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                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 677                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 678                                ETH_ALEN);
 679                         saddr.sa_family = new_active->dev->type;
 680                 } else {
 681                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 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                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 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->jiffies = 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_ATOMIC);
 926         err = -ENOMEM;
 927         if (!np)
 928                 goto out;
 929
 930         err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
 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 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
 950 {
 951         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
 952                 return false;
 953         if (!slave_dev->netdev_ops->ndo_poll_controller)
 954                 return false;
 955         return true;
 956 }
 957
 958 static void bond_poll_controller(struct net_device *bond_dev)
 959 {
 960 }
 961
 962 static void bond_netpoll_cleanup(struct net_device *bond_dev)
 963 {
 964         struct bonding *bond = netdev_priv(bond_dev);
 965         struct list_head *iter;
 966         struct slave *slave;
 967
 968         bond_for_each_slave(bond, slave, iter)
 969                 if (IS_UP(slave->dev))
 970                         slave_disable_netpoll(slave);
 971 }
 972
 973 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
 974 {
 975         struct bonding *bond = netdev_priv(dev);
 976         struct list_head *iter;
 977         struct slave *slave;
 978         int err = 0;
 979
 980         bond_for_each_slave(bond, slave, iter) {
 981                 err = slave_enable_netpoll(slave);
 982                 if (err) {
 983                         bond_netpoll_cleanup(dev);
 984                         break;
 985                 }
 986         }
 987         return err;
 988 }
 989 #else
 990 static inline int slave_enable_netpoll(struct slave *slave)
 991 {
 992         return 0;
 993 }
 994 static inline void slave_disable_netpoll(struct slave *slave)
 995 {
 996 }
 997 static void bond_netpoll_cleanup(struct net_device *bond_dev)
 998 {
 999 }
1000 #endif
1001
1002 /*---------------------------------- IOCTL ----------------------------------*/
1003
1004 static netdev_features_t bond_fix_features(struct net_device *dev,
1005                                            netdev_features_t features)
1006 {
1007         struct bonding *bond = netdev_priv(dev);
1008         struct list_head *iter;
1009         netdev_features_t mask;
1010         struct slave *slave;
1011
1012         if (!bond_has_slaves(bond)) {
1013                 /* Disable adding VLANs to empty bond. But why? --mq */
1014                 features |= NETIF_F_VLAN_CHALLENGED;
1015                 return features;
1016         }
1017
1018         mask = features;
1019         features &= ~NETIF_F_ONE_FOR_ALL;
1020         features |= NETIF_F_ALL_FOR_ALL;
1021
1022         bond_for_each_slave(bond, slave, iter) {
1023                 features = netdev_increment_features(features,
1024                                                      slave->dev->features,
1025                                                      mask);
1026         }
1027         features = netdev_add_tso_features(features, mask);
1028
1029         return features;
1030 }
1031
1032 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1033                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1034                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1035
1036 static void bond_compute_features(struct bonding *bond)
1037 {
1038         unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1039         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1040         struct net_device *bond_dev = bond->dev;
1041         struct list_head *iter;
1042         struct slave *slave;
1043         unsigned short max_hard_header_len = ETH_HLEN;
1044         unsigned int gso_max_size = GSO_MAX_SIZE;
1045         u16 gso_max_segs = GSO_MAX_SEGS;
1046
1047         if (!bond_has_slaves(bond))
1048                 goto done;
1049
1050         bond_for_each_slave(bond, slave, iter) {
1051                 vlan_features = netdev_increment_features(vlan_features,
1052                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1053
1054                 dst_release_flag &= slave->dev->priv_flags;
1055                 if (slave->dev->hard_header_len > max_hard_header_len)
1056                         max_hard_header_len = slave->dev->hard_header_len;
1057
1058                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1059                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1060         }
1061
1062 done:
1063         bond_dev->vlan_features = vlan_features;
1064         bond_dev->hard_header_len = max_hard_header_len;
1065         bond_dev->gso_max_segs = gso_max_segs;
1066         netif_set_gso_max_size(bond_dev, gso_max_size);
1067
1068         flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1069         bond_dev->priv_flags = flags | dst_release_flag;
1070
1071         netdev_change_features(bond_dev);
1072 }
1073
1074 static void bond_setup_by_slave(struct net_device *bond_dev,
1075                                 struct net_device *slave_dev)
1076 {
1077         bond_dev->header_ops        = slave_dev->header_ops;
1078
1079         bond_dev->type              = slave_dev->type;
1080         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1081         bond_dev->addr_len          = slave_dev->addr_len;
1082
1083         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1084                 slave_dev->addr_len);
1085 }
1086
1087 /* On bonding slaves other than the currently active slave, suppress
1088  * duplicates except for alb non-mcast/bcast.
