drivers/net/bonding/bond_main.c

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  35
  36 #include <linux/kernel.h>
  37 #include <linux/module.h>
  38 #include <linux/types.h>
  39 #include <linux/fcntl.h>
  40 #include <linux/interrupt.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/ioport.h>
  43 #include <linux/in.h>
  44 #include <net/ip.h>
  45 #include <linux/ip.h>
  46 #include <linux/tcp.h>
  47 #include <linux/udp.h>
  48 #include <linux/slab.h>
  49 #include <linux/string.h>
  50 #include <linux/init.h>
  51 #include <linux/timer.h>
  52 #include <linux/socket.h>
  53 #include <linux/ctype.h>
  54 #include <linux/inet.h>
  55 #include <linux/bitops.h>
  56 #include <linux/io.h>
  57 #include <asm/dma.h>
  58 #include <linux/uaccess.h>
  59 #include <linux/errno.h>
  60 #include <linux/netdevice.h>
  61 #include <linux/inetdevice.h>
  62 #include <linux/igmp.h>
  63 #include <linux/etherdevice.h>
  64 #include <linux/skbuff.h>
  65 #include <net/sock.h>
  66 #include <linux/rtnetlink.h>
  67 #include <linux/smp.h>
  68 #include <linux/if_ether.h>
  69 #include <net/arp.h>
  70 #include <linux/mii.h>
  71 #include <linux/ethtool.h>
  72 #include <linux/if_vlan.h>
  73 #include <linux/if_bonding.h>
  74 #include <linux/jiffies.h>
  75 #include <linux/preempt.h>
  76 #include <net/route.h>
  77 #include <net/net_namespace.h>
  78 #include <net/netns/generic.h>
  79 #include <net/pkt_sched.h>
  80 #include <linux/rculist.h>
  81 #include <net/flow_keys.h>
  82 #include "bonding.h"
  83 #include "bond_3ad.h"
  84 #include "bond_alb.h"
  85
  86 /*---------------------------- Module parameters ----------------------------*/
  87
  88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  89
  90 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  91 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  92 static int num_peer_notif = 1;
  93 static int miimon;
  94 static int updelay;
  95 static int downdelay;
  96 static int use_carrier  = 1;
  97 static char *mode;
  98 static char *primary;
  99 static char *primary_reselect;
 100 static char *lacp_rate;
 101 static int min_links;
 102 static char *ad_select;
 103 static char *xmit_hash_policy;
 104 static int arp_interval;
 105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 106 static char *arp_validate;
 107 static char *arp_all_targets;
 108 static char *fail_over_mac;
 109 static int all_slaves_active;
 110 static struct bond_params bonding_defaults;
 111 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 112 static int packets_per_slave = 1;
 113 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
 114
 115 module_param(max_bonds, int, 0);
 116 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 117 module_param(tx_queues, int, 0);
 118 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 119 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 120 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 121                                "failover event (alias of num_unsol_na)");
 122 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 123 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 124                                "failover event (alias of num_grat_arp)");
 125 module_param(miimon, int, 0);
 126 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 127 module_param(updelay, int, 0);
 128 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 129 module_param(downdelay, int, 0);
 130 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 131                             "in milliseconds");
 132 module_param(use_carrier, int, 0);
 133 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 134                               "0 for off, 1 for on (default)");
 135 module_param(mode, charp, 0);
 136 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 137                        "1 for active-backup, 2 for balance-xor, "
 138                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 139                        "6 for balance-alb");
 140 module_param(primary, charp, 0);
 141 MODULE_PARM_DESC(primary, "Primary network device to use");
 142 module_param(primary_reselect, charp, 0);
 143 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 144                                    "once it comes up; "
 145                                    "0 for always (default), "
 146                                    "1 for only if speed of primary is "
 147                                    "better, "
 148                                    "2 for only on active slave "
 149                                    "failure");
 150 module_param(lacp_rate, charp, 0);
 151 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 152                             "0 for slow, 1 for fast");
 153 module_param(ad_select, charp, 0);
 154 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 155                             "0 for stable (default), 1 for bandwidth, "
 156                             "2 for count");
 157 module_param(min_links, int, 0);
 158 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 159
 160 module_param(xmit_hash_policy, charp, 0);
 161 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 162                                    "0 for layer 2 (default), 1 for layer 3+4, "
 163                                    "2 for layer 2+3, 3 for encap layer 2+3, "
 164                                    "4 for encap layer 3+4");
 165 module_param(arp_interval, int, 0);
 166 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 167 module_param_array(arp_ip_target, charp, NULL, 0);
 168 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 169 module_param(arp_validate, charp, 0);
 170 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 171                                "0 for none (default), 1 for active, "
 172                                "2 for backup, 3 for all");
 173 module_param(arp_all_targets, charp, 0);
 174 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 175 module_param(fail_over_mac, charp, 0);
 176 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 177                                 "the same MAC; 0 for none (default), "
 178                                 "1 for active, 2 for follow");
 179 module_param(all_slaves_active, int, 0);
 180 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 181                                      "by setting active flag for all slaves; "
 182                                      "0 for never (default), 1 for always.");
 183 module_param(resend_igmp, int, 0);
 184 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 185                               "link failure");
 186 module_param(packets_per_slave, int, 0);
 187 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
 188                                     "mode; 0 for a random slave, 1 packet per "
 189                                     "slave (default), >1 packets per slave.");
 190 module_param(lp_interval, uint, 0);
 191 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
 192                               "the bonding driver sends learning packets to "
 193                               "each slaves peer switch. The default is 1.");
 194
 195 /*----------------------------- Global variables ----------------------------*/
 196
 197 #ifdef CONFIG_NET_POLL_CONTROLLER
 198 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 199 #endif
 200
 201 int bond_net_id __read_mostly;
 202
 203 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 204 static int arp_ip_count;
 205 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 206 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 207 static int lacp_fast;
 208
 209 /*-------------------------- Forward declarations ---------------------------*/
 210
 211 static int bond_init(struct net_device *bond_dev);
 212 static void bond_uninit(struct net_device *bond_dev);
 213
 214 /*---------------------------- General routines -----------------------------*/
 215
 216 const char *bond_mode_name(int mode)
 217 {
 218         static const char *names[] = {
 219                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 220                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 221                 [BOND_MODE_XOR] = "load balancing (xor)",
 222                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 223                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 224                 [BOND_MODE_TLB] = "transmit load balancing",
 225                 [BOND_MODE_ALB] = "adaptive load balancing",
 226         };
 227
 228         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 229                 return "unknown";
 230
 231         return names[mode];
 232 }
 233
 234 /*---------------------------------- VLAN -----------------------------------*/
 235
 236 /**
 237  * bond_dev_queue_xmit - Prepare skb for xmit.
 238  *
 239  * @bond: bond device that got this skb for tx.
 240  * @skb: hw accel VLAN tagged skb to transmit
 241  * @slave_dev: slave that is supposed to xmit this skbuff
 242  */
 243 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 244                         struct net_device *slave_dev)
 245 {
 246         skb->dev = slave_dev;
 247
 248         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 249                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 250         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 251
 252         if (unlikely(netpoll_tx_running(bond->dev)))
 253                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 254         else
 255                 dev_queue_xmit(skb);
 256 }
 257
 258 /*
 259  * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 260  * We don't protect the slave list iteration with a lock because:
 261  * a. This operation is performed in IOCTL context,
 262  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 263  * c. Holding a lock with BH disabled while directly calling a base driver
 264  *    entry point is generally a BAD idea.
 265  *
 266  * The design of synchronization/protection for this operation in the 8021q
 267  * module is good for one or more VLAN devices over a single physical device
 268  * and cannot be extended for a teaming solution like bonding, so there is a
 269  * potential race condition here where a net device from the vlan group might
 270  * be referenced (either by a base driver or the 8021q code) while it is being
 271  * removed from the system. However, it turns out we're not making matters
 272  * worse, and if it works for regular VLAN usage it will work here too.
 273 */
 274
 275 /**
 276  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 277  * @bond_dev: bonding net device that got called
 278  * @vid: vlan id being added
 279  */
 280 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 281                                 __be16 proto, u16 vid)
 282 {
 283         struct bonding *bond = netdev_priv(bond_dev);
 284         struct slave *slave, *rollback_slave;
 285         struct list_head *iter;
 286         int res;
 287
 288         bond_for_each_slave(bond, slave, iter) {
 289                 res = vlan_vid_add(slave->dev, proto, vid);
 290                 if (res)
 291                         goto unwind;
 292         }
 293
 294         return 0;
 295
 296 unwind:
 297         /* unwind to the slave that failed */
 298         bond_for_each_slave(bond, rollback_slave, iter) {
 299                 if (rollback_slave == slave)
 300                         break;
 301
 302                 vlan_vid_del(rollback_slave->dev, proto, vid);
 303         }
 304
 305         return res;
 306 }
 307
 308 /**
 309  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 310  * @bond_dev: bonding net device that got called
 311  * @vid: vlan id being removed
 312  */
 313 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 314                                  __be16 proto, u16 vid)
 315 {
 316         struct bonding *bond = netdev_priv(bond_dev);
 317         struct list_head *iter;
 318         struct slave *slave;
 319
 320         bond_for_each_slave(bond, slave, iter)
 321                 vlan_vid_del(slave->dev, proto, vid);
 322
 323         if (bond_is_lb(bond))
 324                 bond_alb_clear_vlan(bond, vid);
 325
 326         return 0;
 327 }
 328
 329 /*------------------------------- Link status -------------------------------*/
 330
 331 /*
 332  * Set the carrier state for the master according to the state of its
 333  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 334  * do special 802.3ad magic.
 335  *
 336  * Returns zero if carrier state does not change, nonzero if it does.
 337  */
 338 static int bond_set_carrier(struct bonding *bond)
 339 {
 340         struct list_head *iter;
 341         struct slave *slave;
 342
 343         if (!bond_has_slaves(bond))
 344                 goto down;
 345
 346         if (bond->params.mode == BOND_MODE_8023AD)
 347                 return bond_3ad_set_carrier(bond);
 348
 349         bond_for_each_slave(bond, slave, iter) {
 350                 if (slave->link == BOND_LINK_UP) {
 351                         if (!netif_carrier_ok(bond->dev)) {
 352                                 netif_carrier_on(bond->dev);
 353                                 return 1;
 354                         }
 355                         return 0;
 356                 }
 357         }
 358
 359 down:
 360         if (netif_carrier_ok(bond->dev)) {
 361                 netif_carrier_off(bond->dev);
 362                 return 1;
 363         }
 364         return 0;
 365 }
 366
 367 /*
 368  * Get link speed and duplex from the slave's base driver
 369  * using ethtool. If for some reason the call fails or the
 370  * values are invalid, set speed and duplex to -1,
 371  * and return.
 372  */
 373 static void bond_update_speed_duplex(struct slave *slave)
 374 {
 375         struct net_device *slave_dev = slave->dev;
 376         struct ethtool_cmd ecmd;
 377         u32 slave_speed;
 378         int res;
 379
 380         slave->speed = SPEED_UNKNOWN;
 381         slave->duplex = DUPLEX_UNKNOWN;
 382
 383         res = __ethtool_get_settings(slave_dev, &ecmd);
 384         if (res < 0)
 385                 return;
 386
 387         slave_speed = ethtool_cmd_speed(&ecmd);
 388         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 389                 return;
 390
 391         switch (ecmd.duplex) {
 392         case DUPLEX_FULL:
 393         case DUPLEX_HALF:
 394                 break;
 395         default:
 396                 return;
 397         }
 398
 399         slave->speed = slave_speed;
 400         slave->duplex = ecmd.duplex;
 401
 402         return;
 403 }
 404
 405 const char *bond_slave_link_status(s8 link)
 406 {
 407         switch (link) {
 408         case BOND_LINK_UP:
 409                 return "up";
 410         case BOND_LINK_FAIL:
 411                 return "going down";
 412         case BOND_LINK_DOWN:
 413                 return "down";
 414         case BOND_LINK_BACK:
 415                 return "going back";
 416         default:
 417                 return "unknown";
 418         }
 419 }
 420
 421 /*
 422  * if <dev> supports MII link status reporting, check its link status.
 423  *
 424  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 425  * depending upon the setting of the use_carrier parameter.
 426  *
 427  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 428  * can't tell and just pretend it is), or 0, meaning that the link is
 429  * down.
 430  *
 431  * If reporting is non-zero, instead of faking link up, return -1 if
 432  * both ETHTOOL and MII ioctls fail (meaning the device does not
 433  * support them).  If use_carrier is set, return whatever it says.
 434  * It'd be nice if there was a good way to tell if a driver supports
 435  * netif_carrier, but there really isn't.
 436  */
 437 static int bond_check_dev_link(struct bonding *bond,
 438                                struct net_device *slave_dev, int reporting)
 439 {
 440         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 441         int (*ioctl)(struct net_device *, struct ifreq *, int);
 442         struct ifreq ifr;
 443         struct mii_ioctl_data *mii;
 444
 445         if (!reporting && !netif_running(slave_dev))
 446                 return 0;
 447
 448         if (bond->params.use_carrier)
 449                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 450
 451         /* Try to get link status using Ethtool first. */
 452         if (slave_dev->ethtool_ops->get_link)
 453                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 454                         BMSR_LSTATUS : 0;
 455
 456         /* Ethtool can't be used, fallback to MII ioctls. */
 457         ioctl = slave_ops->ndo_do_ioctl;
 458         if (ioctl) {
 459                 /* TODO: set pointer to correct ioctl on a per team member */
 460                 /*       bases to make this more efficient. that is, once  */
 461                 /*       we determine the correct ioctl, we will always    */
 462                 /*       call it and not the others for that team          */
 463                 /*       member.                                           */
 464
 465                 /*
 466                  * We cannot assume that SIOCGMIIPHY will also read a
 467                  * register; not all network drivers (e.g., e100)
 468                  * support that.
 469                  */
 470
 471                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 472                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 473                 mii = if_mii(&ifr);
 474                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 475                         mii->reg_num = MII_BMSR;
 476                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 477                                 return mii->val_out & BMSR_LSTATUS;
 478                 }
 479         }
 480
 481         /*
 482          * If reporting, report that either there's no dev->do_ioctl,
 483          * or both SIOCGMIIREG and get_link failed (meaning that we
 484          * cannot report link status).  If not reporting, pretend
 485          * we're ok.
 486          */
 487         return reporting ? -1 : BMSR_LSTATUS;
 488 }
 489
 490 /*----------------------------- Multicast list ------------------------------*/
 491
 492 /*
 493  * Push the promiscuity flag down to appropriate slaves
 494  */
 495 static int bond_set_promiscuity(struct bonding *bond, int inc)
 496 {
 497         struct list_head *iter;
 498         int err = 0;
 499
 500         if (USES_PRIMARY(bond->params.mode)) {
 501                 /* write lock already acquired */
 502                 if (bond->curr_active_slave) {
 503                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 504                                                   inc);
 505                 }
 506         } else {
 507                 struct slave *slave;
 508
 509                 bond_for_each_slave(bond, slave, iter) {
 510                         err = dev_set_promiscuity(slave->dev, inc);
 511                         if (err)
 512                                 return err;
 513                 }
 514         }
 515         return err;
 516 }
 517
 518 /*
 519  * Push the allmulti flag down to all slaves
 520  */
 521 static int bond_set_allmulti(struct bonding *bond, int inc)
 522 {
 523         struct list_head *iter;
 524         int err = 0;
 525
 526         if (USES_PRIMARY(bond->params.mode)) {
 527                 /* write lock already acquired */
 528                 if (bond->curr_active_slave) {
 529                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 530                                                inc);
 531                 }
 532         } else {
 533                 struct slave *slave;
 534
 535                 bond_for_each_slave(bond, slave, iter) {
 536                         err = dev_set_allmulti(slave->dev, inc);
 537                         if (err)
 538                                 return err;
 539                 }
 540         }
 541         return err;
 542 }
 543
 544 /*
 545  * Retrieve the list of registered multicast addresses for the bonding
 546  * device and retransmit an IGMP JOIN request to the current active
 547  * slave.
 548  */
 549 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 550 {
 551         struct bonding *bond = container_of(work, struct bonding,
 552                                             mcast_work.work);
 553
 554         if (!rtnl_trylock()) {
 555                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 556                 return;
 557         }
 558         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 559
 560         if (bond->igmp_retrans > 1) {
 561                 bond->igmp_retrans--;
 562                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 563         }
 564         rtnl_unlock();
 565 }
 566
 567 /* Flush bond's hardware addresses from slave
 568  */
 569 static void bond_hw_addr_flush(struct net_device *bond_dev,
 570                                struct net_device *slave_dev)
 571 {
 572         struct bonding *bond = netdev_priv(bond_dev);
 573
 574         dev_uc_unsync(slave_dev, bond_dev);
 575         dev_mc_unsync(slave_dev, bond_dev);
 576
 577         if (bond->params.mode == BOND_MODE_8023AD) {
 578                 /* del lacpdu mc addr from mc list */
 579                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 580
 581                 dev_mc_del(slave_dev, lacpdu_multicast);
 582         }
 583 }
 584
 585 /*--------------------------- Active slave change ---------------------------*/
 586
 587 /* Update the hardware address list and promisc/allmulti for the new and
 588  * old active slaves (if any).  Modes that are !USES_PRIMARY keep all
 589  * slaves up date at all times; only the USES_PRIMARY modes need to call
 590  * this function to swap these settings during a failover.
 591  */
 592 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 593                               struct slave *old_active)
 594 {
 595         ASSERT_RTNL();
 596
 597         if (old_active) {
 598                 if (bond->dev->flags & IFF_PROMISC)
 599                         dev_set_promiscuity(old_active->dev, -1);
 600
 601                 if (bond->dev->flags & IFF_ALLMULTI)
 602                         dev_set_allmulti(old_active->dev, -1);
 603
 604                 bond_hw_addr_flush(bond->dev, old_active->dev);
 605         }
 606
 607         if (new_active) {
 608                 /* FIXME: Signal errors upstream. */
 609                 if (bond->dev->flags & IFF_PROMISC)
 610                         dev_set_promiscuity(new_active->dev, 1);
 611
 612                 if (bond->dev->flags & IFF_ALLMULTI)
 613                         dev_set_allmulti(new_active->dev, 1);
 614
 615                 netif_addr_lock_bh(bond->dev);
 616                 dev_uc_sync(new_active->dev, bond->dev);
 617                 dev_mc_sync(new_active->dev, bond->dev);
 618                 netif_addr_unlock_bh(bond->dev);
 619         }
 620 }
 621
 622 /**
 623  * bond_set_dev_addr - clone slave's address to bond
 624  * @bond_dev: bond net device
 625  * @slave_dev: slave net device
 626  *
 627  * Should be called with RTNL held.
