a complete check, but good enough to catch mistakes */
4290                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4291                     !bond_is_ip_target_ok(ip)) {
4292                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4293                                 arp_ip_target[i]);
4294                         arp_interval = 0;
4295                 } else {
4296                         if (bond_get_targets_ip(arp_target, ip) == -1)
4297                                 arp_target[arp_ip_count++] = ip;
4298                         else
4299                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4300                                         &ip);
4301                 }
4302         }
4303
4304         if (arp_interval && !arp_ip_count) {
4305                 /* don't allow arping if no arp_ip_target given... */
4306                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4307                         arp_interval);
4308                 arp_interval = 0;
4309         }
4310
4311         if (arp_validate) {
4312                 if (!arp_interval) {
4313                         pr_err("arp_validate requires arp_interval\n");
4314                         return -EINVAL;
4315                 }
4316
4317                 bond_opt_initstr(&newval, arp_validate);
4318                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4319                                         &newval);
4320                 if (!valptr) {
4321                         pr_err("Error: invalid arp_validate \"%s\"\n",
4322                                arp_validate);
4323                         return -EINVAL;
4324                 }
4325                 arp_validate_value = valptr->value;
4326         } else {
4327                 arp_validate_value = 0;
4328         }
4329
4330         arp_all_targets_value = 0;
4331         if (arp_all_targets) {
4332                 bond_opt_initstr(&newval, arp_all_targets);
4333                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4334                                         &newval);
4335                 if (!valptr) {
4336                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4337                                arp_all_targets);
4338                         arp_all_targets_value = 0;
4339                 } else {
4340                         arp_all_targets_value = valptr->value;
4341                 }
4342         }
4343
4344         if (miimon) {
4345                 pr_info("MII link monitoring set to %d ms\n", miimon);
4346         } else if (arp_interval) {
4347                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4348                                           arp_validate_value);
4349                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4350                         arp_interval, valptr->string, arp_ip_count);
4351
4352                 for (i = 0; i < arp_ip_count; i++)
4353                         pr_cont(" %s", arp_ip_target[i]);
4354
4355                 pr_cont("\n");
4356
4357         } else if (max_bonds) {
4358                 /* miimon and arp_interval not set, we need one so things
4359                  * work as expected, see bonding.txt for details
4360                  */
4361                 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");
4362         }
4363
4364         if (primary && !bond_mode_uses_primary(bond_mode)) {
4365                 /* currently, using a primary only makes sense
4366                  * in active backup, TLB or ALB modes
4367                  */
4368                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4369                         primary, bond_mode_name(bond_mode));
4370                 primary = NULL;
4371         }
4372
4373         if (primary && primary_reselect) {
4374                 bond_opt_initstr(&newval, primary_reselect);
4375                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4376                                         &newval);
4377                 if (!valptr) {
4378                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4379                                primary_reselect);
4380                         return -EINVAL;
4381                 }
4382                 primary_reselect_value = valptr->value;
4383         } else {
4384                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4385         }
4386
4387         if (fail_over_mac) {
4388                 bond_opt_initstr(&newval, fail_over_mac);
4389                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4390                                         &newval);
4391                 if (!valptr) {
4392                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4393                                fail_over_mac);
4394                         return -EINVAL;
4395                 }
4396                 fail_over_mac_value = valptr->value;
4397                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4398                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4399         } else {
4400                 fail_over_mac_value = BOND_FOM_NONE;
4401         }
4402
4403         if (lp_interval == 0) {
4404                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4405                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4406                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4407         }
4408
4409         /* fill params struct with the proper values */
4410         params->mode = bond_mode;
4411         params->xmit_policy = xmit_hashtype;
4412         params->miimon = miimon;
4413         params->num_peer_notif = num_peer_notif;
4414         params->arp_interval = arp_interval;
4415         params->arp_validate = arp_validate_value;
4416         params->arp_all_targets = arp_all_targets_value;
4417         params->updelay = updelay;
4418         params->downdelay = downdelay;
4419         params->use_carrier = use_carrier;
4420         params->lacp_fast = lacp_fast;
4421         params->primary[0] = 0;
4422         params->primary_reselect = primary_reselect_value;
4423         params->fail_over_mac = fail_over_mac_value;
4424         params->tx_queues = tx_queues;
4425         params->all_slaves_active = all_slaves_active;
4426         params->resend_igmp = resend_igmp;
4427         params->min_links = min_links;
4428         params->lp_interval = lp_interval;
4429         params->packets_per_slave = packets_per_slave;
4430         params->tlb_dynamic_lb = 1; /* Default value */
4431         if (packets_per_slave > 0) {
4432                 params->reciprocal_packets_per_slave =
4433                         reciprocal_value(packets_per_slave);
4434         } else {
4435                 /* reciprocal_packets_per_slave is unused if
4436                  * packets_per_slave is 0 or 1, just initialize it
4437                  */
4438                 params->reciprocal_packets_per_slave =
4439                         (struct reciprocal_value) { 0 };
4440         }
4441
4442         if (primary) {
4443                 strncpy(params->primary, primary, IFNAMSIZ);
4444                 params->primary[IFNAMSIZ - 1] = 0;
4445         }
4446
4447         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4448
4449         return 0;
4450 }
4451
4452 static struct lock_class_key bonding_netdev_xmit_lock_key;
4453 static struct lock_class_key bonding_netdev_addr_lock_key;
4454 static struct lock_class_key bonding_tx_busylock_key;
4455
4456 static void bond_set_lockdep_class_one(struct net_device *dev,
4457                                        struct netdev_queue *txq,
4458                                        void *_unused)
4459 {
4460         lockdep_set_class(&txq->_xmit_lock,
4461                           &bonding_netdev_xmit_lock_key);
4462 }
4463
4464 static void bond_set_lockdep_class(struct net_device *dev)
4465 {
4466         lockdep_set_class(&dev->addr_list_lock,
4467                           &bonding_netdev_addr_lock_key);
4468         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4469         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4470 }
4471
4472 /* Called from registration process */
4473 static int bond_init(struct net_device *bond_dev)
4474 {
4475         struct bonding *bond = netdev_priv(bond_dev);
4476         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4477
4478         netdev_dbg(bond_dev, "Begin bond_init\n");
4479
4480         bond->wq = create_singlethread_workqueue(bond_dev->name);
4481         if (!bond->wq)
4482                 return -ENOMEM;
4483
4484         bond_set_lockdep_class(bond_dev);
4485
4486         list_add_tail(&bond->bond_list, &bn->dev_list);
4487
4488         bond_prepare_sysfs_group(bond);
4489
4490         bond_debug_register(bond);
4491
4492         /* Ensure valid dev_addr */
4493         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4494             bond_dev->addr_assign_type == NET_ADDR_PERM)
4495                 eth_hw_addr_random(bond_dev);
4496
4497         return 0;
4498 }
4499
4500 unsigned int bond_get_num_tx_queues(void)
4501 {
4502         return tx_queues;
4503 }
4504
4505 /* Create a new bond based on the specified name and bonding parameters.