1089  */
1090 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1091                                             struct slave *slave,
1092                                             struct bonding *bond)
1093 {
1094         if (bond_is_slave_inactive(slave)) {
1095                 if (bond->params.mode == BOND_MODE_ALB &&
1096                     skb->pkt_type != PACKET_BROADCAST &&
1097                     skb->pkt_type != PACKET_MULTICAST)
1098                         return false;
1099                 return true;
1100         }
1101         return false;
1102 }
1103
1104 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1105 {
1106         struct sk_buff *skb = *pskb;
1107         struct slave *slave;
1108         struct bonding *bond;
1109         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1110                           struct slave *);
1111         int ret = RX_HANDLER_ANOTHER;
1112
1113         skb = skb_share_check(skb, GFP_ATOMIC);
1114         if (unlikely(!skb))
1115                 return RX_HANDLER_CONSUMED;
1116
1117         *pskb = skb;
1118
1119         slave = bond_slave_get_rcu(skb->dev);
1120         bond = slave->bond;
1121
1122         if (bond->params.arp_interval)
1123                 slave->dev->last_rx = jiffies;
1124
1125         recv_probe = ACCESS_ONCE(bond->recv_probe);
1126         if (recv_probe) {
1127                 ret = recv_probe(skb, bond, slave);
1128                 if (ret == RX_HANDLER_CONSUMED) {
1129                         consume_skb(skb);
1130                         return ret;
1131                 }
1132         }
1133
1134         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1135                 return RX_HANDLER_EXACT;
1136         }
1137
1138         skb->dev = bond->dev;
1139
1140         if (bond->params.mode == BOND_MODE_ALB &&
1141             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1142             skb->pkt_type == PACKET_HOST) {
1143
1144                 if (unlikely(skb_cow_head(skb,
1145                                           skb->data - skb_mac_header(skb)))) {
1146                         kfree_skb(skb);
1147                         return RX_HANDLER_CONSUMED;
1148                 }
1149                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1150         }
1151
1152         return ret;
1153 }
1154
1155 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1156                                       struct net_device *slave_dev,
1157                                       struct slave *slave)
1158 {
1159         int err;
1160
1161         err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1162         if (err)
1163                 return err;
1164         slave_dev->flags |= IFF_SLAVE;
1165         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1166         return 0;
1167 }
1168
1169 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1170                                   struct net_device *slave_dev)
1171 {
1172         netdev_upper_dev_unlink(slave_dev, bond_dev);
1173         slave_dev->flags &= ~IFF_SLAVE;
1174         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1175 }
1176
1177 /* enslave device <slave> to bond device <master> */
1178 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1179 {
1180         struct bonding *bond = netdev_priv(bond_dev);
1181         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1182         struct slave *new_slave = NULL, *prev_slave;
1183         struct sockaddr addr;
1184         int link_reporting;
1185         int res = 0, i;
1186
1187         if (!bond->params.use_carrier &&
1188             slave_dev->ethtool_ops->get_link == NULL &&
1189             slave_ops->ndo_do_ioctl == NULL) {
1190                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1191                            bond_dev->name, slave_dev->name);
1192         }
1193
1194         /* already enslaved */
1195         if (slave_dev->flags & IFF_SLAVE) {
1196                 pr_debug("Error, Device was already enslaved\n");
1197                 return -EBUSY;
1198         }
1199
1200         /* vlan challenged mutual exclusion */
1201         /* no need to lock since we're protected by rtnl_lock */
1202         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1203                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1204                 if (vlan_uses_dev(bond_dev)) {
1205                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1206                                bond_dev->name, slave_dev->name, bond_dev->name);
1207                         return -EPERM;
1208                 } else {
1209                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1210                                    bond_dev->name, slave_dev->name,
1211                                    slave_dev->name, bond_dev->name);
1212                 }
1213         } else {
1214                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1215         }
1216
1217         /*
1218          * Old ifenslave binaries are no longer supported.  These can
1219          * be identified with moderate accuracy by the state of the slave:
1220          * the current ifenslave will set the interface down prior to
1221          * enslaving it; the old ifenslave will not.