 628  */
 629 static void bond_set_dev_addr(struct net_device *bond_dev,
 630                               struct net_device *slave_dev)
 631 {
 632         pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 633                  bond_dev, slave_dev, slave_dev->addr_len);
 634         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 635         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 636         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 637 }
 638
 639 /*
 640  * bond_do_fail_over_mac
 641  *
 642  * Perform special MAC address swapping for fail_over_mac settings
 643  *
 644  * Called with RTNL, curr_slave_lock for write_bh.
 645  */
 646 static void bond_do_fail_over_mac(struct bonding *bond,
 647                                   struct slave *new_active,
 648                                   struct slave *old_active)
 649         __releases(&bond->curr_slave_lock)
 650         __acquires(&bond->curr_slave_lock)
 651 {
 652         u8 tmp_mac[ETH_ALEN];
 653         struct sockaddr saddr;
 654         int rv;
 655
 656         switch (bond->params.fail_over_mac) {
 657         case BOND_FOM_ACTIVE:
 658                 if (new_active) {
 659                         write_unlock_bh(&bond->curr_slave_lock);
 660                         bond_set_dev_addr(bond->dev, new_active->dev);
 661                         write_lock_bh(&bond->curr_slave_lock);
 662                 }
 663                 break;
 664         case BOND_FOM_FOLLOW:
 665                 /*
 666                  * if new_active && old_active, swap them
 667                  * if just old_active, do nothing (going to no active slave)
 668                  * if just new_active, set new_active to bond's MAC
 669                  */
 670                 if (!new_active)
 671                         return;
 672
 673                 write_unlock_bh(&bond->curr_slave_lock);
 674
 675                 if (old_active) {
 676                         ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
 677                         ether_addr_copy(saddr.sa_data,
 678                                         old_active->dev->dev_addr);
 679                         saddr.sa_family = new_active->dev->type;
 680                 } else {
 681                         ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
 682                         saddr.sa_family = bond->dev->type;
 683                 }
 684
 685                 rv = dev_set_mac_address(new_active->dev, &saddr);
 686                 if (rv) {
 687                         pr_err("%s: Error %d setting MAC of slave %s\n",
 688                                bond->dev->name, -rv, new_active->dev->name);
 689                         goto out;
 690                 }
 691
 692                 if (!old_active)
 693                         goto out;
 694
 695                 ether_addr_copy(saddr.sa_data, tmp_mac);
 696                 saddr.sa_family = old_active->dev->type;
 697
 698                 rv = dev_set_mac_address(old_active->dev, &saddr);
 699                 if (rv)
 700                         pr_err("%s: Error %d setting MAC of slave %s\n",
 701                                bond->dev->name, -rv, new_active->dev->name);
 702 out:
 703                 write_lock_bh(&bond->curr_slave_lock);
 704                 break;
 705         default:
 706                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 707                        bond->dev->name, bond->params.fail_over_mac);
 708                 break;
 709         }
 710
 711 }
 712
 713 static bool bond_should_change_active(struct bonding *bond)
 714 {
 715         struct slave *prim = bond->primary_slave;
 716         struct slave *curr = bond->curr_active_slave;
 717
 718         if (!prim || !curr || curr->link != BOND_LINK_UP)
 719                 return true;
 720         if (bond->force_primary) {
 721                 bond->force_primary = false;
 722                 return true;
 723         }
 724         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 725             (prim->speed < curr->speed ||
 726              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 727                 return false;
 728         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 729                 return false;
 730         return true;
 731 }
 732
 733 /**
 734  * find_best_interface - select the best available slave to be the active one
 735  * @bond: our bonding struct
 736  */
 737 static struct slave *bond_find_best_slave(struct bonding *bond)
 738 {
 739         struct slave *slave, *bestslave = NULL;
 740         struct list_head *iter;
 741         int mintime = bond->params.updelay;
 742
 743         if (bond->primary_slave && bond->primary_slave->link == BOND_LINK_UP &&
 744             bond_should_change_active(bond))
 745                 return bond->primary_slave;
 746
 747         bond_for_each_slave(bond, slave, iter) {
 748                 if (slave->link == BOND_LINK_UP)
 749                         return slave;
 750                 if (slave->link == BOND_LINK_BACK && IS_UP(slave->dev) &&
 751                     slave->delay < mintime) {
 752                         mintime = slave->delay;
 753                         bestslave = slave;
 754                 }
 755         }
 756
 757         return bestslave;
 758 }
 759
 760 static bool bond_should_notify_peers(struct bonding *bond)
 761 {
 762         struct slave *slave;
 763
 764         rcu_read_lock();
 765         slave = rcu_dereference(bond->curr_active_slave);
 766         rcu_read_unlock();
 767
 768         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
 769                  bond->dev->name, slave ? slave->dev->name : "NULL");
 770
 771         if (!slave || !bond->send_peer_notif ||
 772             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 773                 return false;
 774
 775         return true;
 776 }
 777
 778 /**
 779  * change_active_interface - change the active slave into the specified one
 780  * @bond: our bonding struct
 781  * @new: the new slave to make the active one
 782  *
 783  * Set the new slave to the bond's settings and unset them on the old
 784  * curr_active_slave.
 785  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 786  *
 787  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 788  * because it is apparently the best available slave we have, even though its
 789  * updelay hasn't timed out yet.
 790  *
 791  * If new_active is not NULL, caller must hold curr_slave_lock for write_bh.
 792  */
 793 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 794 {
 795         struct slave *old_active = bond->curr_active_slave;
 796
 797         if (old_active == new_active)
 798                 return;
 799
 800         if (new_active) {
 801                 new_active->last_link_up = jiffies;
 802
 803                 if (new_active->link == BOND_LINK_BACK) {
 804                         if (USES_PRIMARY(bond->params.mode)) {
 805                                 pr_info("%s: making interface %s the new active one %d ms earlier\n",
 806                                         bond->dev->name, new_active->dev->name,
 807                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
 808                         }
 809
 810                         new_active->delay = 0;
 811                         new_active->link = BOND_LINK_UP;
 812
 813                         if (bond->params.mode == BOND_MODE_8023AD)
 814                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 815
 816                         if (bond_is_lb(bond))
 817                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 818                 } else {
 819                         if (USES_PRIMARY(bond->params.mode)) {
 820                                 pr_info("%s: making interface %s the new active one\n",
 821                                         bond->dev->name, new_active->dev->name);
 822                         }
 823                 }
 824         }
 825
 826         if (USES_PRIMARY(bond->params.mode))
 827                 bond_hw_addr_swap(bond, new_active, old_active);
 828
 829         if (bond_is_lb(bond)) {
 830                 bond_alb_handle_active_change(bond, new_active);
 831                 if (old_active)
 832                         bond_set_slave_inactive_flags(old_active,
 833                                                       BOND_SLAVE_NOTIFY_NOW);
 834                 if (new_active)
 835                         bond_set_slave_active_flags(new_active,
 836                                                     BOND_SLAVE_NOTIFY_NOW);
 837         } else {
 838                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 839         }
 840
 841         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
 842                 if (old_active)
 843                         bond_set_slave_inactive_flags(old_active,
 844                                                       BOND_SLAVE_NOTIFY_NOW);
 845
 846                 if (new_active) {
 847                         bool should_notify_peers = false;
 848
 849                         bond_set_slave_active_flags(new_active,
 850                                                     BOND_SLAVE_NOTIFY_NOW);
 851
 852                         if (bond->params.fail_over_mac)
 853                                 bond_do_fail_over_mac(bond, new_active,
 854                                                       old_active);
 855
 856                         if (netif_running(bond->dev)) {
 857                                 bond->send_peer_notif =
 858                                         bond->params.num_peer_notif;
 859                                 should_notify_peers =
 860                                         bond_should_notify_peers(bond);
 861                         }
 862
 863                         write_unlock_bh(&bond->curr_slave_lock);
 864
 865                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 866                         if (should_notify_peers)
 867                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 868                                                          bond->dev);
 869
 870                         write_lock_bh(&bond->curr_slave_lock);
 871                 }
 872         }
 873
 874         /* resend IGMP joins since active slave has changed or
 875          * all were sent on curr_active_slave.
 876          * resend only if bond is brought up with the affected
 877          * bonding modes and the retransmission is enabled */
 878         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 879             ((USES_PRIMARY(bond->params.mode) && new_active) ||
 880              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
 881                 bond->igmp_retrans = bond->params.resend_igmp;
 882                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 883         }
 884 }
 885
 886 /**
 887  * bond_select_active_slave - select a new active slave, if needed
 888  * @bond: our bonding struct
 889  *
 890  * This functions should be called when one of the following occurs:
 891  * - The old curr_active_slave has been released or lost its link.
 892  * - The primary_slave has got its link back.
 893  * - A slave has got its link back and there's no old curr_active_slave.
 894  *
 895  * Caller must hold curr_slave_lock for write_bh.
 896  */
 897 void bond_select_active_slave(struct bonding *bond)
 898 {
 899         struct slave *best_slave;
 900         int rv;
 901
 902         best_slave = bond_find_best_slave(bond);
 903         if (best_slave != bond->curr_active_slave) {
 904                 bond_change_active_slave(bond, best_slave);
 905                 rv = bond_set_carrier(bond);
 906                 if (!rv)
 907                         return;
 908
 909                 if (netif_carrier_ok(bond->dev)) {
 910                         pr_info("%s: first active interface up!\n",
 911                                 bond->dev->name);
 912                 } else {
 913                         pr_info("%s: now running without any active interface!\n",
 914                                 bond->dev->name);
 915                 }
 916         }
 917 }
 918
 919 #ifdef CONFIG_NET_POLL_CONTROLLER
 920 static inline int slave_enable_netpoll(struct slave *slave)
 921 {
 922         struct netpoll *np;
 923         int err = 0;
 924
 925         np = kzalloc(sizeof(*np), GFP_KERNEL);
 926         err = -ENOMEM;
 927         if (!np)
 928                 goto out;
 929
 930         err = __netpoll_setup(np, slave->dev);
 931         if (err) {
 932                 kfree(np);
 933                 goto out;
 934         }
 935         slave->np = np;
 936 out:
 937         return err;
 938 }
 939 static inline void slave_disable_netpoll(struct slave *slave)
 940 {
 941         struct netpoll *np = slave->np;
 942
 943         if (!np)
 944                 return;
 945
 946         slave->np = NULL;
 947         __netpoll_free_async(np);
 948 }
 949
 950 static void bond_poll_controller(struct net_device *bond_dev)
 951 {
 952 }
 953
 954 static void bond_netpoll_cleanup(struct net_device *bond_dev)
 955 {
 956         struct bonding *bond = netdev_priv(bond_dev);
 957         struct list_head *iter;
 958         struct slave *slave;
 959
 960         bond_for_each_slave(bond, slave, iter)
 961                 if (IS_UP(slave->dev))
 962                         slave_disable_netpoll(slave);
 963 }
 964
 965 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
 966 {
 967         struct bonding *bond = netdev_priv(dev);
 968         struct list_head *iter;
 969         struct slave *slave;
 970         int err = 0;
 971
 972         bond_for_each_slave(bond, slave, iter) {
 973                 err = slave_enable_netpoll(slave);
 974                 if (err) {
 975                         bond_netpoll_cleanup(dev);
 976                         break;
 977                 }
 978         }
 979         return err;
 980 }
 981 #else
 982 static inline int slave_enable_netpoll(struct slave *slave)
 983 {
 984         return 0;
 985 }
 986 static inline void slave_disable_netpoll(struct slave *slave)
 987 {
 988 }
 989 static void bond_netpoll_cleanup(struct net_device *bond_dev)
 990 {
 991 }
 992 #endif
 993
 994 /*---------------------------------- IOCTL ----------------------------------*/
 995
 996 static netdev_features_t bond_fix_features(struct net_device *dev,
 997                                            netdev_features_t features)
 998 {
 999         struct bonding *bond = netdev_priv(dev);
1000         struct list_head *iter;
1001         netdev_features_t mask;
1002         struct slave *slave;
1003
1004         if (!bond_has_slaves(bond)) {
1005                 /* Disable adding VLANs to empty bond. But why? --mq */
1006                 features |= NETIF_F_VLAN_CHALLENGED;
1007                 return features;
1008         }
1009
1010         mask = features;
1011         features &= ~NETIF_F_ONE_FOR_ALL;
1012         features |= NETIF_F_ALL_FOR_ALL;
1013
1014         bond_for_each_slave(bond, slave, iter) {
1015                 features = netdev_increment_features(features,
1016                                                      slave->dev->features,
1017                                                      mask);
1018         }
1019         features = netdev_add_tso_features(features, mask);
1020
1021         return features;
1022 }
1023
1024 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1025                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1026                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1027
1028 static void bond_compute_features(struct bonding *bond)
1029 {
1030         unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1031         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1032         struct net_device *bond_dev = bond->dev;
1033         struct list_head *iter;
1034         struct slave *slave;
1035         unsigned short max_hard_header_len = ETH_HLEN;
1036         unsigned int gso_max_size = GSO_MAX_SIZE;
1037         u16 gso_max_segs = GSO_MAX_SEGS;
1038
1039         if (!bond_has_slaves(bond))
1040                 goto done;
1041
1042         bond_for_each_slave(bond, slave, iter) {
1043                 vlan_features = netdev_increment_features(vlan_features,
1044                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1045
1046                 dst_release_flag &= slave->dev->priv_flags;
1047                 if (slave->dev->hard_header_len > max_hard_header_len)
1048                         max_hard_header_len = slave->dev->hard_header_len;
1049
1050                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1051                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1052         }
1053
1054 done:
1055         bond_dev->vlan_features = vlan_features;
1056         bond_dev->hard_header_len = max_hard_header_len;
1057         bond_dev->gso_max_segs = gso_max_segs;
1058         netif_set_gso_max_size(bond_dev, gso_max_size);
1059
1060         flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1061         bond_dev->priv_flags = flags | dst_release_flag;
1062
1063         netdev_change_features(bond_dev);
1064 }
1065
1066 static void bond_setup_by_slave(struct net_device *bond_dev,
1067                                 struct net_device *slave_dev)
1068 {
1069         bond_dev->header_ops        = slave_dev->header_ops;
1070
1071         bond_dev->type              = slave_dev->type;
1072         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1073         bond_dev->addr_len          = slave_dev->addr_len;
1074
1075         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1076                 slave_dev->addr_len);
1077 }
1078
1079 /* On bonding slaves other than the currently active slave, suppress
1080  * duplicates except for alb non-mcast/bcast.
1081  */
1082 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1083                                             struct slave *slave,
1084                                             struct bonding *bond)
1085 {
1086         if (bond_is_slave_inactive(slave)) {
1087                 if (bond->params.mode == BOND_MODE_ALB &&
1088                     skb->pkt_type != PACKET_BROADCAST &&
1089                     skb->pkt_type != PACKET_MULTICAST)
1090                         return false;
1091                 return true;
1092         }
1093         return false;
1094 }
1095
1096 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1097 {
1098         struct sk_buff *skb = *pskb;
1099         struct slave *slave;
1100         struct bonding *bond;
1101         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1102                           struct slave *);
1103         int ret = RX_HANDLER_ANOTHER;
1104
1105         skb = skb_share_check(skb, GFP_ATOMIC);
1106         if (unlikely(!skb))
1107                 return RX_HANDLER_CONSUMED;
1108
1109         *pskb = skb;
1110
1111         slave = bond_slave_get_rcu(skb->dev);
1112         bond = slave->bond;
1113
1114         recv_probe = ACCESS_ONCE(bond->recv_probe);
1115         if (recv_probe) {
1116                 ret = recv_probe(skb, bond, slave);
1117                 if (ret == RX_HANDLER_CONSUMED) {
1118                         consume_skb(skb);
1119                         return ret;
1120                 }
1121         }
1122
1123         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1124                 return RX_HANDLER_EXACT;
1125         }
1126
1127         skb->dev = bond->dev;
1128
1129         if (bond->params.mode == BOND_MODE_ALB &&
1130             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1131             skb->pkt_type == PACKET_HOST) {
1132
1133                 if (unlikely(skb_cow_head(skb,
1134                                           skb->data - skb_mac_header(skb)))) {
1135                         kfree_skb(skb);
1136                         return RX_HANDLER_CONSUMED;
1137                 }
1138                 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1139         }
1140
1141         return ret;
1142 }
1143
1144 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1145                                       struct net_device *slave_dev,
1146                                       struct slave *slave)
1147 {
1148         int err;
1149
1150         err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1151         if (err)
1152                 return err;
1153         slave_dev->flags |= IFF_SLAVE;
1154         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1155         return 0;
1156 }
1157
1158 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1159                                   struct net_device *slave_dev)
1160 {
1161         netdev_upper_dev_unlink(slave_dev, bond_dev);
1162         slave_dev->flags &= ~IFF_SLAVE;
1163         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1164 }
1165
1166 /* enslave device <slave> to bond device <master> */
1167 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1168 {
1169         struct bonding *bond = netdev_priv(bond_dev);
1170         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1171         struct slave *new_slave = NULL, *prev_slave;
1172         struct sockaddr addr;
1173         int link_reporting;
1174         int res = 0, i;
1175
1176         if (!bond->params.use_carrier &&
1177             slave_dev->ethtool_ops->get_link == NULL &&
1178             slave_ops->ndo_do_ioctl == NULL) {
1179                 pr_warn("%s: Warning: no link monitoring support for %s\n",
1180                         bond_dev->name, slave_dev->name);
1181         }
1182
1183         /* already enslaved */
1184         if (slave_dev->flags & IFF_SLAVE) {
1185                 pr_debug("Error: Device was already enslaved\n");
1186                 return -EBUSY;
1187         }
1188
1189         if (bond_dev == slave_dev) {
1190                 pr_err("%s: cannot enslave bond to itself.\n", bond_dev->name);
1191                 return -EPERM;
1192         }
1193
1194         /* vlan challenged mutual exclusion */
1195         /* no need to lock since we're protected by rtnl_lock */
1196         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1197                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1198                 if (vlan_uses_dev(bond_dev)) {
1199                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1200                                bond_dev->name, slave_dev->name, bond_dev->name);
1201                         return -EPERM;
1202                 } else {
1203                         pr_warn("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1204                                 bond_dev->name, slave_dev->name,
1205                                 slave_dev->name, bond_dev->name);
1206                 }
1207         } else {
1208                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1209         }
1210
1211         /*
1212          * Old ifenslave binaries are no longer supported.  These can
1213          * be identified with moderate accuracy by the state of the slave:
1214          * the current ifenslave will set the interface down prior to
1215          * enslaving it; the old ifenslave will not.