4506  * If name is NULL, obtain a suitable "bond%d" name for us.
4507  * Caller must NOT hold rtnl_lock; we need to release it here before we
4508  * set up our sysfs entries.
4509  */
4510 int bond_create(struct net *net, const char *name)
4511 {
4512         struct net_device *bond_dev;
4513         int res;
4514
4515         rtnl_lock();
4516
4517         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4518                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
4519                                    bond_setup, tx_queues);
4520         if (!bond_dev) {
4521                 pr_err("%s: eek! can't alloc netdev!\n", name);
4522                 rtnl_unlock();
4523                 return -ENOMEM;
4524         }
4525
4526         dev_net_set(bond_dev, net);
4527         bond_dev->rtnl_link_ops = &bond_link_ops;
4528
4529         res = register_netdevice(bond_dev);
4530
4531         netif_carrier_off(bond_dev);
4532
4533         rtnl_unlock();
4534         if (res < 0)
4535                 bond_destructor(bond_dev);
4536         return res;
4537 }
4538
4539 static int __net_init bond_net_init(struct net *net)
4540 {
4541         struct bond_net *bn = net_generic(net, bond_net_id);
4542
4543         bn->net = net;
4544         INIT_LIST_HEAD(&bn->dev_list);
4545
4546         bond_create_proc_dir(bn);
4547         bond_create_sysfs(bn);
4548
4549         return 0;
4550 }
4551
4552 static void __net_exit bond_net_exit(struct net *net)
4553 {
4554         struct bond_net *bn = net_generic(net, bond_net_id);
4555         struct bonding *bond, *tmp_bond;
4556         LIST_HEAD(list);
4557
4558         bond_destroy_sysfs(bn);
4559
4560         /* Kill off any bonds created after unregistering bond rtnl ops */
4561         rtnl_lock();
4562         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4563                 unregister_netdevice_queue(bond->dev, &list);
4564         unregister_netdevice_many(&list);
4565         rtnl_unlock();
4566
4567         bond_destroy_proc_dir(bn);
4568 }
4569
4570 static struct pernet_operations bond_net_ops = {
4571         .init = bond_net_init,
4572         .exit = bond_net_exit,
4573         .id   = &bond_net_id,
4574         .size = sizeof(struct bond_net),
4575 };
4576
4577 static int __init bonding_init(void)
4578 {
4579         int i;
4580         int res;
4581
4582         pr_info("%s", bond_version);
4583
4584         res = bond_check_params(&bonding_defaults);
4585         if (res)
4586                 goto out;
4587
4588         res = register_pernet_subsys(&bond_net_ops);
4589         if (res)
4590                 goto out;
4591
4592         res = bond_netlink_init();
4593         if (res)
4594                 goto err_link;
4595
4596         bond_create_debugfs();
4597
4598         for (i = 0; i < max_bonds; i++) {
4599                 res = bond_create(&init_net, NULL);
4600                 if (res)
4601                         goto err;
4602         }
4603
4604         register_netdevice_notifier(&bond_netdev_notifier);
4605 out:
4606         return res;
4607 err:
4608         bond_destroy_debugfs();
4609         bond_netlink_fini();
4610 err_link:
4611         unregister_pernet_subsys(&bond_net_ops);
4612         goto out;
4613
4614 }
4615
4616 static void __exit bonding_exit(void)
4617 {
4618         unregister_netdevice_notifier(&bond_netdev_notifier);
4619
4620         bond_destroy_debugfs();
4621
4622         bond_netlink_fini();
4623         unregister_pernet_subsys(&bond_net_ops);
4624
4625 #ifdef CONFIG_NET_POLL_CONTROLLER
4626         /* Make sure we don't have an imbalance on our netpoll blocking */
4627         WARN_ON(atomic_read(&netpoll_block_tx));
4628 #endif
4629 }
4630
4631 module_init(bonding_init);
4632 module_exit(bonding_exit);
4633 MODULE_LICENSE("GPL");
4634 MODULE_VERSION(DRV_VERSION);
4635 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4636 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");