1222          */
1223         if ((slave_dev->flags & IFF_UP)) {
1224                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1225                        slave_dev->name);
1226                 res = -EPERM;
1227                 goto err_undo_flags;
1228         }
1229
1230         /* set bonding device ether type by slave - bonding netdevices are
1231          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1232          * there is a need to override some of the type dependent attribs/funcs.
1233          *
1234          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1235          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1236          */
1237         if (!bond_has_slaves(bond)) {
1238                 if (bond_dev->type != slave_dev->type) {
1239                         pr_debug("%s: change device type from %d to %d\n",
1240                                  bond_dev->name,
1241                                  bond_dev->type, slave_dev->type);
1242
1243                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1244                                                        bond_dev);
1245                         res = notifier_to_errno(res);
1246                         if (res) {
1247                                 pr_err("%s: refused to change device type\n",
1248                                        bond_dev->name);
1249                                 res = -EBUSY;
1250                                 goto err_undo_flags;
1251                         }
1252
1253                         /* Flush unicast and multicast addresses */
1254                         dev_uc_flush(bond_dev);
1255                         dev_mc_flush(bond_dev);
1256
1257                         if (slave_dev->type != ARPHRD_ETHER)
1258                                 bond_setup_by_slave(bond_dev, slave_dev);
1259                         else {
1260                                 ether_setup(bond_dev);
1261                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1262                         }
1263
1264                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1265                                                  bond_dev);
1266                 }
1267         } else if (bond_dev->type != slave_dev->type) {
1268                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1269                        slave_dev->name,
1270                        slave_dev->type, bond_dev->type);
1271                 res = -EINVAL;
1272                 goto err_undo_flags;
1273         }
1274
1275         if (slave_ops->ndo_set_mac_address == NULL) {
1276                 if (!bond_has_slaves(bond)) {
1277                         pr_warn("%s: Warning: The first slave device specified does not support setting the MAC address.\n",
1278                                 bond_dev->name);
1279                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1280                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1281                                 pr_warn("%s: Setting fail_over_mac to active for active-backup mode.\n",
1282                                         bond_dev->name);
1283                         }
1284                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1285                         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",
1286                                bond_dev->name);
1287                         res = -EOPNOTSUPP;
1288                         goto err_undo_flags;
1289                 }
1290         }
1291
1292         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1293
1294         /* If this is the first slave, then we need to set the master's hardware
1295          * address to be the same as the slave's. */
1296         if (!bond_has_slaves(bond) &&
1297             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1298                 bond_set_dev_addr(bond->dev, slave_dev);
1299
1300         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1301         if (!new_slave) {
1302                 res = -ENOMEM;
1303                 goto err_undo_flags;
1304         }
1305         /*
1306          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1307          * is set via sysfs or module option if desired.
1308          */
1309         new_slave->queue_id = 0;
1310
1311         /* Save slave's original mtu and then set it to match the bond */
1312         new_slave->original_mtu = slave_dev->mtu;
1313         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1314         if (res) {
1315                 pr_debug("Error %d calling dev_set_mtu\n", res);
1316                 goto err_free;
1317         }
1318
1319         /*
1320          * Save slave's original ("permanent") mac address for modes
1321          * that need it, and for restoring it upon release, and then
1322          * set it to the master's address
1323          */
1324         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1325
1326         if (!bond->params.fail_over_mac ||
1327             bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1328                 /*
1329                  * Set slave to master's mac address.  The application already
1330                  * set the master's mac address to that of the first slave
1331                  */
1332                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1333                 addr.sa_family = slave_dev->type;
1334                 res = dev_set_mac_address(slave_dev, &addr);
1335                 if (res) {
1336                         pr_debug("Error %d calling set_mac_address\n", res);
1337                         goto err_restore_mtu;
1338                 }
1339         }
1340
1341         /* open the slave since the application closed it */
1342         res = dev_open(slave_dev);
1343         if (res) {
1344                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1345                 goto err_restore_mac;
1346         }
1347
1348         new_slave->bond = bond;
1349         new_slave->dev = slave_dev;
1350         slave_dev->priv_flags |= IFF_BONDING;
1351
1352         if (bond_is_lb(bond)) {
1353                 /* bond_alb_init_slave() must be called before all other stages since
1354                  * it might fail and we do not want to have to undo everything
1355                  */
1356                 res = bond_alb_init_slave(bond, new_slave);
1357                 if (res)
1358                         goto err_close;
1359         }
1360
1361         /* If the mode USES_PRIMARY, then the following is handled by
1362          * bond_change_active_slave().