1216          */
1217         if ((slave_dev->flags & IFF_UP)) {
1218                 pr_err("%s is up - this may be due to an out of date ifenslave\n",
1219                        slave_dev->name);
1220                 res = -EPERM;
1221                 goto err_undo_flags;
1222         }
1223
1224         /* set bonding device ether type by slave - bonding netdevices are
1225          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1226          * there is a need to override some of the type dependent attribs/funcs.
1227          *
1228          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1229          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1230          */
1231         if (!bond_has_slaves(bond)) {
1232                 if (bond_dev->type != slave_dev->type) {
1233                         pr_debug("%s: change device type from %d to %d\n",
1234                                  bond_dev->name,
1235                                  bond_dev->type, slave_dev->type);
1236
1237                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1238                                                        bond_dev);
1239                         res = notifier_to_errno(res);
1240                         if (res) {
1241                                 pr_err("%s: refused to change device type\n",
1242                                        bond_dev->name);
1243                                 res = -EBUSY;
1244                                 goto err_undo_flags;
1245                         }
1246
1247                         /* Flush unicast and multicast addresses */
1248                         dev_uc_flush(bond_dev);
1249                         dev_mc_flush(bond_dev);
1250
1251                         if (slave_dev->type != ARPHRD_ETHER)
1252                                 bond_setup_by_slave(bond_dev, slave_dev);
1253                         else {
1254                                 ether_setup(bond_dev);
1255                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1256                         }
1257
1258                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1259                                                  bond_dev);
1260                 }
1261         } else if (bond_dev->type != slave_dev->type) {
1262                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1263                        slave_dev->name, slave_dev->type, bond_dev->type);
1264                 res = -EINVAL;
1265                 goto err_undo_flags;
1266         }
1267
1268         if (slave_ops->ndo_set_mac_address == NULL) {
1269                 if (!bond_has_slaves(bond)) {
1270                         pr_warn("%s: Warning: The first slave device specified does not support setting the MAC address\n",
1271                                 bond_dev->name);
1272                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1273                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1274                                 pr_warn("%s: Setting fail_over_mac to active for active-backup mode\n",
1275                                         bond_dev->name);
1276                         }
1277                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1278                         pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n",
1279                                bond_dev->name);
1280                         res = -EOPNOTSUPP;
1281                         goto err_undo_flags;
1282                 }
1283         }
1284
1285         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1286
1287         /* If this is the first slave, then we need to set the master's hardware
1288          * address to be the same as the slave's. */
1289         if (!bond_has_slaves(bond) &&
1290             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1291                 bond_set_dev_addr(bond->dev, slave_dev);
1292
1293         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1294         if (!new_slave) {
1295                 res = -ENOMEM;
1296                 goto err_undo_flags;
1297         }
1298         /*
1299          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1300          * is set via sysfs or module option if desired.
1301          */
1302         new_slave->queue_id = 0;
1303
1304         /* Save slave's original mtu and then set it to match the bond */
1305         new_slave->original_mtu = slave_dev->mtu;
1306         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1307         if (res) {
1308                 pr_debug("Error %d calling dev_set_mtu\n", res);
1309                 goto err_free;
1310         }
1311
1312         /*
1313          * Save slave's original ("permanent") mac address for modes
1314          * that need it, and for restoring it upon release, and then
1315          * set it to the master's address
1316          */
1317         ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1318
1319         if (!bond->params.fail_over_mac ||
1320             bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1321                 /*
1322                  * Set slave to master's mac address.  The application already
1323                  * set the master's mac address to that of the first slave
1324                  */
1325                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1326                 addr.sa_family = slave_dev->type;
1327                 res = dev_set_mac_address(slave_dev, &addr);
1328                 if (res) {
1329                         pr_debug("Error %d calling set_mac_address\n", res);
1330                         goto err_restore_mtu;
1331                 }
1332         }
1333
1334         /* open the slave since the application closed it */
1335         res = dev_open(slave_dev);
1336         if (res) {
1337                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1338                 goto err_restore_mac;
1339         }
1340
1341         new_slave->bond = bond;
1342         new_slave->dev = slave_dev;
1343         slave_dev->priv_flags |= IFF_BONDING;
1344
1345         if (bond_is_lb(bond)) {
1346                 /* bond_alb_init_slave() must be called before all other stages since
1347                  * it might fail and we do not want to have to undo everything
1348                  */
1349                 res = bond_alb_init_slave(bond, new_slave);
1350                 if (res)
1351                         goto err_close;
1352         }
1353
1354         /* If the mode USES_PRIMARY, then the following is handled by
1355          * bond_change_active_slave().
1356          */
1357         if (!USES_PRIMARY(bond->params.mode)) {
1358                 /* set promiscuity level to new slave */
1359                 if (bond_dev->flags & IFF_PROMISC) {
1360                         res = dev_set_promiscuity(slave_dev, 1);
1361                         if (res)
1362                                 goto err_close;
1363                 }
1364
1365                 /* set allmulti level to new slave */
1366                 if (bond_dev->flags & IFF_ALLMULTI) {
1367                         res = dev_set_allmulti(slave_dev, 1);
1368                         if (res)
1369                                 goto err_close;
1370                 }
1371
1372                 netif_addr_lock_bh(bond_dev);
1373
1374                 dev_mc_sync_multiple(slave_dev, bond_dev);
1375                 dev_uc_sync_multiple(slave_dev, bond_dev);
1376
1377                 netif_addr_unlock_bh(bond_dev);
1378         }
1379
1380         if (bond->params.mode == BOND_MODE_8023AD) {
1381                 /* add lacpdu mc addr to mc list */
1382                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1383
1384                 dev_mc_add(slave_dev, lacpdu_multicast);
1385         }
1386
1387         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1388         if (res) {
1389                 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1390                        bond_dev->name, slave_dev->name);
1391                 goto err_close;
1392         }
1393
1394         prev_slave = bond_last_slave(bond);
1395
1396         new_slave->delay = 0;
1397         new_slave->link_failure_count = 0;
1398
1399         bond_update_speed_duplex(new_slave);
1400
1401         new_slave->last_rx = jiffies -
1402                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1403         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1404                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1405
1406         if (bond->params.miimon && !bond->params.use_carrier) {
1407                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1408
1409                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1410                         /*
1411                          * miimon is set but a bonded network driver
1412                          * does not support ETHTOOL/MII and
1413                          * arp_interval is not set.  Note: if
1414                          * use_carrier is enabled, we will never go
1415                          * here (because netif_carrier is always
1416                          * supported); thus, we don't need to change
1417                          * the messages for netif_carrier.
1418                          */
1419                         pr_warn("%s: Warning: MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n",
1420                                 bond_dev->name, slave_dev->name);
1421                 } else if (link_reporting == -1) {
1422                         /* unable get link status using mii/ethtool */
1423                         pr_warn("%s: Warning: can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface\n",
1424                                 bond_dev->name, slave_dev->name);
1425                 }
1426         }
1427
1428         /* check for initial state */
1429         if (bond->params.miimon) {
1430                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1431                         if (bond->params.updelay) {
1432                                 new_slave->link = BOND_LINK_BACK;
1433                                 new_slave->delay = bond->params.updelay;
1434                         } else {
1435                                 new_slave->link = BOND_LINK_UP;
1436                         }
1437                 } else {
1438                         new_slave->link = BOND_LINK_DOWN;
1439                 }
1440         } else if (bond->params.arp_interval) {
1441                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1442                         BOND_LINK_UP : BOND_LINK_DOWN);
1443         } else {
1444                 new_slave->link = BOND_LINK_UP;
1445         }
1446
1447         if (new_slave->link != BOND_LINK_DOWN)
1448                 new_slave->last_link_up = jiffies;
1449         pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1450                  new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1451                  (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1452
1453         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1454                 /* if there is a primary slave, remember it */
1455                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1456                         bond->primary_slave = new_slave;
1457                         bond->force_primary = true;
1458                 }
1459         }
1460
1461         switch (bond->params.mode) {
1462         case BOND_MODE_ACTIVEBACKUP:
1463                 bond_set_slave_inactive_flags(new_slave,
1464                                               BOND_SLAVE_NOTIFY_NOW);
1465                 break;
1466         case BOND_MODE_8023AD:
1467                 /* in 802.3ad mode, the internal mechanism
1468                  * will activate the slaves in the selected
1469                  * aggregator
1470                  */
1471                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1472                 /* if this is the first slave */
1473                 if (!prev_slave) {
1474                         SLAVE_AD_INFO(new_slave).id = 1;
1475                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1476                          * can be called only after the mac address of the bond is set
1477                          */
1478                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1479                 } else {
1480                         SLAVE_AD_INFO(new_slave).id =
1481                                 SLAVE_AD_INFO(prev_slave).id + 1;
1482                 }
1483
1484                 bond_3ad_bind_slave(new_slave);
1485                 break;
1486         case BOND_MODE_TLB:
1487         case BOND_MODE_ALB:
1488                 bond_set_active_slave(new_slave);
1489                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1490                 break;
1491         default:
1492                 pr_debug("This slave is always active in trunk mode\n");
1493
1494                 /* always active in trunk mode */
1495                 bond_set_active_slave(new_slave);
1496
1497                 /* In trunking mode there is little meaning to curr_active_slave
1498                  * anyway (it holds no special properties of the bond device),
1499                  * so we can change it without calling change_active_interface()
1500                  */
1501                 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1502                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1503
1504                 break;
1505         } /* switch(bond_mode) */
1506
1507 #ifdef CONFIG_NET_POLL_CONTROLLER
1508         slave_dev->npinfo = bond->dev->npinfo;
1509         if (slave_dev->npinfo) {
1510                 if (slave_enable_netpoll(new_slave)) {
1511                         pr_info("Error, %s: master_dev is using netpoll, but new slave device does not support netpoll\n",
1512                                 bond_dev->name);
1513                         res = -EBUSY;
1514                         goto err_detach;
1515                 }
1516         }
1517 #endif
1518
1519         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1520                                          new_slave);
1521         if (res) {
1522                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1523                 goto err_detach;
1524         }
1525
1526         res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1527         if (res) {
1528                 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1529                 goto err_unregister;
1530         }
1531
1532         res = bond_sysfs_slave_add(new_slave);
1533         if (res) {
1534                 pr_debug("Error %d calling bond_sysfs_slave_add\n", res);
1535                 goto err_upper_unlink;
1536         }
1537
1538         bond->slave_cnt++;
1539         bond_compute_features(bond);
1540         bond_set_carrier(bond);
1541
1542         if (USES_PRIMARY(bond->params.mode)) {
1543                 block_netpoll_tx();
1544                 write_lock_bh(&bond->curr_slave_lock);
1545                 bond_select_active_slave(bond);
1546                 write_unlock_bh(&bond->curr_slave_lock);
1547                 unblock_netpoll_tx();
1548         }
1549
1550         pr_info("%s: Enslaving %s as %s interface with %s link\n",
1551                 bond_dev->name, slave_dev->name,
1552                 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1553                 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1554
1555         /* enslave is successful */
1556         return 0;
1557
1558 /* Undo stages on error */
1559 err_upper_unlink:
1560         bond_upper_dev_unlink(bond_dev, slave_dev);
1561
1562 err_unregister:
1563         netdev_rx_handler_unregister(slave_dev);
1564
1565 err_detach:
1566         if (!USES_PRIMARY(bond->params.mode))
1567                 bond_hw_addr_flush(bond_dev, slave_dev);
1568
1569         vlan_vids_del_by_dev(slave_dev, bond_dev);
1570         if (bond->primary_slave == new_slave)
1571                 bond->primary_slave = NULL;
1572         if (bond->curr_active_slave == new_slave) {
1573                 block_netpoll_tx();
1574                 write_lock_bh(&bond->curr_slave_lock);
1575                 bond_change_active_slave(bond, NULL);
1576                 bond_select_active_slave(bond);
1577                 write_unlock_bh(&bond->curr_slave_lock);
1578                 unblock_netpoll_tx();
1579         }
1580         slave_disable_netpoll(new_slave);
1581
1582 err_close:
1583         slave_dev->priv_flags &= ~IFF_BONDING;
1584         dev_close(slave_dev);
1585
1586 err_restore_mac:
1587         if (!bond->params.fail_over_mac ||
1588             bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1589                 /* XXX TODO - fom follow mode needs to change master's
1590                  * MAC if this slave's MAC is in use by the bond, or at
1591                  * least print a warning.
1592                  */
1593                 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1594                 addr.sa_family = slave_dev->type;
1595                 dev_set_mac_address(slave_dev, &addr);
1596         }
1597
1598 err_restore_mtu:
1599         dev_set_mtu(slave_dev, new_slave->original_mtu);
1600
1601 err_free:
1602         kfree(new_slave);
1603
1604 err_undo_flags:
1605         /* Enslave of first slave has failed and we need to fix master's mac */
1606         if (!bond_has_slaves(bond) &&
1607             ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
1608                 eth_hw_addr_random(bond_dev);
1609
1610         return res;
1611 }
1612
1613 /*
1614  * Try to release the slave device <slave> from the bond device <master>
1615  * It is legal to access curr_active_slave without a lock because all the function
1616  * is write-locked. If "all" is true it means that the function is being called
1617  * while destroying a bond interface and all slaves are being released.
1618  *
1619  * The rules for slave state should be:
1620  *   for Active/Backup:
1621  *     Active stays on all backups go down
1622  *   for Bonded connections:
1623  *     The first up interface should be left on and all others downed.
1624  */
1625 static int __bond_release_one(struct net_device *bond_dev,
1626                               struct net_device *slave_dev,
1627                               bool all)
1628 {
1629         struct bonding *bond = netdev_priv(bond_dev);
1630         struct slave *slave, *oldcurrent;
1631         struct sockaddr addr;
1632         int old_flags = bond_dev->flags;
1633         netdev_features_t old_features = bond_dev->features;
1634
1635         /* slave is not a slave or master is not master of this slave */
1636         if (!(slave_dev->flags & IFF_SLAVE) ||
1637             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1638                 pr_err("%s: Error: cannot release %s\n",
1639                        bond_dev->name, slave_dev->name);
1640                 return -EINVAL;
1641         }
1642
1643         block_netpoll_tx();
1644
1645         slave = bond_get_slave_by_dev(bond, slave_dev);
1646         if (!slave) {
1647                 /* not a slave of this bond */
1648                 pr_info("%s: %s not enslaved\n",
1649                         bond_dev->name, slave_dev->name);
1650                 unblock_netpoll_tx();
1651                 return -EINVAL;
1652         }
1653
1654         bond_sysfs_slave_del(slave);
1655
1656         bond_upper_dev_unlink(bond_dev, slave_dev);
1657         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1658          * for this slave anymore.
1659          */
1660         netdev_rx_handler_unregister(slave_dev);
1661         write_lock_bh(&bond->lock);
1662
1663         /* Inform AD package of unbinding of slave. */
1664         if (bond->params.mode == BOND_MODE_8023AD)
1665                 bond_3ad_unbind_slave(slave);
1666
1667         write_unlock_bh(&bond->lock);
1668
1669         pr_info("%s: Releasing %s interface %s\n",
1670                 bond_dev->name,
1671                 bond_is_active_slave(slave) ? "active" : "backup",
1672                 slave_dev->name);
1673
1674         oldcurrent = bond->curr_active_slave;
1675
1676         bond->current_arp_slave = NULL;
1677
1678         if (!all && (!bond->params.fail_over_mac ||
1679                      bond->params.mode != BOND_MODE_ACTIVEBACKUP)) {
1680                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1681                     bond_has_slaves(bond))
1682                         pr_warn("%s: Warning: the permanent HWaddr of %s - %pM - is still in use by %s - set the HWaddr of %s to a different address to avoid conflicts\n",
1683                                 bond_dev->name, slave_dev->name,
1684                                 slave->perm_hwaddr,
1685                                 bond_dev->name, slave_dev->name);
1686         }
1687
1688         if (bond->primary_slave == slave)
1689                 bond->primary_slave = NULL;
1690
1691         if (oldcurrent == slave) {
1692                 write_lock_bh(&bond->curr_slave_lock);
1693                 bond_change_active_slave(bond, NULL);
1694                 write_unlock_bh(&bond->curr_slave_lock);
1695         }
1696
1697         if (bond_is_lb(bond)) {
1698                 /* Must be called only after the slave has been
1699                  * detached from the list and the curr_active_slave
1700                  * has been cleared (if our_slave == old_current),
1701                  * but before a new active slave is selected.
1702                  */
1703                 bond_alb_deinit_slave(bond, slave);
1704         }
1705
1706         if (all) {
1707                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1708         } else if (oldcurrent == slave) {
1709                 /*
1710                  * Note that we hold RTNL over this sequence, so there
1711                  * is no concern that another slave add/remove event
1712                  * will interfere.