1363          */
1364         if (!USES_PRIMARY(bond->params.mode)) {
1365                 /* set promiscuity level to new slave */
1366                 if (bond_dev->flags & IFF_PROMISC) {
1367                         res = dev_set_promiscuity(slave_dev, 1);
1368                         if (res)
1369                                 goto err_close;
1370                 }
1371
1372                 /* set allmulti level to new slave */
1373                 if (bond_dev->flags & IFF_ALLMULTI) {
1374                         res = dev_set_allmulti(slave_dev, 1);
1375                         if (res)
1376                                 goto err_close;
1377                 }
1378
1379                 netif_addr_lock_bh(bond_dev);
1380
1381                 dev_mc_sync_multiple(slave_dev, bond_dev);
1382                 dev_uc_sync_multiple(slave_dev, bond_dev);
1383
1384                 netif_addr_unlock_bh(bond_dev);
1385         }
1386
1387         if (bond->params.mode == BOND_MODE_8023AD) {
1388                 /* add lacpdu mc addr to mc list */
1389                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1390
1391                 dev_mc_add(slave_dev, lacpdu_multicast);
1392         }
1393
1394         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1395         if (res) {
1396                 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1397                        bond_dev->name, slave_dev->name);
1398                 goto err_close;
1399         }
1400
1401         prev_slave = bond_last_slave(bond);
1402
1403         new_slave->delay = 0;
1404         new_slave->link_failure_count = 0;
1405
1406         bond_update_speed_duplex(new_slave);
1407
1408         new_slave->last_arp_rx = jiffies -
1409                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1410         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1411                 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
1412
1413         if (bond->params.miimon && !bond->params.use_carrier) {
1414                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1415
1416                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1417                         /*
1418                          * miimon is set but a bonded network driver
1419                          * does not support ETHTOOL/MII and
1420                          * arp_interval is not set.  Note: if
1421                          * use_carrier is enabled, we will never go
1422                          * here (because netif_carrier is always
1423                          * supported); thus, we don't need to change
1424                          * the messages for netif_carrier.
1425                          */
1426                         pr_warning("%s: Warning: MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details.\n",
1427                                bond_dev->name, slave_dev->name);
1428                 } else if (link_reporting == -1) {
1429                         /* unable get link status using mii/ethtool */
1430                         pr_warning("%s: Warning: can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface.\n",
1431                                    bond_dev->name, slave_dev->name);
1432                 }
1433         }
1434
1435         /* check for initial state */
1436         if (bond->params.miimon) {
1437                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1438                         if (bond->params.updelay) {
1439                                 new_slave->link = BOND_LINK_BACK;
1440                                 new_slave->delay = bond->params.updelay;
1441                         } else {
1442                                 new_slave->link = BOND_LINK_UP;
1443                         }
1444                 } else {
1445                         new_slave->link = BOND_LINK_DOWN;
1446                 }
1447         } else if (bond->params.arp_interval) {
1448                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1449                         BOND_LINK_UP : BOND_LINK_DOWN);
1450         } else {
1451                 new_slave->link = BOND_LINK_UP;
1452         }
1453
1454         if (new_slave->link != BOND_LINK_DOWN)
1455                 new_slave->jiffies = jiffies;
1456         pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1457                 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1458                         (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1459
1460         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1461                 /* if there is a primary slave, remember it */
1462                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1463                         bond->primary_slave = new_slave;
1464                         bond->force_primary = true;
1465                 }
1466         }
1467
1468         switch (bond->params.mode) {
1469         case BOND_MODE_ACTIVEBACKUP:
1470                 bond_set_slave_inactive_flags(new_slave,
1471                                               BOND_SLAVE_NOTIFY_NOW);
1472                 break;
1473         case BOND_MODE_8023AD:
1474                 /* in 802.3ad mode, the internal mechanism
1475                  * will activate the slaves in the selected
1476                  * aggregator
1477                  */
1478                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1479                 /* if this is the first slave */
1480                 if (!prev_slave) {
1481                         SLAVE_AD_INFO(new_slave).id = 1;
1482                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1483                          * can be called only after the mac address of the bond is set