1713                  */
1714                 write_lock_bh(&bond->curr_slave_lock);
1715
1716                 bond_select_active_slave(bond);
1717
1718                 write_unlock_bh(&bond->curr_slave_lock);
1719         }
1720
1721         if (!bond_has_slaves(bond)) {
1722                 bond_set_carrier(bond);
1723                 eth_hw_addr_random(bond_dev);
1724
1725                 if (vlan_uses_dev(bond_dev)) {
1726                         pr_warn("%s: Warning: clearing HW address of %s while it still has VLANs\n",
1727                                 bond_dev->name, bond_dev->name);
1728                         pr_warn("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs\n",
1729                                 bond_dev->name);
1730                 }
1731         }
1732
1733         unblock_netpoll_tx();
1734         synchronize_rcu();
1735         bond->slave_cnt--;
1736
1737         if (!bond_has_slaves(bond)) {
1738                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1739                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1740         }
1741
1742         bond_compute_features(bond);
1743         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1744             (old_features & NETIF_F_VLAN_CHALLENGED))
1745                 pr_info("%s: last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1746                         bond_dev->name, slave_dev->name, bond_dev->name);
1747
1748         /* must do this from outside any spinlocks */
1749         vlan_vids_del_by_dev(slave_dev, bond_dev);
1750
1751         /* If the mode USES_PRIMARY, then this cases was handled above by
1752          * bond_change_active_slave(..., NULL)
1753          */
1754         if (!USES_PRIMARY(bond->params.mode)) {
1755                 /* unset promiscuity level from slave
1756                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1757                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1758                  * value of that flag before that change, as that was the value
1759                  * when this slave was attached, so we cache at the start of the
1760                  * function and use it here. Same goes for ALLMULTI below
1761                  */
1762                 if (old_flags & IFF_PROMISC)
1763                         dev_set_promiscuity(slave_dev, -1);
1764
1765                 /* unset allmulti level from slave */
1766                 if (old_flags & IFF_ALLMULTI)
1767                         dev_set_allmulti(slave_dev, -1);
1768
1769                 bond_hw_addr_flush(bond_dev, slave_dev);
1770         }
1771
1772         slave_disable_netpoll(slave);
1773
1774         /* close slave before restoring its mac address */
1775         dev_close(slave_dev);
1776
1777         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1778             bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1779                 /* restore original ("permanent") mac address */
1780                 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1781                 addr.sa_family = slave_dev->type;
1782                 dev_set_mac_address(slave_dev, &addr);
1783         }
1784
1785         dev_set_mtu(slave_dev, slave->original_mtu);
1786
1787         slave_dev->priv_flags &= ~IFF_BONDING;
1788
1789         kfree(slave);
1790
1791         return 0;  /* deletion OK */
1792 }
1793
1794 /* A wrapper used because of ndo_del_link */
1795 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1796 {
1797         return __bond_release_one(bond_dev, slave_dev, false);
1798 }
1799
1800 /*
1801 * First release a slave and then destroy the bond if no more slaves are left.
1802 * Must be under rtnl_lock when this function is called.
1803 */
1804 static int  bond_release_and_destroy(struct net_device *bond_dev,
1805                                      struct net_device *slave_dev)
1806 {
1807         struct bonding *bond = netdev_priv(bond_dev);
1808         int ret;
1809
1810         ret = bond_release(bond_dev, slave_dev);
1811         if (ret == 0 && !bond_has_slaves(bond)) {
1812                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1813                 pr_info("%s: Destroying bond %s\n",
1814                         bond_dev->name, bond_dev->name);
1815                 unregister_netdevice(bond_dev);
1816         }
1817         return ret;
1818 }
1819
1820 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1821 {
1822         struct bonding *bond = netdev_priv(bond_dev);
1823
1824         info->bond_mode = bond->params.mode;
1825         info->miimon = bond->params.miimon;
1826
1827         info->num_slaves = bond->slave_cnt;
1828
1829         return 0;
1830 }
1831
1832 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1833 {
1834         struct bonding *bond = netdev_priv(bond_dev);
1835         struct list_head *iter;
1836         int i = 0, res = -ENODEV;
1837         struct slave *slave;
1838
1839         bond_for_each_slave(bond, slave, iter) {
1840                 if (i++ == (int)info->slave_id) {
1841                         res = 0;
1842                         strcpy(info->slave_name, slave->dev->name);
1843                         info->link = slave->link;
1844                         info->state = bond_slave_state(slave);
1845                         info->link_failure_count = slave->link_failure_count;
1846                         break;
1847                 }
1848         }
1849
1850         return res;
1851 }
1852
1853 /*-------------------------------- Monitoring -------------------------------*/
1854
1855
1856 static int bond_miimon_inspect(struct bonding *bond)
1857 {
1858         int link_state, commit = 0;
1859         struct list_head *iter;
1860         struct slave *slave;
1861         bool ignore_updelay;
1862
1863         ignore_updelay = !bond->curr_active_slave ? true : false;
1864
1865         bond_for_each_slave_rcu(bond, slave, iter) {
1866                 slave->new_link = BOND_LINK_NOCHANGE;
1867
1868                 link_state = bond_check_dev_link(bond, slave->dev, 0);
1869
1870                 switch (slave->link) {
1871                 case BOND_LINK_UP:
1872                         if (link_state)
1873                                 continue;
1874
1875                         slave->link = BOND_LINK_FAIL;
1876                         slave->delay = bond->params.downdelay;
1877                         if (slave->delay) {
1878                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms\n",
1879                                         bond->dev->name,
1880                                         (bond->params.mode ==
1881                                          BOND_MODE_ACTIVEBACKUP) ?
1882                                         (bond_is_active_slave(slave) ?
1883                                          "active " : "backup ") : "",
1884                                         slave->dev->name,
1885                                         bond->params.downdelay * bond->params.miimon);
1886                         }
1887                         /*FALLTHRU*/
1888                 case BOND_LINK_FAIL:
1889                         if (link_state) {
1890                                 /*
1891                                  * recovered before downdelay expired
1892                                  */
1893                                 slave->link = BOND_LINK_UP;
1894                                 slave->last_link_up = jiffies;
1895                                 pr_info("%s: link status up again after %d ms for interface %s\n",
1896                                         bond->dev->name,
1897                                         (bond->params.downdelay - slave->delay) *
1898                                         bond->params.miimon,
1899                                         slave->dev->name);
1900                                 continue;
1901                         }
1902
1903                         if (slave->delay <= 0) {
1904                                 slave->new_link = BOND_LINK_DOWN;
1905                                 commit++;
1906                                 continue;
1907                         }
1908
1909                         slave->delay--;
1910                         break;
1911
1912                 case BOND_LINK_DOWN:
1913                         if (!link_state)
1914                                 continue;
1915
1916                         slave->link = BOND_LINK_BACK;
1917                         slave->delay = bond->params.updelay;
1918
1919                         if (slave->delay) {
1920                                 pr_info("%s: link status up for interface %s, enabling it in %d ms\n",
1921                                         bond->dev->name, slave->dev->name,
1922                                         ignore_updelay ? 0 :
1923                                         bond->params.updelay *
1924                                         bond->params.miimon);
1925                         }
1926                         /*FALLTHRU*/
1927                 case BOND_LINK_BACK:
1928                         if (!link_state) {
1929                                 slave->link = BOND_LINK_DOWN;
1930                                 pr_info("%s: link status down again after %d ms for interface %s\n",
1931                                         bond->dev->name,
1932                                         (bond->params.updelay - slave->delay) *
1933                                         bond->params.miimon,
1934                                         slave->dev->name);
1935
1936                                 continue;
1937                         }
1938
1939                         if (ignore_updelay)
1940                                 slave->delay = 0;
1941
1942                         if (slave->delay <= 0) {
1943                                 slave->new_link = BOND_LINK_UP;
1944                                 commit++;
1945                                 ignore_updelay = false;
1946                                 continue;
1947                         }
1948
1949                         slave->delay--;
1950                         break;
1951                 }
1952         }
1953
1954         return commit;
1955 }
1956
1957 static void bond_miimon_commit(struct bonding *bond)
1958 {
1959         struct list_head *iter;
1960         struct slave *slave;
1961
1962         bond_for_each_slave(bond, slave, iter) {
1963                 switch (slave->new_link) {
1964                 case BOND_LINK_NOCHANGE:
1965                         continue;
1966
1967                 case BOND_LINK_UP:
1968                         slave->link = BOND_LINK_UP;
1969                         slave->last_link_up = jiffies;
1970
1971                         if (bond->params.mode == BOND_MODE_8023AD) {
1972                                 /* prevent it from being the active one */
1973                                 bond_set_backup_slave(slave);
1974                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1975                                 /* make it immediately active */
1976                                 bond_set_active_slave(slave);
1977                         } else if (slave != bond->primary_slave) {
1978                                 /* prevent it from being the active one */
1979                                 bond_set_backup_slave(slave);
1980                         }
1981
1982                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex\n",
1983                                 bond->dev->name, slave->dev->name,
1984                                 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
1985                                 slave->duplex ? "full" : "half");
1986
1987                         /* notify ad that the link status has changed */
1988                         if (bond->params.mode == BOND_MODE_8023AD)
1989                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
1990
1991                         if (bond_is_lb(bond))
1992                                 bond_alb_handle_link_change(bond, slave,
1993                                                             BOND_LINK_UP);
1994
1995                         if (!bond->curr_active_slave ||
1996                             (slave == bond->primary_slave))
1997                                 goto do_failover;
1998
1999                         continue;
2000
2001                 case BOND_LINK_DOWN:
2002                         if (slave->link_failure_count < UINT_MAX)
2003                                 slave->link_failure_count++;
2004
2005                         slave->link = BOND_LINK_DOWN;
2006
2007                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2008                             bond->params.mode == BOND_MODE_8023AD)
2009                                 bond_set_slave_inactive_flags(slave,
2010                                                               BOND_SLAVE_NOTIFY_NOW);
2011
2012                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2013                                 bond->dev->name, slave->dev->name);
2014
2015                         if (bond->params.mode == BOND_MODE_8023AD)
2016                                 bond_3ad_handle_link_change(slave,
2017                                                             BOND_LINK_DOWN);
2018
2019                         if (bond_is_lb(bond))
2020                                 bond_alb_handle_link_change(bond, slave,
2021                                                             BOND_LINK_DOWN);
2022
2023                         if (slave == bond->curr_active_slave)
2024                                 goto do_failover;
2025
2026                         continue;
2027
2028                 default:
2029                         pr_err("%s: invalid new link %d on slave %s\n",
2030                                bond->dev->name, slave->new_link,
2031                                slave->dev->name);
2032                         slave->new_link = BOND_LINK_NOCHANGE;
2033
2034                         continue;
2035                 }
2036
2037 do_failover:
2038                 ASSERT_RTNL();
2039                 block_netpoll_tx();
2040                 write_lock_bh(&bond->curr_slave_lock);
2041                 bond_select_active_slave(bond);
2042                 write_unlock_bh(&bond->curr_slave_lock);
2043                 unblock_netpoll_tx();
2044         }
2045
2046         bond_set_carrier(bond);
2047 }
2048
2049 /*
2050  * bond_mii_monitor
2051  *
2052  * Really a wrapper that splits the mii monitor into two phases: an
2053  * inspection, then (if inspection indicates something needs to be done)
2054  * an acquisition of appropriate locks followed by a commit phase to
2055  * implement whatever link state changes are indicated.
2056  */
2057 static void bond_mii_monitor(struct work_struct *work)
2058 {
2059         struct bonding *bond = container_of(work, struct bonding,
2060                                             mii_work.work);
2061         bool should_notify_peers = false;
2062         unsigned long delay;
2063
2064         delay = msecs_to_jiffies(bond->params.miimon);
2065
2066         if (!bond_has_slaves(bond))
2067                 goto re_arm;
2068
2069         rcu_read_lock();
2070
2071         should_notify_peers = bond_should_notify_peers(bond);
2072
2073         if (bond_miimon_inspect(bond)) {
2074                 rcu_read_unlock();
2075
2076                 /* Race avoidance with bond_close cancel of workqueue */
2077                 if (!rtnl_trylock()) {
2078                         delay = 1;
2079                         should_notify_peers = false;
2080                         goto re_arm;
2081                 }
2082
2083                 bond_miimon_commit(bond);
2084
2085                 rtnl_unlock();  /* might sleep, hold no other locks */
2086         } else
2087                 rcu_read_unlock();
2088
2089 re_arm:
2090         if (bond->params.miimon)
2091                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2092
2093         if (should_notify_peers) {
2094                 if (!rtnl_trylock())
2095                         return;
2096                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2097                 rtnl_unlock();
2098         }
2099 }
2100
2101 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2102 {
2103         struct net_device *upper;
2104         struct list_head *iter;
2105         bool ret = false;
2106
2107         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2108                 return true;
2109
2110         rcu_read_lock();
2111         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2112                 if (ip == bond_confirm_addr(upper, 0, ip)) {
2113                         ret = true;
2114                         break;
2115                 }
2116         }
2117         rcu_read_unlock();
2118
2119         return ret;
2120 }
2121
2122 /*
2123  * We go to the (large) trouble of VLAN tagging ARP frames because
2124  * switches in VLAN mode (especially if ports are configured as
2125  * "native" to a VLAN) might not pass non-tagged frames.
2126  */
2127 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2128                           __be32 dest_ip, __be32 src_ip,
2129                           struct bond_vlan_tag *inner,
2130                           struct bond_vlan_tag *outer)
2131 {
2132         struct sk_buff *skb;
2133
2134         pr_debug("arp %d on slave %s: dst %pI4 src %pI4\n",
2135                  arp_op, slave_dev->name, &dest_ip, &src_ip);
2136
2137         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2138                          NULL, slave_dev->dev_addr, NULL);
2139
2140         if (!skb) {
2141                 net_err_ratelimited("ARP packet allocation failed\n");
2142                 return;
2143         }
2144         if (outer->vlan_id) {
2145                 if (inner->vlan_id) {
2146                         pr_debug("inner tag: proto %X vid %X\n",
2147                                  ntohs(inner->vlan_proto), inner->vlan_id);
2148                         skb = __vlan_put_tag(skb, inner->vlan_proto,
2149                                              inner->vlan_id);
2150                         if (!skb) {
2151                                 net_err_ratelimited("failed to insert inner VLAN tag\n");
2152                                 return;
2153                         }
2154                 }
2155
2156                 pr_debug("outer reg: proto %X vid %X\n",
2157                          ntohs(outer->vlan_proto), outer->vlan_id);
2158                 skb = vlan_put_tag(skb, outer->vlan_proto, outer->vlan_id);
2159                 if (!skb) {
2160                         net_err_ratelimited("failed to insert outer VLAN tag\n");
2161                         return;
2162                 }
2163         }
2164         arp_xmit(skb);
2165 }
2166
2167
2168 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2169 {
2170         struct net_device *upper, *vlan_upper;
2171         struct list_head *iter, *vlan_iter;
2172         struct rtable *rt;
2173         struct bond_vlan_tag inner, outer;
2174         __be32 *targets = bond->params.arp_targets, addr;
2175         int i;
2176
2177         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2178                 pr_debug("basa: target %pI4\n", &targets[i]);
2179                 inner.vlan_proto = 0;
2180                 inner.vlan_id = 0;
2181                 outer.vlan_proto = 0;
2182                 outer.vlan_id = 0;
2183
2184                 /* Find out through which dev should the packet go */
2185                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2186                                      RTO_ONLINK, 0);
2187                 if (IS_ERR(rt)) {
2188                         /* there's no route to target - try to send arp
2189                          * probe to generate any traffic (arp_validate=0)
2190                          */
2191                         if (bond->params.arp_validate)
2192                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2193                                                      bond->dev->name,
2194                                                      &targets[i]);
2195                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i], 0, &inner, &outer);
2196                         continue;
2197                 }
2198
2199                 /* bond device itself */
2200                 if (rt->dst.dev == bond->dev)
2201                         goto found;
2202
2203                 rcu_read_lock();
2204                 /* first we search only for vlan devices. for every vlan
2205                  * found we verify its upper dev list, searching for the
2206                  * rt->dst.dev. If found we save the tag of the vlan and
2207                  * proceed to send the packet.
2208                  */
2209                 netdev_for_each_all_upper_dev_rcu(bond->dev, vlan_upper,
2210                                                   vlan_iter) {
2211                         if (!is_vlan_dev(vlan_upper))
2212                                 continue;
2213
2214                         if (vlan_upper == rt->dst.dev) {
2215                                 outer.vlan_proto = vlan_dev_vlan_proto(vlan_upper);
2216                                 outer.vlan_id = vlan_dev_vlan_id(vlan_upper);
2217                                 rcu_read_unlock();
2218                                 goto found;
2219                         }
2220                         netdev_for_each_all_upper_dev_rcu(vlan_upper, upper,
2221                                                           iter) {
2222                                 if (upper == rt->dst.dev) {
2223                                         /* If the upper dev is a vlan dev too,
2224                                          *  set the vlan tag to inner tag.
2225                                          */
2226                                         if (is_vlan_dev(upper)) {
2227                                                 inner.vlan_proto = vlan_dev_vlan_proto(upper);
2228                                                 inner.vlan_id = vlan_dev_vlan_id(upper);
2229                                         }
2230                                         outer.vlan_proto = vlan_dev_vlan_proto(vlan_upper);
2231                                         outer.vlan_id = vlan_dev_vlan_id(vlan_upper);
2232                                         rcu_read_unlock();
2233                                         goto found;
2234                                 }
2235                         }
2236                 }
2237
2238                 /* if the device we're looking for is not on top of any of
2239                  * our upper vlans, then just search for any dev that
2240                  * matches, and in case it's a vlan - save the id
2241                  */
2242                 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2243                         if (upper == rt->dst.dev) {
2244                                 rcu_read_unlock();
2245                                 goto found;
2246                         }
2247                 }
2248                 rcu_read_unlock();
2249
2250                 /* Not our device - skip */
2251                 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2252                          bond->dev->name, &targets[i],
2253                          rt->dst.dev ? rt->dst.dev->name : "NULL");
2254
2255                 ip_rt_put(rt);
2256                 continue;
2257
2258 found:
2259                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2260                 ip_rt_put(rt);
2261                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2262                               addr, &inner, &outer);
2263         }
2264 }
2265
2266 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2267 {
2268         int i;
2269
2270         if (!sip || !bond_has_this_ip(bond, tip)) {
2271                 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2272                 return;
2273         }
2274
2275         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2276         if (i == -1) {
2277                 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2278                 return;
2279         }
2280         slave->last_rx = jiffies;
2281         slave->target_last_arp_rx[i] = jiffies;
2282 }
2283
2284 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2285                  struct slave *slave)
2286 {
2287         struct arphdr *arp = (struct arphdr *)skb->data;
2288         struct slave *curr_active_slave;
2289         unsigned char *arp_ptr;
2290         __be32 sip, tip;
2291         int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2292
2293         if (!slave_do_arp_validate(bond, slave)) {
2294                 if ((slave_do_arp_validate_only(bond, slave) && is_arp) ||
2295                     !slave_do_arp_validate_only(bond, slave))
2296                         slave->last_rx = jiffies;
2297                 return RX_HANDLER_ANOTHER;
2298         } else if (!is_arp) {
2299                 return RX_HANDLER_ANOTHER;
2300         }
2301
2302         alen = arp_hdr_len(bond->dev);
2303
2304         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2305                  bond->dev->name, skb->dev->name);
2306
2307         if (alen > skb_headlen(skb)) {
2308                 arp = kmalloc(alen, GFP_ATOMIC);
2309                 if (!arp)
2310                         goto out_unlock;
2311                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2312                         goto out_unlock;
2313         }
2314
2315         if (arp->ar_hln != bond->dev->addr_len ||
2316             skb->pkt_type == PACKET_OTHERHOST ||
2317             skb->pkt_type == PACKET_LOOPBACK ||
2318             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2319             arp->ar_pro != htons(ETH_P_IP) ||
2320             arp->ar_pln != 4)
2321                 goto out_unlock;
2322
2323         arp_ptr = (unsigned char *)(arp + 1);
2324         arp_ptr += bond->dev->addr_len;
2325         memcpy(&sip, arp_ptr, 4);
2326         arp_ptr += 4 + bond->dev->addr_len;
2327         memcpy(&tip, arp_ptr, 4);
2328
2329         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2330                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2331                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2332                  &sip, &tip);
2333
2334         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2335
2336         /*
2337          * Backup slaves won't see the ARP reply, but do come through
2338          * here for each ARP probe (so we swap the sip/tip to validate
2339          * the probe).  In a "redundant switch, common router" type of
2340          * configuration, the ARP probe will (hopefully) travel from
2341          * the active, through one switch, the router, then the other
2342          * switch before reaching the backup.
2343          *
2344          * We 'trust' the arp requests if there is an active slave and
2345          * it received valid arp reply(s) after it became active. This
2346          * is done to avoid endless looping when we can't reach the
2347          * arp_ip_target and fool ourselves with our own arp requests.
2348          */
2349
2350         if (bond_is_active_slave(slave))
2351                 bond_validate_arp(bond, slave, sip, tip);
2352         else if (curr_active_slave &&
2353                  time_after(slave_last_rx(bond, curr_active_slave),
2354                             curr_active_slave->last_link_up))
2355                 bond_validate_arp(bond, slave, tip, sip);
2356
2357 out_unlock:
2358         if (arp != (struct arphdr *)skb->data)
2359                 kfree(arp);
2360         return RX_HANDLER_ANOTHER;
2361 }
2362
2363 /* function to verify if we're in the arp_interval timeslice, returns true if
2364  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2365  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2366  */
2367 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2368                                   int mod)
2369 {
2370         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2371
2372         return time_in_range(jiffies,
2373                              last_act - delta_in_ticks,
2374                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2375 }
2376
2377 /*
2378  * this function is called regularly to monitor each slave's link
2379  * ensuring that traffic is being sent and received when arp monitoring
2380  * is used in load-balancing mode. if the adapter has been dormant, then an
2381  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2382  * arp monitoring in active backup mode.
2383  */
2384 static void bond_loadbalance_arp_mon(struct work_struct *work)
2385 {
2386         struct bonding *bond = container_of(work, struct bonding,
2387                                             arp_work.work);
2388         struct slave *slave, *oldcurrent;
2389         struct list_head *iter;
2390         int do_failover = 0, slave_state_changed = 0;
2391
2392         if (!bond_has_slaves(bond))
2393                 goto re_arm;
2394
2395         rcu_read_lock();
2396
2397         oldcurrent = ACCESS_ONCE(bond->curr_active_slave);
2398         /* see if any of the previous devices are up now (i.e. they have
2399          * xmt and rcv traffic). the curr_active_slave does not come into
2400          * the picture unless it is null. also, slave->last_link_up is not
2401          * needed here because we send an arp on each slave and give a slave
2402          * as long as it needs to get the tx/rx within the delta.
2403          * TODO: what about up/down delay in arp mode? it wasn't here before
2404          *       so it can wait
2405          */
2406         bond_for_each_slave_rcu(bond, slave, iter) {
2407                 unsigned long trans_start = dev_trans_start(slave->dev);
2408
2409                 if (slave->link != BOND_LINK_UP) {
2410                         if (bond_time_in_interval(bond, trans_start, 1) &&
2411                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2412
2413                                 slave->link  = BOND_LINK_UP;
2414                                 slave_state_changed = 1;
2415
2416                                 /* primary_slave has no meaning in round-robin
2417                                  * mode. the window of a slave being up and
2418                                  * curr_active_slave being null after enslaving
2419                                  * is closed.
2420                                  */
2421                                 if (!oldcurrent) {
2422                                         pr_info("%s: link status definitely up for interface %s\n",
2423                                                 bond->dev->name,
2424                                                 slave->dev->name);
2425                                         do_failover = 1;
2426                                 } else {
2427                                         pr_info("%s: interface %s is now up\n",
2428                                                 bond->dev->name,
2429                                                 slave->dev->name);
2430                                 }
2431                         }
2432                 } else {
2433                         /* slave->link == BOND_LINK_UP */
2434
2435                         /* not all switches will respond to an arp request
2436                          * when the source ip is 0, so don't take the link down
2437                          * if we don't know our ip yet
2438                          */
2439                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2440                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2441
2442                                 slave->link  = BOND_LINK_DOWN;
2443                                 slave_state_changed = 1;
2444
2445                                 if (slave->link_failure_count < UINT_MAX)
2446                                         slave->link_failure_count++;
2447
2448                                 pr_info("%s: interface %s is now down\n",
2449                                         bond->dev->name, slave->dev->name);
2450
2451                                 if (slave == oldcurrent)
2452                                         do_failover = 1;
2453                         }
2454                 }
2455
2456                 /* note: if switch is in round-robin mode, all links
2457                  * must tx arp to ensure all links rx an arp - otherwise
2458                  * links may oscillate or not come up at all; if switch is
2459                  * in something like xor mode, there is nothing we can
2460                  * do - all replies will be rx'ed on same link causing slaves
2461                  * to be unstable during low/no traffic periods
2462                  */
2463                 if (IS_UP(slave->dev))
2464                         bond_arp_send_all(bond, slave);
2465         }
2466
2467         rcu_read_unlock();
2468
2469         if (do_failover || slave_state_changed) {
2470                 if (!rtnl_trylock())
2471                         goto re_arm;
2472
2473                 if (slave_state_changed) {
2474                         bond_slave_state_change(bond);
2475                 } else if (do_failover) {
2476                         /* the bond_select_active_slave must hold RTNL
2477                          * and curr_slave_lock for write.
2478                          */
2479                         block_netpoll_tx();
2480                         write_lock_bh(&bond->curr_slave_lock);
2481
2482                         bond_select_active_slave(bond);
2483
2484                         write_unlock_bh(&bond->curr_slave_lock);
2485                         unblock_netpoll_tx();
2486                 }
2487                 rtnl_unlock();
2488         }
2489
2490 re_arm:
2491         if (bond->params.arp_interval)
2492                 queue_delayed_work(bond->wq, &bond->arp_work,
2493                                    msecs_to_jiffies(bond->params.arp_interval));
2494 }
2495
2496 /*
2497  * Called to inspect slaves for active-backup mode ARP monitor link state
2498  * changes.  Sets new_link in slaves to specify what action should take
2499  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2500  * to link states must be committed.
2501  *
2502  * Called with rcu_read_lock hold.
2503  */
2504 static int bond_ab_arp_inspect(struct bonding *bond)
2505 {
2506         unsigned long trans_start, last_rx;
2507         struct list_head *iter;
2508         struct slave *slave;
2509         int commit = 0;
2510
2511         bond_for_each_slave_rcu(bond, slave, iter) {
2512                 slave->new_link = BOND_LINK_NOCHANGE;
2513                 last_rx = slave_last_rx(bond, slave);
2514
2515                 if (slave->link != BOND_LINK_UP) {
2516                         if (bond_time_in_interval(bond, last_rx, 1)) {
2517                                 slave->new_link = BOND_LINK_UP;
2518                                 commit++;
2519                         }
2520                         continue;
2521                 }
2522
2523                 /*
2524                  * Give slaves 2*delta after being enslaved or made
2525                  * active.  This avoids bouncing, as the last receive
2526                  * times need a full ARP monitor cycle to be updated.
2527                  */
2528                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2529                         continue;
2530
2531                 /*
2532                  * Backup slave is down if:
2533                  * - No current_arp_slave AND
2534                  * - more than 3*delta since last receive AND
2535                  * - the bond has an IP address
2536                  *
2537                  * Note: a non-null current_arp_slave indicates
2538                  * the curr_active_slave went down and we are
2539                  * searching for a new one; under this condition
2540                  * we only take the curr_active_slave down - this
2541                  * gives each slave a chance to tx/rx traffic
2542                  * before being taken out
2543                  */
2544                 if (!bond_is_active_slave(slave) &&
2545                     !bond->current_arp_slave &&
2546                     !bond_time_in_interval(bond, last_rx, 3)) {
2547                         slave->new_link = BOND_LINK_DOWN;
2548                         commit++;
2549                 }
2550
2551                 /*
2552                  * Active slave is down if:
2553                  * - more than 2*delta since transmitting OR
2554                  * - (more than 2*delta since receive AND
2555                  *    the bond has an IP address)
2556                  */
2557                 trans_start = dev_trans_start(slave->dev);
2558                 if (bond_is_active_slave(slave) &&
2559                     (!bond_time_in_interval(bond, trans_start, 2) ||
2560                      !bond_time_in_interval(bond, last_rx, 2))) {
2561                         slave->new_link = BOND_LINK_DOWN;
2562                         commit++;
2563                 }
2564         }
2565
2566         return commit;
2567 }
2568
2569 /*
2570  * Called to commit link state changes noted by inspection step of
2571  * active-backup mode ARP monitor.
2572  *
2573  * Called with RTNL hold.
2574  */
2575 static void bond_ab_arp_commit(struct bonding *bond)
2576 {
2577         unsigned long trans_start;
2578         struct list_head *iter;
2579         struct slave *slave;
2580
2581         bond_for_each_slave(bond, slave, iter) {
2582                 switch (slave->new_link) {
2583                 case BOND_LINK_NOCHANGE:
2584                         continue;
2585
2586                 case BOND_LINK_UP:
2587                         trans_start = dev_trans_start(slave->dev);
2588                         if (bond->curr_active_slave != slave ||
2589                             (!bond->curr_active_slave &&
2590                              bond_time_in_interval(bond, trans_start, 1))) {
2591                                 slave->link = BOND_LINK_UP;
2592                                 if (bond->current_arp_slave) {
2593                                         bond_set_slave_inactive_flags(
2594                                                 bond->current_arp_slave,
2595                                                 BOND_SLAVE_NOTIFY_NOW);
2596                                         bond->current_arp_slave = NULL;
2597                                 }
2598
2599                                 pr_info("%s: link status definitely up for interface %s\n",
2600                                         bond->dev->name, slave->dev->name);
2601
2602                                 if (!bond->curr_active_slave ||
2603                                     (slave == bond->primary_slave))
2604                                         goto do_failover;
2605
2606                         }
2607
2608                         continue;
2609
2610                 case BOND_LINK_DOWN:
2611                         if (slave->link_failure_count < UINT_MAX)
2612                                 slave->link_failure_count++;
2613
2614                         slave->link = BOND_LINK_DOWN;
2615                         bond_set_slave_inactive_flags(slave,
2616                                                       BOND_SLAVE_NOTIFY_NOW);
2617
2618                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2619                                 bond->dev->name, slave->dev->name);
2620
2621                         if (slave == bond->curr_active_slave) {
2622                                 bond->current_arp_slave = NULL;
2623                                 goto do_failover;
2624                         }
2625
2626                         continue;
2627
2628                 default:
2629                         pr_err("%s: impossible: new_link %d on slave %s\n",
2630                                bond->dev->name, slave->new_link,
2631                                slave->dev->name);
2632                         continue;
2633                 }
2634
2635 do_failover:
2636                 ASSERT_RTNL();
2637                 block_netpoll_tx();
2638                 write_lock_bh(&bond->curr_slave_lock);
2639                 bond_select_active_slave(bond);
2640                 write_unlock_bh(&bond->curr_slave_lock);
2641                 unblock_netpoll_tx();
2642         }
2643
2644         bond_set_carrier(bond);
2645 }
2646
2647 /*
2648  * Send ARP probes for active-backup mode ARP monitor.
2649  *
2650  * Called with rcu_read_lock hold.
2651  */
2652 static bool bond_ab_arp_probe(struct bonding *bond)
2653 {
2654         struct slave *slave, *before = NULL, *new_slave = NULL,
2655                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2656                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2657         struct list_head *iter;
2658         bool found = false;
2659         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2660
2661         if (curr_arp_slave && curr_active_slave)
2662                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2663                         curr_arp_slave->dev->name,
2664                         curr_active_slave->dev->name);
2665
2666         if (curr_active_slave) {
2667                 bond_arp_send_all(bond, curr_active_slave);
2668                 return should_notify_rtnl;
2669         }
2670
2671         /* if we don't have a curr_active_slave, search for the next available
2672          * backup slave from the current_arp_slave and make it the candidate
2673          * for becoming the curr_active_slave
2674          */
2675
2676         if (!curr_arp_slave) {
2677                 curr_arp_slave = bond_first_slave_rcu(bond);
2678                 if (!curr_arp_slave)
2679                         return should_notify_rtnl;
2680         }
2681
2682         bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2683
2684         bond_for_each_slave_rcu(bond, slave, iter) {
2685                 if (!found && !before && IS_UP(slave->dev))
2686                         before = slave;
2687
2688                 if (found && !new_slave && IS_UP(slave->dev))
2689                         new_slave = slave;
2690                 /* if the link state is up at this point, we
2691                  * mark it down - this can happen if we have
2692                  * simultaneous link failures and
2693                  * reselect_active_interface doesn't make this
2694                  * one the current slave so it is still marked
2695                  * up when it is actually down
2696                  */
2697                 if (!IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
2698                         slave->link = BOND_LINK_DOWN;
2699                         if (slave->link_failure_count < UINT_MAX)
2700                                 slave->link_failure_count++;
2701
2702                         bond_set_slave_inactive_flags(slave,
2703                                                       BOND_SLAVE_NOTIFY_LATER);
2704
2705                         pr_info("%s: backup interface %s is now down\n",
2706                                 bond->dev->name, slave->dev->name);
2707                 }
2708                 if (slave == curr_arp_slave)
2709                         found = true;
2710         }
2711
2712         if (!new_slave && before)
2713                 new_slave = before;
2714
2715         if (!new_slave)
2716                 goto check_state;
2717
2718         new_slave->link = BOND_LINK_BACK;
2719         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2720         bond_arp_send_all(bond, new_slave);
2721         new_slave->last_link_up = jiffies;
2722         rcu_assign_pointer(bond->current_arp_slave, new_slave);
2723
2724 check_state:
2725         bond_for_each_slave_rcu(bond, slave, iter) {
2726                 if (slave->should_notify) {
2727                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2728                         break;
2729                 }
2730         }
2731         return should_notify_rtnl;
2732 }
2733
2734 static void bond_activebackup_arp_mon(struct work_struct *work)
2735 {
2736         struct bonding *bond = container_of(work, struct bonding,
2737                                             arp_work.work);
2738         bool should_notify_peers = false;
2739         bool should_notify_rtnl = false;
2740         int delta_in_ticks;
2741
2742         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2743
2744         if (!bond_has_slaves(bond))
2745                 goto re_arm;
2746
2747         rcu_read_lock();
2748
2749         should_notify_peers = bond_should_notify_peers(bond);
2750
2751         if (bond_ab_arp_inspect(bond)) {
2752                 rcu_read_unlock();
2753
2754                 /* Race avoidance with bond_close flush of workqueue */
2755                 if (!rtnl_trylock()) {
2756                         delta_in_ticks = 1;
2757                         should_notify_peers = false;
2758                         goto re_arm;
2759                 }
2760
2761                 bond_ab_arp_commit(bond);
2762
2763                 rtnl_unlock();
2764                 rcu_read_lock();
2765         }
2766
2767         should_notify_rtnl = bond_ab_arp_probe(bond);
2768         rcu_read_unlock();
2769
2770 re_arm:
2771         if (bond->params.arp_interval)
2772                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2773
2774         if (should_notify_peers || should_notify_rtnl) {
2775                 if (!rtnl_trylock())
2776                         return;
2777
2778                 if (should_notify_peers)
2779                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2780                                                  bond->dev);
2781                 if (should_notify_rtnl)
2782                         bond_slave_state_notify(bond);
2783
2784                 rtnl_unlock();
2785         }
2786 }
2787
2788 /*-------------------------- netdev event handling --------------------------*/
2789
2790 /*
2791  * Change device name
2792  */
2793 static int bond_event_changename(struct bonding *bond)
2794 {
2795         bond_remove_proc_entry(bond);
2796         bond_create_proc_entry(bond);
2797
2798         bond_debug_reregister(bond);
2799
2800         return NOTIFY_DONE;
2801 }
2802
2803 static int bond_master_netdev_event(unsigned long event,
2804                                     struct net_device *bond_dev)
2805 {
2806         struct bonding *event_bond = netdev_priv(bond_dev);
2807
2808         switch (event) {
2809         case NETDEV_CHANGENAME:
2810                 return bond_event_changename(event_bond);
2811         case NETDEV_UNREGISTER:
2812                 bond_remove_proc_entry(event_bond);
2813                 break;
2814         case NETDEV_REGISTER:
2815                 bond_create_proc_entry(event_bond);
2816                 break;
2817         case NETDEV_NOTIFY_PEERS:
2818                 if (event_bond->send_peer_notif)
2819                         event_bond->send_peer_notif--;
2820                 break;
2821         default:
2822                 break;
2823         }
2824
2825         return NOTIFY_DONE;
2826 }
2827
2828 static int bond_slave_netdev_event(unsigned long event,
2829                                    struct net_device *slave_dev)
2830 {
2831         struct slave *slave = bond_slave_get_rtnl(slave_dev);
2832         struct bonding *bond;
2833         struct net_device *bond_dev;
2834         u32 old_speed;
2835         u8 old_duplex;
2836
2837         /* A netdev event can be generated while enslaving a device
2838          * before netdev_rx_handler_register is called in which case
2839          * slave will be NULL
2840          */
2841         if (!slave)
2842                 return NOTIFY_DONE;
2843         bond_dev = slave->bond->dev;
2844         bond = slave->bond;
2845
2846         switch (event) {
2847         case NETDEV_UNREGISTER:
2848                 if (bond_dev->type != ARPHRD_ETHER)
2849                         bond_release_and_destroy(bond_dev, slave_dev);
2850                 else
2851                         bond_release(bond_dev, slave_dev);
2852                 break;
2853         case NETDEV_UP:
2854         case NETDEV_CHANGE:
2855                 old_speed = slave->speed;
2856                 old_duplex = slave->duplex;
2857
2858                 bond_update_speed_duplex(slave);
2859
2860                 if (bond->params.mode == BOND_MODE_8023AD) {
2861                         if (old_speed != slave->speed)
2862                                 bond_3ad_adapter_speed_changed(slave);
2863                         if (old_duplex != slave->duplex)
2864                                 bond_3ad_adapter_duplex_changed(slave);
2865                 }
2866                 break;
2867         case NETDEV_DOWN:
2868                 /*
2869                  * ... Or is it this?