1484                          */
1485                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1486                 } else {
1487                         SLAVE_AD_INFO(new_slave).id =
1488                                 SLAVE_AD_INFO(prev_slave).id + 1;
1489                 }
1490
1491                 bond_3ad_bind_slave(new_slave);
1492                 break;
1493         case BOND_MODE_TLB:
1494         case BOND_MODE_ALB:
1495                 bond_set_active_slave(new_slave);
1496                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1497                 break;
1498         default:
1499                 pr_debug("This slave is always active in trunk mode\n");
1500
1501                 /* always active in trunk mode */
1502                 bond_set_active_slave(new_slave);
1503
1504                 /* In trunking mode there is little meaning to curr_active_slave
1505                  * anyway (it holds no special properties of the bond device),
1506                  * so we can change it without calling change_active_interface()
1507                  */
1508                 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1509                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1510
1511                 break;
1512         } /* switch(bond_mode) */
1513
1514 #ifdef CONFIG_NET_POLL_CONTROLLER
1515         slave_dev->npinfo = bond->dev->npinfo;
1516         if (slave_dev->npinfo) {
1517                 if (slave_enable_netpoll(new_slave)) {
1518                         pr_info("Error, %s: master_dev is using netpoll, "
1519                                  "but new slave device does not support netpoll.\n",
1520                                  bond_dev->name);
1521                         res = -EBUSY;
1522                         goto err_detach;
1523                 }
1524         }
1525 #endif
1526
1527         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1528                                          new_slave);
1529         if (res) {
1530                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1531                 goto err_detach;
1532         }
1533
1534         res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1535         if (res) {
1536                 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1537                 goto err_unregister;
1538         }
1539
1540         res = bond_sysfs_slave_add(new_slave);
1541         if (res) {
1542                 pr_debug("Error %d calling bond_sysfs_slave_add\n", res);
1543                 goto err_upper_unlink;
1544         }
1545
1546         bond->slave_cnt++;
1547         bond_compute_features(bond);
1548         bond_set_carrier(bond);
1549
1550         if (USES_PRIMARY(bond->params.mode)) {
1551                 block_netpoll_tx();
1552                 write_lock_bh(&bond->curr_slave_lock);
1553                 bond_select_active_slave(bond);
1554                 write_unlock_bh(&bond->curr_slave_lock);
1555                 unblock_netpoll_tx();
1556         }
1557
1558         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1559                 bond_dev->name, slave_dev->name,
1560                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1561                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1562
1563         /* enslave is successful */
1564         return 0;
1565
1566 /* Undo stages on error */
1567 err_upper_unlink:
1568         bond_upper_dev_unlink(bond_dev, slave_dev);
1569
1570 err_unregister:
1571         netdev_rx_handler_unregister(slave_dev);
1572
1573 err_detach:
1574         if (!USES_PRIMARY(bond->params.mode))
1575                 bond_hw_addr_flush(bond_dev, slave_dev);
1576
1577         vlan_vids_del_by_dev(slave_dev, bond_dev);
1578         if (bond->primary_slave == new_slave)
1579                 bond->primary_slave = NULL;
1580         if (bond->curr_active_slave == new_slave) {
1581                 block_netpoll_tx();
1582                 write_lock_bh(&bond->curr_slave_lock);
1583                 bond_change_active_slave(bond, NULL);
1584                 bond_select_active_slave(bond);
1585                 write_unlock_bh(&bond->curr_slave_lock);
1586                 unblock_netpoll_tx();
1587         }
1588         slave_disable_netpoll(new_slave);
1589
1590 err_close:
1591         slave_dev->priv_flags &= ~IFF_BONDING;
1592         dev_close(slave_dev);
1593
1594 err_restore_mac:
1595         if (!bond->params.fail_over_mac ||
1596             bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1597                 /* XXX TODO - fom follow mode needs to change master's
1598                  * MAC if this slave's MAC is in use by the bond, or at
1599                  * least print a warning.
1600                  */
1601                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1602                 addr.sa_family = slave_dev->type;
1603                 dev_set_mac_address(slave_dev, &addr);
1604         }
1605
1606 err_restore_mtu:
1607         dev_set_mtu(slave_dev, new_slave->original_mtu);
1608
1609 err_free:
1610         kfree(new_slave);
1611
1612 err_undo_flags:
1613         /* Enslave of first slave has failed and we need to fix master's mac */
1614         if (!bond_has_slaves(bond) &&
1615             ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
1616                 eth_hw_addr_random(bond_dev);
1617
1618         return res;
1619 }
1620
1621 /*
1622  * Try to release the slave device <slave> from the bond device <master>
1623  * It is legal to access curr_active_slave without a lock because all the function
1624  * is write-locked. If "all" is true it means that the function is being called
1625  * while destroying a bond interface and all slaves are being released.