2870                  */
2871                 break;
2872         case NETDEV_CHANGEMTU:
2873                 /*
2874                  * TODO: Should slaves be allowed to
2875                  * independently alter their MTU?  For
2876                  * an active-backup bond, slaves need
2877                  * not be the same type of device, so
2878                  * MTUs may vary.  For other modes,
2879                  * slaves arguably should have the
2880                  * same MTUs. To do this, we'd need to
2881                  * take over the slave's change_mtu
2882                  * function for the duration of their
2883                  * servitude.
2884                  */
2885                 break;
2886         case NETDEV_CHANGENAME:
2887                 /* we don't care if we don't have primary set */
2888                 if (!USES_PRIMARY(bond->params.mode) ||
2889                     !bond->params.primary[0])
2890                         break;
2891
2892                 if (slave == bond->primary_slave) {
2893                         /* slave's name changed - he's no longer primary */
2894                         bond->primary_slave = NULL;
2895                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2896                         /* we have a new primary slave */
2897                         bond->primary_slave = slave;
2898                 } else { /* we didn't change primary - exit */
2899                         break;
2900                 }
2901
2902                 pr_info("%s: Primary slave changed to %s, reselecting active slave\n",
2903                         bond->dev->name,
2904                         bond->primary_slave ? slave_dev->name : "none");
2905
2906                 block_netpoll_tx();
2907                 write_lock_bh(&bond->curr_slave_lock);
2908                 bond_select_active_slave(bond);
2909                 write_unlock_bh(&bond->curr_slave_lock);
2910                 unblock_netpoll_tx();
2911                 break;
2912         case NETDEV_FEAT_CHANGE:
2913                 bond_compute_features(bond);
2914                 break;
2915         case NETDEV_RESEND_IGMP:
2916                 /* Propagate to master device */
2917                 call_netdevice_notifiers(event, slave->bond->dev);
2918                 break;
2919         default:
2920                 break;
2921         }
2922
2923         return NOTIFY_DONE;
2924 }
2925
2926 /*
2927  * bond_netdev_event: handle netdev notifier chain events.
2928  *
2929  * This function receives events for the netdev chain.  The caller (an
2930  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2931  * locks for us to safely manipulate the slave devices (RTNL lock,
2932  * dev_probe_lock).
2933  */
2934 static int bond_netdev_event(struct notifier_block *this,
2935                              unsigned long event, void *ptr)
2936 {
2937         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2938
2939         pr_debug("event_dev: %s, event: %lx\n",
2940                  event_dev ? event_dev->name : "None", event);
2941
2942         if (!(event_dev->priv_flags & IFF_BONDING))
2943                 return NOTIFY_DONE;
2944
2945         if (event_dev->flags & IFF_MASTER) {
2946                 pr_debug("IFF_MASTER\n");
2947                 return bond_master_netdev_event(event, event_dev);
2948         }
2949
2950         if (event_dev->flags & IFF_SLAVE) {
2951                 pr_debug("IFF_SLAVE\n");
2952                 return bond_slave_netdev_event(event, event_dev);
2953         }
2954
2955         return NOTIFY_DONE;
2956 }
2957
2958 static struct notifier_block bond_netdev_notifier = {
2959         .notifier_call = bond_netdev_event,
2960 };
2961
2962 /*---------------------------- Hashing Policies -----------------------------*/
2963
2964 /* L2 hash helper */
2965 static inline u32 bond_eth_hash(struct sk_buff *skb)
2966 {
2967         struct ethhdr *data = (struct ethhdr *)skb->data;
2968
2969         if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
2970                 return data->h_dest[5] ^ data->h_source[5];
2971
2972         return 0;
2973 }
2974
2975 /* Extract the appropriate headers based on bond's xmit policy */
2976 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
2977                               struct flow_keys *fk)
2978 {
2979         const struct ipv6hdr *iph6;
2980         const struct iphdr *iph;
2981         int noff, proto = -1;
2982
2983         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
2984                 return skb_flow_dissect(skb, fk);
2985
2986         fk->ports = 0;
2987         noff = skb_network_offset(skb);
2988         if (skb->protocol == htons(ETH_P_IP)) {
2989                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
2990                         return false;
2991                 iph = ip_hdr(skb);
2992                 fk->src = iph->saddr;
2993                 fk->dst = iph->daddr;
2994                 noff += iph->ihl << 2;
2995                 if (!ip_is_fragment(iph))
2996                         proto = iph->protocol;
2997         } else if (skb->protocol == htons(ETH_P_IPV6)) {
2998                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
2999                         return false;
3000                 iph6 = ipv6_hdr(skb);
3001                 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
3002                 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
3003                 noff += sizeof(*iph6);
3004                 proto = iph6->nexthdr;
3005         } else {
3006                 return false;
3007         }
3008         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3009                 fk->ports = skb_flow_get_ports(skb, noff, proto);
3010
3011         return true;
3012 }
3013
3014 /**
3015  * bond_xmit_hash - generate a hash value based on the xmit policy
3016  * @bond: bonding device
3017  * @skb: buffer to use for headers
3018  * @count: modulo value
3019  *
3020  * This function will extract the necessary headers from the skb buffer and use
3021  * them to generate a hash based on the xmit_policy set in the bonding device
3022  * which will be reduced modulo count before returning.
3023  */
3024 int bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, int count)
3025 {
3026         struct flow_keys flow;
3027         u32 hash;
3028
3029         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3030             !bond_flow_dissect(bond, skb, &flow))
3031                 return bond_eth_hash(skb) % count;
3032
3033         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3034             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3035                 hash = bond_eth_hash(skb);
3036         else
3037                 hash = (__force u32)flow.ports;
3038         hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
3039         hash ^= (hash >> 16);
3040         hash ^= (hash >> 8);
3041
3042         return hash % count;
3043 }
3044
3045 /*-------------------------- Device entry points ----------------------------*/
3046
3047 static void bond_work_init_all(struct bonding *bond)
3048 {
3049         INIT_DELAYED_WORK(&bond->mcast_work,
3050                           bond_resend_igmp_join_requests_delayed);
3051         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3052         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3053         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3054                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3055         else
3056                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3057         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3058 }
3059
3060 static void bond_work_cancel_all(struct bonding *bond)
3061 {
3062         cancel_delayed_work_sync(&bond->mii_work);
3063         cancel_delayed_work_sync(&bond->arp_work);
3064         cancel_delayed_work_sync(&bond->alb_work);
3065         cancel_delayed_work_sync(&bond->ad_work);
3066         cancel_delayed_work_sync(&bond->mcast_work);
3067 }
3068
3069 static int bond_open(struct net_device *bond_dev)
3070 {
3071         struct bonding *bond = netdev_priv(bond_dev);
3072         struct list_head *iter;
3073         struct slave *slave;
3074
3075         /* reset slave->backup and slave->inactive */
3076         read_lock(&bond->lock);
3077         if (bond_has_slaves(bond)) {
3078                 read_lock(&bond->curr_slave_lock);
3079                 bond_for_each_slave(bond, slave, iter) {
3080                         if (USES_PRIMARY(bond->params.mode)
3081                                 && (slave != bond->curr_active_slave)) {
3082                                 bond_set_slave_inactive_flags(slave,
3083                                                               BOND_SLAVE_NOTIFY_NOW);
3084                         } else {
3085                                 bond_set_slave_active_flags(slave,
3086                                                             BOND_SLAVE_NOTIFY_NOW);
3087                         }
3088                 }
3089                 read_unlock(&bond->curr_slave_lock);
3090         }
3091         read_unlock(&bond->lock);
3092
3093         bond_work_init_all(bond);
3094
3095         if (bond_is_lb(bond)) {
3096                 /* bond_alb_initialize must be called before the timer
3097                  * is started.
3098                  */
3099                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3100                         return -ENOMEM;
3101                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3102         }
3103
3104         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3105                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3106
3107         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3108                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3109                 bond->recv_probe = bond_arp_rcv;
3110         }
3111
3112         if (bond->params.mode == BOND_MODE_8023AD) {
3113                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3114                 /* register to receive LACPDUs */
3115                 bond->recv_probe = bond_3ad_lacpdu_recv;
3116                 bond_3ad_initiate_agg_selection(bond, 1);
3117         }
3118
3119         return 0;
3120 }
3121
3122 static int bond_close(struct net_device *bond_dev)
3123 {
3124         struct bonding *bond = netdev_priv(bond_dev);
3125
3126         bond_work_cancel_all(bond);
3127         bond->send_peer_notif = 0;
3128         if (bond_is_lb(bond))
3129                 bond_alb_deinitialize(bond);
3130         bond->recv_probe = NULL;
3131
3132         return 0;
3133 }
3134
3135 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3136                                                 struct rtnl_link_stats64 *stats)
3137 {
3138         struct bonding *bond = netdev_priv(bond_dev);
3139         struct rtnl_link_stats64 temp;
3140         struct list_head *iter;
3141         struct slave *slave;
3142
3143         memset(stats, 0, sizeof(*stats));
3144
3145         read_lock_bh(&bond->lock);
3146         bond_for_each_slave(bond, slave, iter) {
3147                 const struct rtnl_link_stats64 *sstats =
3148                         dev_get_stats(slave->dev, &temp);
3149
3150                 stats->rx_packets += sstats->rx_packets;
3151                 stats->rx_bytes += sstats->rx_bytes;
3152                 stats->rx_errors += sstats->rx_errors;
3153                 stats->rx_dropped += sstats->rx_dropped;
3154
3155                 stats->tx_packets += sstats->tx_packets;
3156                 stats->tx_bytes += sstats->tx_bytes;
3157                 stats->tx_errors += sstats->tx_errors;
3158                 stats->tx_dropped += sstats->tx_dropped;
3159
3160                 stats->multicast += sstats->multicast;
3161                 stats->collisions += sstats->collisions;
3162
3163                 stats->rx_length_errors += sstats->rx_length_errors;
3164                 stats->rx_over_errors += sstats->rx_over_errors;
3165                 stats->rx_crc_errors += sstats->rx_crc_errors;
3166                 stats->rx_frame_errors += sstats->rx_frame_errors;
3167                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3168                 stats->rx_missed_errors += sstats->rx_missed_errors;
3169
3170                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3171                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3172                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3173                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3174                 stats->tx_window_errors += sstats->tx_window_errors;
3175         }
3176         read_unlock_bh(&bond->lock);
3177
3178         return stats;
3179 }
3180
3181 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3182 {
3183         struct bonding *bond = netdev_priv(bond_dev);
3184         struct net_device *slave_dev = NULL;
3185         struct ifbond k_binfo;
3186         struct ifbond __user *u_binfo = NULL;
3187         struct ifslave k_sinfo;
3188         struct ifslave __user *u_sinfo = NULL;
3189         struct mii_ioctl_data *mii = NULL;
3190         struct bond_opt_value newval;
3191         struct net *net;
3192         int res = 0;
3193
3194         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3195
3196         switch (cmd) {
3197         case SIOCGMIIPHY:
3198                 mii = if_mii(ifr);
3199                 if (!mii)
3200                         return -EINVAL;
3201
3202                 mii->phy_id = 0;
3203                 /* Fall Through */
3204         case SIOCGMIIREG:
3205                 /*
3206                  * We do this again just in case we were called by SIOCGMIIREG
3207                  * instead of SIOCGMIIPHY.
3208                  */
3209                 mii = if_mii(ifr);
3210                 if (!mii)
3211                         return -EINVAL;
3212
3213
3214                 if (mii->reg_num == 1) {
3215                         mii->val_out = 0;
3216                         read_lock(&bond->lock);
3217                         read_lock(&bond->curr_slave_lock);
3218                         if (netif_carrier_ok(bond->dev))
3219                                 mii->val_out = BMSR_LSTATUS;
3220
3221                         read_unlock(&bond->curr_slave_lock);
3222                         read_unlock(&bond->lock);
3223                 }
3224
3225                 return 0;
3226         case BOND_INFO_QUERY_OLD:
3227         case SIOCBONDINFOQUERY:
3228                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3229
3230                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3231                         return -EFAULT;
3232
3233                 res = bond_info_query(bond_dev, &k_binfo);
3234                 if (res == 0 &&
3235                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3236                         return -EFAULT;
3237
3238                 return res;
3239         case BOND_SLAVE_INFO_QUERY_OLD:
3240         case SIOCBONDSLAVEINFOQUERY:
3241                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3242
3243                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3244                         return -EFAULT;
3245
3246                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3247                 if (res == 0 &&
3248                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3249                         return -EFAULT;
3250
3251                 return res;
3252         default:
3253                 /* Go on */
3254                 break;
3255         }
3256
3257         net = dev_net(bond_dev);
3258
3259         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3260                 return -EPERM;
3261
3262         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3263
3264         pr_debug("slave_dev=%p:\n", slave_dev);
3265
3266         if (!slave_dev)
3267                 return -ENODEV;
3268
3269         pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3270         switch (cmd) {
3271         case BOND_ENSLAVE_OLD:
3272         case SIOCBONDENSLAVE:
3273                 res = bond_enslave(bond_dev, slave_dev);
3274                 break;
3275         case BOND_RELEASE_OLD:
3276         case SIOCBONDRELEASE:
3277                 res = bond_release(bond_dev, slave_dev);
3278                 break;
3279         case BOND_SETHWADDR_OLD:
3280         case SIOCBONDSETHWADDR:
3281                 bond_set_dev_addr(bond_dev, slave_dev);
3282                 res = 0;
3283                 break;
3284         case BOND_CHANGE_ACTIVE_OLD:
3285         case SIOCBONDCHANGEACTIVE:
3286                 bond_opt_initstr(&newval, slave_dev->name);
3287                 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3288                 break;
3289         default:
3290                 res = -EOPNOTSUPP;
3291         }
3292
3293         return res;
3294 }
3295
3296 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3297 {
3298         struct bonding *bond = netdev_priv(bond_dev);
3299
3300         if (change & IFF_PROMISC)
3301                 bond_set_promiscuity(bond,
3302                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3303
3304         if (change & IFF_ALLMULTI)
3305                 bond_set_allmulti(bond,
3306                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3307 }
3308
3309 static void bond_set_rx_mode(struct net_device *bond_dev)
3310 {
3311         struct bonding *bond = netdev_priv(bond_dev);
3312         struct list_head *iter;
3313         struct slave *slave;
3314
3315
3316         rcu_read_lock();
3317         if (USES_PRIMARY(bond->params.mode)) {
3318                 slave = rcu_dereference(bond->curr_active_slave);
3319                 if (slave) {
3320                         dev_uc_sync(slave->dev, bond_dev);
3321                         dev_mc_sync(slave->dev, bond_dev);
3322                 }
3323         } else {
3324                 bond_for_each_slave_rcu(bond, slave, iter) {
3325                         dev_uc_sync_multiple(slave->dev, bond_dev);
3326                         dev_mc_sync_multiple(slave->dev, bond_dev);
3327                 }
3328         }
3329         rcu_read_unlock();
3330 }
3331
3332 static int bond_neigh_init(struct neighbour *n)
3333 {
3334         struct bonding *bond = netdev_priv(n->dev);
3335         const struct net_device_ops *slave_ops;
3336         struct neigh_parms parms;
3337         struct slave *slave;
3338         int ret;
3339
3340         slave = bond_first_slave(bond);
3341         if (!slave)
3342                 return 0;
3343         slave_ops = slave->dev->netdev_ops;
3344         if (!slave_ops->ndo_neigh_setup)
3345                 return 0;
3346
3347         parms.neigh_setup = NULL;
3348         parms.neigh_cleanup = NULL;
3349         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3350         if (ret)
3351                 return ret;
3352
3353         /*
3354          * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3355          * after the last slave has been detached.  Assumes that all slaves
3356          * utilize the same neigh_cleanup (true at this writing as only user
3357          * is ipoib).
3358          */
3359         n->parms->neigh_cleanup = parms.neigh_cleanup;
3360
3361         if (!parms.neigh_setup)
3362                 return 0;
3363
3364         return parms.neigh_setup(n);
3365 }
3366
3367 /*
3368  * The bonding ndo_neigh_setup is called at init time beofre any
3369  * slave exists. So we must declare proxy setup function which will
3370  * be used at run time to resolve the actual slave neigh param setup.