1626  *
1627  * The rules for slave state should be:
1628  *   for Active/Backup:
1629  *     Active stays on all backups go down
1630  *   for Bonded connections:
1631  *     The first up interface should be left on and all others downed.
1632  */
1633 static int __bond_release_one(struct net_device *bond_dev,
1634                               struct net_device *slave_dev,
1635                               bool all)
1636 {
1637         struct bonding *bond = netdev_priv(bond_dev);
1638         struct slave *slave, *oldcurrent;
1639         struct sockaddr addr;
1640         int old_flags = bond_dev->flags;
1641         netdev_features_t old_features = bond_dev->features;
1642
1643         /* slave is not a slave or master is not master of this slave */
1644         if (!(slave_dev->flags & IFF_SLAVE) ||
1645             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1646                 pr_err("%s: Error: cannot release %s.\n",
1647                        bond_dev->name, slave_dev->name);
1648                 return -EINVAL;
1649         }
1650
1651         block_netpoll_tx();
1652
1653         slave = bond_get_slave_by_dev(bond, slave_dev);
1654         if (!slave) {
1655                 /* not a slave of this bond */
1656                 pr_info("%s: %s not enslaved\n",
1657                         bond_dev->name, slave_dev->name);
1658                 unblock_netpoll_tx();
1659                 return -EINVAL;
1660         }
1661
1662         /* release the slave from its bond */
1663         bond->slave_cnt--;
1664
1665         bond_sysfs_slave_del(slave);
1666
1667         bond_upper_dev_unlink(bond_dev, slave_dev);
1668         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1669          * for this slave anymore.
1670          */
1671         netdev_rx_handler_unregister(slave_dev);
1672         write_lock_bh(&bond->lock);
1673
1674         /* Inform AD package of unbinding of slave. */
1675         if (bond->params.mode == BOND_MODE_8023AD)
1676                 bond_3ad_unbind_slave(slave);
1677
1678         write_unlock_bh(&bond->lock);
1679
1680         pr_info("%s: releasing %s interface %s\n",
1681                 bond_dev->name,
1682                 bond_is_active_slave(slave) ? "active" : "backup",
1683                 slave_dev->name);
1684
1685         oldcurrent = bond->curr_active_slave;
1686
1687         bond->current_arp_slave = NULL;
1688
1689         if (!all && (!bond->params.fail_over_mac ||
1690                      bond->params.mode != BOND_MODE_ACTIVEBACKUP)) {
1691                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1692                     bond_has_slaves(bond))
1693                         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",
1694                                    bond_dev->name, slave_dev->name,
1695                                    slave->perm_hwaddr,
1696                                    bond_dev->name, slave_dev->name);
1697         }
1698
1699         if (bond->primary_slave == slave)
1700                 bond->primary_slave = NULL;
1701
1702         if (oldcurrent == slave) {
1703                 write_lock_bh(&bond->curr_slave_lock);
1704                 bond_change_active_slave(bond, NULL);
1705                 write_unlock_bh(&bond->curr_slave_lock);
1706         }
1707
1708         if (bond_is_lb(bond)) {
1709                 /* Must be called only after the slave has been
1710                  * detached from the list and the curr_active_slave
1711                  * has been cleared (if our_slave == old_current),
1712                  * but before a new active slave is selected.
1713                  */
1714                 bond_alb_deinit_slave(bond, slave);
1715         }
1716
1717         if (all) {
1718                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1719         } else if (oldcurrent == slave) {
1720                 /*
1721                  * Note that we hold RTNL over this sequence, so there
1722                  * is no concern that another slave add/remove event
1723                  * will interfere.
1724                  */
1725                 write_lock_bh(&bond->curr_slave_lock);
1726
1727                 bond_select_active_slave(bond);
1728
1729                 write_unlock_bh(&bond->curr_slave_lock);
1730         }
1731
1732         if (!bond_has_slaves(bond)) {
1733                 bond_set_carrier(bond);
1734                 eth_hw_addr_random(bond_dev);
1735
1736                 if (vlan_uses_dev(bond_dev)) {
1737                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1738                                    bond_dev->name, bond_dev->name);
1739                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1740                                    bond_dev->name);
1741                 }
1742         }
1743
1744         unblock_netpoll_tx();
1745         synchronize_rcu();
1746
1747         if (!bond_has_slaves(bond)) {
1748                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1749                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1750         }
1751
1752         bond_compute_features(bond);
1753         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1754             (old_features & NETIF_F_VLAN_CHALLENGED))
1755                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1756                         bond_dev->name, slave_dev->name, bond_dev->name);
1757
1758         /* must do this from outside any spinlocks */
1759         vlan_vids_del_by_dev(slave_dev, bond_dev);
1760
1761         /* If the mode USES_PRIMARY, then this cases was handled above by
1762          * bond_change_active_slave(..., NULL)
1763          */
1764         if (!USES_PRIMARY(bond->params.mode)) {
1765                 /* unset promiscuity level from slave
1766                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1767                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1768                  * value of that flag before that change, as that was the value
1769                  * when this slave was attached, so we cache at the start of the
1770                  * function and use it here. Same goes for ALLMULTI below
1771                  */
1772                 if (old_flags & IFF_PROMISC)
1773                         dev_set_promiscuity(slave_dev, -1);
1774
1775                 /* unset allmulti level from slave */
1776                 if (old_flags & IFF_ALLMULTI)
1777                         dev_set_allmulti(slave_dev, -1);
1778
1779                 bond_hw_addr_flush(bond_dev, slave_dev);
1780         }
1781
1782         slave_disable_netpoll(slave);
1783
1784         /* close slave before restoring its mac address */
1785         dev_close(slave_dev);
1786
1787         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1788             bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1789                 /* restore original ("permanent") mac address */
1790                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1791                 addr.sa_family = slave_dev->type;
1792                 dev_set_mac_address(slave_dev, &addr);
1793         }
1794
1795         dev_set_mtu(slave_dev, slave->original_mtu);
1796
1797         slave_dev->priv_flags &= ~IFF_BONDING;
1798
1799         kfree(slave);
1800
1801         return 0;  /* deletion OK */
1802 }
1803
1804 /* A wrapper used because of ndo_del_link */
1805 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1806 {
1807         return __bond_release_one(bond_dev, slave_dev, false);
1808 }
1809
1810 /*
1811 * First release a slave and then destroy the bond if no more slaves are left.
1812 * Must be under rtnl_lock when this function is called.
1813 */
1814 static int  bond_release_and_destroy(struct net_device *bond_dev,
1815                                      struct net_device *slave_dev)
1816 {
1817         struct bonding *bond = netdev_priv(bond_dev);
1818         int ret;
1819
1820         ret = bond_release(bond_dev, slave_dev);
1821         if (ret == 0 && !bond_has_slaves(bond)) {
1822                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1823                 pr_info("%s: destroying bond %s.\n",
1824                         bond_dev->name, bond_dev->name);
1825                 unregister_netdevice(bond_dev);
1826         }
1827         return ret;
1828 }
1829
1830 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1831 {
1832         struct bonding *bond = netdev_priv(bond_dev);
1833
1834         info->bond_mode = bond->params.mode;
1835         info->miimon = bond->params.miimon;
1836
1837         read_lock(&bond->lock);
1838         info->num_slaves = bond->slave_cnt;
1839         read_unlock(&bond->lock);
1840
1841         return 0;
1842 }
1843
1844 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1845 {
1846         struct bonding *bond = netdev_priv(bond_dev);
1847         struct list_head *iter;
1848         int i = 0, res = -ENODEV;
1849         struct slave *slave;
1850
1851         read_lock(&bond->lock);
1852         bond_for_each_slave(bond, slave, iter) {
1853                 if (i++ == (int)info->slave_id) {
1854                         res = 0;
1855                         strcpy(info->slave_name, slave->dev->name);
1856                         info->link = slave->link;
1857                         info->state = bond_slave_state(slave);
1858                         info->link_failure_count = slave->link_failure_count;
1859                         break;
1860                 }
1861         }
1862         read_unlock(&bond->lock);
1863
1864         return res;
1865 }
1866
1867 /*-------------------------------- Monitoring -------------------------------*/
1868
1869
1870 static int bond_miimon_inspect(struct bonding *bond)
1871 {
1872         int link_state, commit = 0;
1873         struct list_head *iter;
1874         struct slave *slave;
1875         bool ignore_updelay;
1876
1877         ignore_updelay = !bond->curr_active_slave ? true : false;
1878
1879         bond_for_each_slave_rcu(bond, slave, iter) {
1880                 slave->new_link = BOND_LINK_NOCHANGE;
1881
1882                 link_state = bond_check_dev_link(bond, slave->dev, 0);
1883
1884                 switch (slave->link) {
1885                 case BOND_LINK_UP:
1886                         if (link_state)
1887                                 continue;
1888
1889                         slave->link = BOND_LINK_FAIL;
1890                         slave->delay = bond->params.downdelay;
1891                         if (slave->delay) {
1892                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
1893                                         bond->dev->name,
1894                                         (bond->params.mode ==
1895                                          BOND_MODE_ACTIVEBACKUP) ?
1896                                         (bond_is_active_slave(slave) ?