3371  *
3372  * It's also called by master devices (such as vlans) to setup their
3373  * underlying devices. In that case - do nothing, we're already set up from
3374  * our init.
3375  */
3376 static int bond_neigh_setup(struct net_device *dev,
3377                             struct neigh_parms *parms)
3378 {
3379         /* modify only our neigh_parms */
3380         if (parms->dev == dev)
3381                 parms->neigh_setup = bond_neigh_init;
3382
3383         return 0;
3384 }
3385
3386 /*
3387  * Change the MTU of all of a master's slaves to match the master
3388  */
3389 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3390 {
3391         struct bonding *bond = netdev_priv(bond_dev);
3392         struct slave *slave, *rollback_slave;
3393         struct list_head *iter;
3394         int res = 0;
3395
3396         pr_debug("bond=%p, name=%s, new_mtu=%d\n",
3397                  bond, bond_dev ? bond_dev->name : "None", new_mtu);
3398
3399         /* Can't hold bond->lock with bh disabled here since
3400          * some base drivers panic. On the other hand we can't
3401          * hold bond->lock without bh disabled because we'll
3402          * deadlock. The only solution is to rely on the fact
3403          * that we're under rtnl_lock here, and the slaves
3404          * list won't change. This doesn't solve the problem
3405          * of setting the slave's MTU while it is
3406          * transmitting, but the assumption is that the base
3407          * driver can handle that.
3408          *
3409          * TODO: figure out a way to safely iterate the slaves
3410          * list, but without holding a lock around the actual
3411          * call to the base driver.
3412          */
3413
3414         bond_for_each_slave(bond, slave, iter) {
3415                 pr_debug("s %p c_m %p\n",
3416                          slave, slave->dev->netdev_ops->ndo_change_mtu);
3417
3418                 res = dev_set_mtu(slave->dev, new_mtu);
3419
3420                 if (res) {
3421                         /* If we failed to set the slave's mtu to the new value
3422                          * we must abort the operation even in ACTIVE_BACKUP
3423                          * mode, because if we allow the backup slaves to have
3424                          * different mtu values than the active slave we'll
3425                          * need to change their mtu when doing a failover. That
3426                          * means changing their mtu from timer context, which
3427                          * is probably not a good idea.
3428                          */
3429                         pr_debug("err %d %s\n", res, slave->dev->name);
3430                         goto unwind;
3431                 }
3432         }
3433
3434         bond_dev->mtu = new_mtu;
3435
3436         return 0;
3437
3438 unwind:
3439         /* unwind from head to the slave that failed */
3440         bond_for_each_slave(bond, rollback_slave, iter) {
3441                 int tmp_res;
3442
3443                 if (rollback_slave == slave)
3444                         break;
3445
3446                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3447                 if (tmp_res) {
3448                         pr_debug("unwind err %d dev %s\n",
3449                                  tmp_res, rollback_slave->dev->name);
3450                 }
3451         }
3452
3453         return res;
3454 }
3455
3456 /*
3457  * Change HW address
3458  *
3459  * Note that many devices must be down to change the HW address, and
3460  * downing the master releases all slaves.  We can make bonds full of
3461  * bonding devices to test this, however.
3462  */
3463 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3464 {
3465         struct bonding *bond = netdev_priv(bond_dev);
3466         struct slave *slave, *rollback_slave;
3467         struct sockaddr *sa = addr, tmp_sa;
3468         struct list_head *iter;
3469         int res = 0;
3470
3471         if (bond->params.mode == BOND_MODE_ALB)
3472                 return bond_alb_set_mac_address(bond_dev, addr);
3473
3474
3475         pr_debug("bond=%p, name=%s\n",
3476                  bond, bond_dev ? bond_dev->name : "None");
3477
3478         /* If fail_over_mac is enabled, do nothing and return success.
3479          * Returning an error causes ifenslave to fail.
3480          */
3481         if (bond->params.fail_over_mac &&
3482             bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3483                 return 0;
3484
3485         if (!is_valid_ether_addr(sa->sa_data))
3486                 return -EADDRNOTAVAIL;
3487
3488         /* Can't hold bond->lock with bh disabled here since
3489          * some base drivers panic. On the other hand we can't
3490          * hold bond->lock without bh disabled because we'll
3491          * deadlock. The only solution is to rely on the fact
3492          * that we're under rtnl_lock here, and the slaves
3493          * list won't change. This doesn't solve the problem
3494          * of setting the slave's hw address while it is
3495          * transmitting, but the assumption is that the base
3496          * driver can handle that.
3497          *
3498          * TODO: figure out a way to safely iterate the slaves
3499          * list, but without holding a lock around the actual
3500          * call to the base driver.
3501          */
3502
3503         bond_for_each_slave(bond, slave, iter) {
3504                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3505                 res = dev_set_mac_address(slave->dev, addr);
3506                 if (res) {
3507                         /* TODO: consider downing the slave
3508                          * and retry ?
3509                          * User should expect communications
3510                          * breakage anyway until ARP finish
3511                          * updating, so...
3512                          */
3513                         pr_debug("err %d %s\n", res, slave->dev->name);
3514                         goto unwind;
3515                 }
3516         }
3517
3518         /* success */
3519         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3520         return 0;
3521
3522 unwind:
3523         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3524         tmp_sa.sa_family = bond_dev->type;
3525
3526         /* unwind from head to the slave that failed */
3527         bond_for_each_slave(bond, rollback_slave, iter) {
3528                 int tmp_res;
3529
3530                 if (rollback_slave == slave)
3531                         break;
3532
3533                 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3534                 if (tmp_res) {
3535                         pr_debug("unwind err %d dev %s\n",
3536                                  tmp_res, rollback_slave->dev->name);
3537                 }
3538         }
3539
3540         return res;
3541 }
3542
3543 /**
3544  * bond_xmit_slave_id - transmit skb through slave with slave_id
3545  * @bond: bonding device that is transmitting
3546  * @skb: buffer to transmit
3547  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3548  *
3549  * This function tries to transmit through slave with slave_id but in case
3550  * it fails, it tries to find the first available slave for transmission.
3551  * The skb is consumed in all cases, thus the function is void.
3552  */
3553 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3554 {
3555         struct list_head *iter;
3556         struct slave *slave;
3557         int i = slave_id;
3558
3559         /* Here we start from the slave with slave_id */
3560         bond_for_each_slave_rcu(bond, slave, iter) {
3561                 if (--i < 0) {
3562                         if (slave_can_tx(slave)) {
3563                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3564                                 return;
3565                         }
3566                 }
3567         }
3568
3569         /* Here we start from the first slave up to slave_id */
3570         i = slave_id;
3571         bond_for_each_slave_rcu(bond, slave, iter) {
3572                 if (--i < 0)
3573                         break;
3574                 if (slave_can_tx(slave)) {
3575                         bond_dev_queue_xmit(bond, skb, slave->dev);
3576                         return;
3577                 }
3578         }
3579         /* no slave that can tx has been found */
3580         dev_kfree_skb_any(skb);
3581 }
3582
3583 /**
3584  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3585  * @bond: bonding device to use
3586  *
3587  * Based on the value of the bonding device's packets_per_slave parameter
3588  * this function generates a slave id, which is usually used as the next
3589  * slave to transmit through.
3590  */
3591 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3592 {
3593         u32 slave_id;
3594         struct reciprocal_value reciprocal_packets_per_slave;
3595         int packets_per_slave = bond->params.packets_per_slave;
3596
3597         switch (packets_per_slave) {
3598         case 0:
3599                 slave_id = prandom_u32();
3600                 break;
3601         case 1:
3602                 slave_id = bond->rr_tx_counter;
3603                 break;
3604         default:
3605                 reciprocal_packets_per_slave =
3606                         bond->params.reciprocal_packets_per_slave;
3607                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3608                                              reciprocal_packets_per_slave);
3609                 break;
3610         }
3611         bond->rr_tx_counter++;
3612
3613         return slave_id;
3614 }
3615
3616 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3617 {
3618         struct bonding *bond = netdev_priv(bond_dev);
3619         struct iphdr *iph = ip_hdr(skb);
3620         struct slave *slave;
3621         u32 slave_id;
3622
3623         /* Start with the curr_active_slave that joined the bond as the
3624          * default for sending IGMP traffic.  For failover purposes one
3625          * needs to maintain some consistency for the interface that will
3626          * send the join/membership reports.  The curr_active_slave found
3627          * will send all of this type of traffic.
3628          */
3629         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3630                 slave = rcu_dereference(bond->curr_active_slave);
3631                 if (slave && slave_can_tx(slave))
3632                         bond_dev_queue_xmit(bond, skb, slave->dev);
3633                 else
3634                         bond_xmit_slave_id(bond, skb, 0);
3635         } else {
3636                 slave_id = bond_rr_gen_slave_id(bond);
3637                 bond_xmit_slave_id(bond, skb, slave_id % bond->slave_cnt);
3638         }
3639
3640         return NETDEV_TX_OK;
3641 }
3642
3643 /*
3644  * in active-backup mode, we know that bond->curr_active_slave is always valid if
3645  * the bond has a usable interface.
3646  */
3647 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3648 {
3649         struct bonding *bond = netdev_priv(bond_dev);
3650         struct slave *slave;
3651
3652         slave = rcu_dereference(bond->curr_active_slave);
3653         if (slave)
3654                 bond_dev_queue_xmit(bond, skb, slave->dev);
3655         else
3656                 dev_kfree_skb_any(skb);
3657
3658         return NETDEV_TX_OK;
3659 }
3660
3661 /* In bond_xmit_xor() , we determine the output device by using a pre-
3662  * determined xmit_hash_policy(), If the selected device is not enabled,
3663  * find the next active slave.
3664  */
3665 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3666 {
3667         struct bonding *bond = netdev_priv(bond_dev);
3668
3669         bond_xmit_slave_id(bond, skb, bond_xmit_hash(bond, skb, bond->slave_cnt));
3670
3671         return NETDEV_TX_OK;
3672 }
3673
3674 /* in broadcast mode, we send everything to all usable interfaces. */
3675 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3676 {
3677         struct bonding *bond = netdev_priv(bond_dev);
3678         struct slave *slave = NULL;
3679         struct list_head *iter;
3680
3681         bond_for_each_slave_rcu(bond, slave, iter) {
3682                 if (bond_is_last_slave(bond, slave))
3683                         break;
3684                 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3685                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3686
3687                         if (!skb2) {
3688                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3689                                                     bond_dev->name, __func__);
3690                                 continue;
3691                         }
3692                         /* bond_dev_queue_xmit always returns 0 */
3693                         bond_dev_queue_xmit(bond, skb2, slave->dev);
3694                 }
3695         }
3696         if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3697                 bond_dev_queue_xmit(bond, skb, slave->dev);
3698         else
3699                 dev_kfree_skb_any(skb);
3700
3701         return NETDEV_TX_OK;
3702 }
3703
3704 /*------------------------- Device initialization ---------------------------*/
3705
3706 /*
3707  * Lookup the slave that corresponds to a qid
3708  */
3709 static inline int bond_slave_override(struct bonding *bond,
3710                                       struct sk_buff *skb)
3711 {
3712         struct slave *slave = NULL;
3713         struct list_head *iter;
3714
3715         if (!skb->queue_mapping)
3716                 return 1;
3717
3718         /* Find out if any slaves have the same mapping as this skb. */
3719         bond_for_each_slave_rcu(bond, slave, iter) {
3720                 if (slave->queue_id == skb->queue_mapping) {
3721                         if (slave_can_tx(slave)) {
3722                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3723                                 return 0;
3724                         }
3725                         /* If the slave isn't UP, use default transmit policy. */
3726                         break;
3727                 }
3728         }
3729
3730         return 1;
3731 }
3732
3733
3734 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3735                              void *accel_priv, select_queue_fallback_t fallback)
3736 {
3737         /*
3738          * This helper function exists to help dev_pick_tx get the correct
3739          * destination queue.  Using a helper function skips a call to
3740          * skb_tx_hash and will put the skbs in the queue we expect on their
3741          * way down to the bonding driver.
3742          */
3743         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3744
3745         /*
3746          * Save the original txq to restore before passing to the driver
3747          */
3748         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3749
3750         if (unlikely(txq >= dev->real_num_tx_queues)) {
3751                 do {
3752                         txq -= dev->real_num_tx_queues;
3753                 } while (txq >= dev->real_num_tx_queues);
3754         }
3755         return txq;
3756 }
3757
3758 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3759 {
3760         struct bonding *bond = netdev_priv(dev);
3761
3762         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
3763                 if (!bond_slave_override(bond, skb))
3764                         return NETDEV_TX_OK;
3765         }
3766
3767         switch (bond->params.mode) {
3768         case BOND_MODE_ROUNDROBIN:
3769                 return bond_xmit_roundrobin(skb, dev);
3770         case BOND_MODE_ACTIVEBACKUP:
3771                 return bond_xmit_activebackup(skb, dev);
3772         case BOND_MODE_XOR:
3773                 return bond_xmit_xor(skb, dev);
3774         case BOND_MODE_BROADCAST:
3775                 return bond_xmit_broadcast(skb, dev);
3776         case BOND_MODE_8023AD:
3777                 return bond_3ad_xmit_xor(skb, dev);
3778         case BOND_MODE_ALB:
3779         case BOND_MODE_TLB:
3780                 return bond_alb_xmit(skb, dev);
3781         default:
3782                 /* Should never happen, mode already checked */
3783                 pr_err("%s: Error: Unknown bonding mode %d\n",
3784                        dev->name, bond->params.mode);
3785                 WARN_ON_ONCE(1);
3786                 dev_kfree_skb_any(skb);
3787                 return NETDEV_TX_OK;
3788         }
3789 }
3790
3791 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3792 {
3793         struct bonding *bond = netdev_priv(dev);
3794         netdev_tx_t ret = NETDEV_TX_OK;
3795
3796         /*
3797          * If we risk deadlock from transmitting this in the
3798          * netpoll path, tell netpoll to queue the frame for later tx
3799          */
3800         if (unlikely(is_netpoll_tx_blocked(dev)))
3801                 return NETDEV_TX_BUSY;
3802
3803         rcu_read_lock();
3804         if (bond_has_slaves(bond))
3805                 ret = __bond_start_xmit(skb, dev);
3806         else
3807                 dev_kfree_skb_any(skb);
3808         rcu_read_unlock();
3809
3810         return ret;
3811 }
3812
3813 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3814                                      struct ethtool_cmd *ecmd)
3815 {
3816         struct bonding *bond = netdev_priv(bond_dev);
3817         unsigned long speed = 0;
3818         struct list_head *iter;
3819         struct slave *slave;
3820
3821         ecmd->duplex = DUPLEX_UNKNOWN;
3822         ecmd->port = PORT_OTHER;
3823
3824         /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3825          * do not need to check mode.  Though link speed might not represent
3826          * the true receive or transmit bandwidth (not all modes are symmetric)
3827          * this is an accurate maximum.
3828          */
3829         read_lock(&bond->lock);
3830         bond_for_each_slave(bond, slave, iter) {
3831                 if (SLAVE_IS_OK(slave)) {
3832                         if (slave->speed != SPEED_UNKNOWN)
3833                                 speed += slave->speed;
3834                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
3835                             slave->duplex != DUPLEX_UNKNOWN)
3836                                 ecmd->duplex = slave->duplex;
3837                 }
3838         }
3839         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3840         read_unlock(&bond->lock);
3841
3842         return 0;
3843 }
3844
3845 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3846                                      struct ethtool_drvinfo *drvinfo)
3847 {
3848         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3849         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3850         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3851                  BOND_ABI_VERSION);
3852 }
3853
3854 static const struct ethtool_ops bond_ethtool_ops = {
3855         .get_drvinfo            = bond_ethtool_get_drvinfo,
3856         .get_settings           = bond_ethtool_get_settings,
3857         .get_link               = ethtool_op_get_link,
3858 };
3859
3860 static const struct net_device_ops bond_netdev_ops = {
3861         .ndo_init               = bond_init,
3862         .ndo_uninit             = bond_uninit,
3863         .ndo_open               = bond_open,
3864         .ndo_stop               = bond_close,
3865         .ndo_start_xmit         = bond_start_xmit,
3866         .ndo_select_queue       = bond_select_queue,
3867         .ndo_get_stats64        = bond_get_stats,
3868         .ndo_do_ioctl           = bond_do_ioctl,
3869         .ndo_change_rx_flags    = bond_change_rx_flags,
3870         .ndo_set_rx_mode        = bond_set_rx_mode,
3871         .ndo_change_mtu         = bond_change_mtu,
3872         .ndo_set_mac_address    = bond_set_mac_address,
3873         .ndo_neigh_setup        = bond_neigh_setup,
3874         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
3875         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
3876 #ifdef CONFIG_NET_POLL_CONTROLLER
3877         .ndo_netpoll_setup      = bond_netpoll_setup,
3878         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
3879         .ndo_poll_controller    = bond_poll_controller,
3880 #endif
3881         .ndo_add_slave          = bond_enslave,
3882         .ndo_del_slave          = bond_release,
3883         .ndo_fix_features       = bond_fix_features,
3884 };
3885
3886 static const struct device_type bond_type = {
3887         .name = "bond",
3888 };
3889
3890 static void bond_destructor(struct net_device *bond_dev)
3891 {
3892         struct bonding *bond = netdev_priv(bond_dev);
3893         if (bond->wq)
3894                 destroy_workqueue(bond->wq);
3895         free_netdev(bond_dev);
3896 }
3897
3898 void bond_setup(struct net_device *bond_dev)
3899 {
3900         struct bonding *bond = netdev_priv(bond_dev);
3901
3902         /* initialize rwlocks */
3903         rwlock_init(&bond->lock);
3904         rwlock_init(&bond->curr_slave_lock);
3905         bond->params = bonding_defaults;
3906
3907         /* Initialize pointers */
3908         bond->dev = bond_dev;
3909
3910         /* Initialize the device entry points */
3911         ether_setup(bond_dev);
3912         bond_dev->netdev_ops = &bond_netdev_ops;
3913         bond_dev->ethtool_ops = &bond_ethtool_ops;
3914
3915         bond_dev->destructor = bond_destructor;
3916
3917         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
3918
3919         /* Initialize the device options */
3920         bond_dev->tx_queue_len = 0;
3921         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
3922         bond_dev->priv_flags |= IFF_BONDING;
3923         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
3924
3925         /* At first, we block adding VLANs. That's the only way to
3926          * prevent problems that occur when adding VLANs over an
3927          * empty bond. The block will be removed once non-challenged
3928          * slaves are enslaved.
3929          */
3930         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
3931
3932         /* don't acquire bond device's netif_tx_lock when
3933          * transmitting */
3934         bond_dev->features |= NETIF_F_LLTX;
3935
3936         /* By default, we declare the bond to be fully
3937          * VLAN hardware accelerated capable. Special
3938          * care is taken in the various xmit functions
3939          * when there are slaves that are not hw accel
3940          * capable
3941          */
3942
3943         /* Don't allow bond devices to change network namespaces. */
3944         bond_dev->features |= NETIF_F_NETNS_LOCAL;
3945
3946         bond_dev->hw_features = BOND_VLAN_FEATURES |
3947                                 NETIF_F_HW_VLAN_CTAG_TX |
3948                                 NETIF_F_HW_VLAN_CTAG_RX |
3949                                 NETIF_F_HW_VLAN_CTAG_FILTER;
3950
3951         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
3952         bond_dev->features |= bond_dev->hw_features;
3953 }
3954
3955 /*
3956 * Destroy a bonding device.
3957 * Must be under rtnl_lock when this function is called.
3958 */
3959 static void bond_uninit(struct net_device *bond_dev)
3960 {
3961         struct bonding *bond = netdev_priv(bond_dev);
3962         struct list_head *iter;
3963         struct slave *slave;
3964
3965         bond_netpoll_cleanup(bond_dev);
3966
3967         /* Release the bonded slaves */
3968         bond_for_each_slave(bond, slave, iter)
3969                 __bond_release_one(bond_dev, slave->dev, true);
3970         pr_info("%s: Released all slaves\n", bond_dev->name);
3971
3972         list_del(&bond->bond_list);
3973
3974         bond_debug_unregister(bond);
3975 }
3976
3977 /*------------------------- Module initialization ---------------------------*/
3978
3979 static int bond_check_params(struct bond_params *params)
3980 {
3981         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
3982         struct bond_opt_value newval;
3983         const struct bond_opt_value *valptr;
3984         int arp_all_targets_value;
3985
3986         /*
3987          * Convert string parameters.
3988          */
3989         if (mode) {
3990                 bond_opt_initstr(&newval, mode);
3991                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
3992                 if (!valptr) {
3993                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
3994                         return -EINVAL;
3995                 }
3996                 bond_mode = valptr->value;
3997         }
3998
3999         if (xmit_hash_policy) {
4000                 if ((bond_mode != BOND_MODE_XOR) &&
4001                     (bond_mode != BOND_MODE_8023AD)) {
4002                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4003                                 bond_mode_name(bond_mode));
4004                 } else {
4005                         bond_opt_initstr(&newval, xmit_hash_policy);
4006                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4007                                                 &newval);
4008                         if (!valptr) {
4009                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4010                                        xmit_hash_policy);
4011                                 return -EINVAL;
4012                         }
4013                         xmit_hashtype = valptr->value;
4014                 }
4015         }
4016
4017         if (lacp_rate) {
4018                 if (bond_mode != BOND_MODE_8023AD) {
4019                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4020                                 bond_mode_name(bond_mode));
4021                 } else {
4022                         bond_opt_initstr(&newval, lacp_rate);
4023                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4024                                                 &newval);
4025                         if (!valptr) {
4026                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4027                                        lacp_rate);
4028                                 return -EINVAL;
4029                         }
4030                         lacp_fast = valptr->value;
4031                 }
4032         }
4033
4034         if (ad_select) {
4035                 bond_opt_initstr(&newval, lacp_rate);
4036                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4037                                         &newval);
4038                 if (!valptr) {
4039                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4040                         return -EINVAL;
4041                 }
4042                 params->ad_select = valptr->value;
4043                 if (bond_mode != BOND_MODE_8023AD)
4044                         pr_warn("ad_select param only affects 802.3ad mode\n");
4045         } else {
4046                 params->ad_select = BOND_AD_STABLE;
4047         }
4048
4049         if (max_bonds < 0) {
4050                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4051                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4052                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4053         }
4054
4055         if (miimon < 0) {
4056                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4057                         miimon, INT_MAX);
4058                 miimon = 0;
4059         }
4060
4061         if (updelay < 0) {
4062                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4063                         updelay, INT_MAX);
4064                 updelay = 0;
4065         }
4066
4067         if (downdelay < 0) {
4068                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4069                         downdelay, INT_MAX);
4070                 downdelay = 0;
4071         }
4072
4073         if ((use_carrier != 0) && (use_carrier != 1)) {
4074                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4075                         use_carrier);
4076                 use_carrier = 1;
4077         }
4078
4079         if (num_peer_notif < 0 || num_peer_notif > 255) {
4080                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4081                         num_peer_notif);
4082                 num_peer_notif = 1;
4083         }
4084
4085         /* reset values for 802.3ad/TLB/ALB */
4086         if (BOND_NO_USES_ARP(bond_mode)) {
4087                 if (!miimon) {
4088                         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");
4089                         pr_warn("Forcing miimon to 100msec\n");
4090                         miimon = BOND_DEFAULT_MIIMON;
4091                 }
4092         }
4093
4094         if (tx_queues < 1 || tx_queues > 255) {
4095                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4096                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4097                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4098         }
4099
4100         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4101                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4102                         all_slaves_active);
4103                 all_slaves_active = 0;
4104         }
4105
4106         if (resend_igmp < 0 || resend_igmp > 255) {
4107                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4108                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4109                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4110         }
4111
4112         bond_opt_initval(&newval, packets_per_slave);
4113         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4114                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4115                         packets_per_slave, USHRT_MAX);
4116                 packets_per_slave = 1;
4117         }
4118
4119         if (bond_mode == BOND_MODE_ALB) {
4120                 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",
4121                           updelay);
4122         }
4123
4124         if (!miimon) {
4125                 if (updelay || downdelay) {
4126                         /* just warn the user the up/down delay will have
4127                          * no effect since miimon is zero...
4128                          */
4129                         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",
4130                                 updelay, downdelay);
4131                 }
4132         } else {
4133                 /* don't allow arp monitoring */
4134                 if (arp_interval) {
4135                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4136                                 miimon, arp_interval);
4137                         arp_interval = 0;
4138                 }
4139
4140                 if ((updelay % miimon) != 0) {
4141                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4142                                 updelay, miimon, (updelay / miimon) * miimon);
4143                 }
4144
4145                 updelay /= miimon;
4146
4147                 if ((downdelay % miimon) != 0) {
4148                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4149                                 downdelay, miimon,
4150                                 (downdelay / miimon) * miimon);
4151                 }
4152
4153                 downdelay /= miimon;
4154         }
4155
4156         if (arp_interval < 0) {
4157                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4158                         arp_interval, INT_MAX);
4159                 arp_interval = 0;
4160         }
4161
4162         for (arp_ip_count = 0, i = 0;
4163              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4164                 /* not complete check, but should be good enough to
4165                    catch mistakes */
4166                 __be32 ip;
4167                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4168                     IS_IP_TARGET_UNUSABLE_ADDRESS(ip)) {
4169                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4170                                 arp_ip_target[i]);
4171                         arp_interval = 0;
4172                 } else {
4173                         if (bond_get_targets_ip(arp_target, ip) == -1)
4174                                 arp_target[arp_ip_count++] = ip;
4175                         else
4176                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4177                                         &ip);
4178                 }
4179         }
4180
4181         if (arp_interval && !arp_ip_count) {
4182                 /* don't allow arping if no arp_ip_target given... */
4183                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4184                         arp_interval);
4185                 arp_interval = 0;
4186         }
4187
4188         if (arp_validate) {
4189                 if (!arp_interval) {
4190                         pr_err("arp_validate requires arp_interval\n");
4191                         return -EINVAL;
4192                 }
4193
4194                 bond_opt_initstr(&newval, arp_validate);
4195                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4196                                         &newval);
4197                 if (!valptr) {
4198                         pr_err("Error: invalid arp_validate \"%s\"\n",
4199                                arp_validate);
4200                         return -EINVAL;
4201                 }
4202                 arp_validate_value = valptr->value;
4203         } else {
4204                 arp_validate_value = 0;
4205         }
4206
4207         arp_all_targets_value = 0;
4208         if (arp_all_targets) {
4209                 bond_opt_initstr(&newval, arp_all_targets);
4210                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4211                                         &newval);
4212                 if (!valptr) {
4213                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4214                                arp_all_targets);
4215                         arp_all_targets_value = 0;
4216                 } else {
4217                         arp_all_targets_value = valptr->value;
4218                 }
4219         }
4220
4221         if (miimon) {
4222                 pr_info("MII link monitoring set to %d ms\n", miimon);
4223         } else if (arp_interval) {
4224                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4225                                           arp_validate_value);
4226                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4227                         arp_interval, valptr->string, arp_ip_count);
4228
4229                 for (i = 0; i < arp_ip_count; i++)
4230                         pr_cont(" %s", arp_ip_target[i]);
4231
4232                 pr_cont("\n");
4233
4234         } else if (max_bonds) {
4235                 /* miimon and arp_interval not set, we need one so things
4236                  * work as expected, see bonding.txt for details
4237                  */
4238                 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");
4239         }
4240
4241         if (primary && !USES_PRIMARY(bond_mode)) {
4242                 /* currently, using a primary only makes sense
4243                  * in active backup, TLB or ALB modes
4244                  */
4245                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4246                         primary, bond_mode_name(bond_mode));
4247                 primary = NULL;
4248         }
4249
4250         if (primary && primary_reselect) {
4251                 bond_opt_initstr(&newval, primary_reselect);
4252                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4253                                         &newval);
4254                 if (!valptr) {
4255                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4256                                primary_reselect);
4257                         return -EINVAL;
4258                 }
4259                 primary_reselect_value = valptr->value;
4260         } else {
4261                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4262         }
4263
4264         if (fail_over_mac) {
4265                 bond_opt_initstr(&newval, fail_over_mac);
4266                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4267                                         &newval);
4268                 if (!valptr) {
4269                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4270                                fail_over_mac);
4271                         return -EINVAL;
4272                 }
4273                 fail_over_mac_value = valptr->value;
4274                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4275                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4276         } else {
4277                 fail_over_mac_value = BOND_FOM_NONE;
4278         }
4279
4280         if (lp_interval == 0) {
4281                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4282                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4283                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4284         }
4285
4286         /* fill params struct with the proper values */
4287         params->mode = bond_mode;
4288         params->xmit_policy = xmit_hashtype;
4289         params->miimon = miimon;
4290         params->num_peer_notif = num_peer_notif;
4291         params->arp_interval = arp_interval;
4292         params->arp_validate = arp_validate_value;
4293         params->arp_all_targets = arp_all_targets_value;
4294         params->updelay = updelay;
4295         params->downdelay = downdelay;
4296         params->use_carrier = use_carrier;
4297         params->lacp_fast = lacp_fast;
4298         params->primary[0] = 0;
4299         params->primary_reselect = primary_reselect_value;
4300         params->fail_over_mac = fail_over_mac_value;
4301         params->tx_queues = tx_queues;
4302         params->all_slaves_active = all_slaves_active;
4303         params->resend_igmp = resend_igmp;
4304         params->min_links = min_links;
4305         params->lp_interval = lp_interval;
4306         params->packets_per_slave = packets_per_slave;
4307         if (packets_per_slave > 0) {
4308                 params->reciprocal_packets_per_slave =
4309                         reciprocal_value(packets_per_slave);
4310         } else {
4311                 /* reciprocal_packets_per_slave is unused if
4312                  * packets_per_slave is 0 or 1, just initialize it
4313                  */
4314                 params->reciprocal_packets_per_slave =
4315                         (struct reciprocal_value) { 0 };
4316         }
4317
4318         if (primary) {
4319                 strncpy(params->primary, primary, IFNAMSIZ);
4320                 params->primary[IFNAMSIZ - 1] = 0;
4321         }
4322
4323         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4324
4325         return 0;
4326 }
4327
4328 static struct lock_class_key bonding_netdev_xmit_lock_key;
4329 static struct lock_class_key bonding_netdev_addr_lock_key;
4330 static struct lock_class_key bonding_tx_busylock_key;
4331
4332 static void bond_set_lockdep_class_one(struct net_device *dev,
4333                                        struct netdev_queue *txq,
4334                                        void *_unused)
4335 {
4336         lockdep_set_class(&txq->_xmit_lock,
4337                           &bonding_netdev_xmit_lock_key);
4338 }
4339
4340 static void bond_set_lockdep_class(struct net_device *dev)
4341 {
4342         lockdep_set_class(&dev->addr_list_lock,
4343                           &bonding_netdev_addr_lock_key);
4344         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4345         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4346 }
4347
4348 /*
4349  * Called from registration process
4350  */
4351 static int bond_init(struct net_device *bond_dev)
4352 {
4353         struct bonding *bond = netdev_priv(bond_dev);
4354         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4355         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4356
4357         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4358
4359         /*
4360          * Initialize locks that may be required during
4361          * en/deslave operations.  All of the bond_open work
4362          * (of which this is part) should really be moved to
4363          * a phase prior to dev_open
4364          */
4365         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4366         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4367
4368         bond->wq = create_singlethread_workqueue(bond_dev->name);
4369         if (!bond->wq)
4370                 return -ENOMEM;
4371
4372         bond_set_lockdep_class(bond_dev);
4373
4374         list_add_tail(&bond->bond_list, &bn->dev_list);
4375
4376         bond_prepare_sysfs_group(bond);
4377
4378         bond_debug_register(bond);
4379
4380         /* Ensure valid dev_addr */
4381         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4382             bond_dev->addr_assign_type == NET_ADDR_PERM)
4383                 eth_hw_addr_random(bond_dev);
4384
4385         return 0;
4386 }
4387
4388 unsigned int bond_get_num_tx_queues(void)
4389 {
4390         return tx_queues;
4391 }
4392
4393 /* Create a new bond based on the specified name and bonding parameters.
4394  * If name is NULL, obtain a suitable "bond%d" name for us.
4395  * Caller must NOT hold rtnl_lock; we need to release it here before we
4396  * set up our sysfs entries.
4397  */
4398 int bond_create(struct net *net, const char *name)
4399 {
4400         struct net_device *bond_dev;
4401         int res;
4402
4403         rtnl_lock();
4404
4405         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4406                                    name ? name : "bond%d",
4407                                    bond_setup, tx_queues);
4408         if (!bond_dev) {
4409                 pr_err("%s: eek! can't alloc netdev!\n", name);
4410                 rtnl_unlock();
4411                 return -ENOMEM;
4412         }
4413
4414         dev_net_set(bond_dev, net);
4415         bond_dev->rtnl_link_ops = &bond_link_ops;
4416
4417         res = register_netdevice(bond_dev);
4418
4419         netif_carrier_off(bond_dev);
4420
4421         rtnl_unlock();
4422         if (res < 0)
4423                 bond_destructor(bond_dev);
4424         return res;
4425 }
4426
4427 static int __net_init bond_net_init(struct net *net)
4428 {
4429         struct bond_net *bn = net_generic(net, bond_net_id);
4430
4431         bn->net = net;
4432         INIT_LIST_HEAD(&bn->dev_list);
4433
4434         bond_create_proc_dir(bn);
4435         bond_create_sysfs(bn);
4436
4437         return 0;
4438 }
4439
4440 static void __net_exit bond_net_exit(struct net *net)
4441 {
4442         struct bond_net *bn = net_generic(net, bond_net_id);
4443         struct bonding *bond, *tmp_bond;
4444         LIST_HEAD(list);
4445
4446         bond_destroy_sysfs(bn);
4447         bond_destroy_proc_dir(bn);
4448
4449         /* Kill off any bonds created after unregistering bond rtnl ops */
4450         rtnl_lock();
4451         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4452                 unregister_netdevice_queue(bond->dev, &list);
4453         unregister_netdevice_many(&list);
4454         rtnl_unlock();
4455 }
4456
4457 static struct pernet_operations bond_net_ops = {
4458         .init = bond_net_init,
4459         .exit = bond_net_exit,
4460         .id   = &bond_net_id,
4461         .size = sizeof(struct bond_net),
4462 };
4463
4464 static int __init bonding_init(void)
4465 {
4466         int i;
4467         int res;
4468
4469         pr_info("%s", bond_version);
4470
4471         res = bond_check_params(&bonding_defaults);
4472         if (res)
4473                 goto out;
4474
4475         res = register_pernet_subsys(&bond_net_ops);
4476         if (res)
4477                 goto out;
4478
4479         res = bond_netlink_init();
4480         if (res)
4481                 goto err_link;
4482
4483         bond_create_debugfs();
4484
4485         for (i = 0; i < max_bonds; i++) {
4486                 res = bond_create(&init_net, NULL);
4487                 if (res)
4488                         goto err;
4489         }
4490
4491         register_netdevice_notifier(&bond_netdev_notifier);
4492 out:
4493         return res;
4494 err:
4495         bond_destroy_debugfs();
4496         bond_netlink_fini();
4497 err_link:
4498         unregister_pernet_subsys(&bond_net_ops);
4499         goto out;
4500
4501 }
4502
4503 static void __exit bonding_exit(void)
4504 {
4505         unregister_netdevice_notifier(&bond_netdev_notifier);
4506
4507         bond_destroy_debugfs();
4508
4509         bond_netlink_fini();
4510         unregister_pernet_subsys(&bond_net_ops);
4511
4512 #ifdef CONFIG_NET_POLL_CONTROLLER
4513         /*
4514          * Make sure we don't have an imbalance on our netpoll blocking
4515          */
4516         WARN_ON(atomic_read(&netpoll_block_tx));
4517 #endif
4518 }
4519
4520 module_init(bonding_init);
4521 module_exit(bonding_exit);
4522 MODULE_LICENSE("GPL");
4523 MODULE_VERSION(DRV_VERSION);
4524 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4525 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");