1897                                          "active " : "backup ") : "",
1898                                         slave->dev->name,
1899                                         bond->params.downdelay * bond->params.miimon);
1900                         }
1901                         /*FALLTHRU*/
1902                 case BOND_LINK_FAIL:
1903                         if (link_state) {
1904                                 /*
1905                                  * recovered before downdelay expired
1906                                  */
1907                                 slave->link = BOND_LINK_UP;
1908                                 slave->jiffies = jiffies;
1909                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
1910                                         bond->dev->name,
1911                                         (bond->params.downdelay - slave->delay) *
1912                                         bond->params.miimon,
1913                                         slave->dev->name);
1914                                 continue;
1915                         }
1916
1917                         if (slave->delay <= 0) {
1918                                 slave->new_link = BOND_LINK_DOWN;
1919                                 commit++;
1920                                 continue;
1921                         }
1922
1923                         slave->delay--;
1924                         break;
1925
1926                 case BOND_LINK_DOWN:
1927                         if (!link_state)
1928                                 continue;
1929
1930                         slave->link = BOND_LINK_BACK;
1931                         slave->delay = bond->params.updelay;
1932
1933                         if (slave->delay) {
1934                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
1935                                         bond->dev->name, slave->dev->name,
1936                                         ignore_updelay ? 0 :
1937                                         bond->params.updelay *
1938                                         bond->params.miimon);
1939                         }
1940                         /*FALLTHRU*/
1941                 case BOND_LINK_BACK:
1942                         if (!link_state) {
1943                                 slave->link = BOND_LINK_DOWN;
1944                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
1945                                         bond->dev->name,
1946                                         (bond->params.updelay - slave->delay) *
1947                                         bond->params.miimon,
1948                                         slave->dev->name);
1949
1950                                 continue;
1951                         }
1952
1953                         if (ignore_updelay)
1954                                 slave->delay = 0;
1955
1956                         if (slave->delay <= 0) {
1957                                 slave->new_link = BOND_LINK_UP;
1958                                 commit++;
1959                                 ignore_updelay = false;
1960                                 continue;
1961                         }
1962
1963                         slave->delay--;
1964                         break;
1965                 }
1966         }
1967
1968         return commit;
1969 }
1970
1971 static void bond_miimon_commit(struct bonding *bond)
1972 {
1973         struct list_head *iter;
1974         struct slave *slave;
1975
1976         bond_for_each_slave(bond, slave, iter) {
1977                 switch (slave->new_link) {
1978                 case BOND_LINK_NOCHANGE:
1979                         continue;
1980
1981                 case BOND_LINK_UP:
1982                         slave->link = BOND_LINK_UP;
1983                         slave->jiffies = jiffies;
1984
1985                         if (bond->params.mode == BOND_MODE_8023AD) {
1986                                 /* prevent it from being the active one */
1987                                 bond_set_backup_slave(slave);
1988                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1989                                 /* make it immediately active */
1990                                 bond_set_active_slave(slave);
1991                         } else if (slave != bond->primary_slave) {
1992                                 /* prevent it from being the active one */
1993                                 bond_set_backup_slave(slave);
1994                         }
1995
1996                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
1997                                 bond->dev->name, slave->dev->name,
1998                                 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
1999                                 slave->duplex ? "full" : "half");
2000
2001                         /* notify ad that the link status has changed */
2002                         if (bond->params.mode == BOND_MODE_8023AD)
2003                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2004
2005                         if (bond_is_lb(bond))
2006                                 bond_alb_handle_link_change(bond, slave,
2007                                                             BOND_LINK_UP);
2008
2009                         if (!bond->curr_active_slave ||
2010                             (slave == bond->primary_slave))
2011                                 goto do_failover;
2012
2013                         continue;
2014
2015                 case BOND_LINK_DOWN:
2016                         if (slave->link_failure_count < UINT_MAX)
2017                                 slave->link_failure_count++;
2018
2019                         slave->link = BOND_LINK_DOWN;
2020
2021                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2022                             bond->params.mode == BOND_MODE_8023AD)
2023                                 bond_set_slave_inactive_flags(slave,
2024                                                               BOND_SLAVE_NOTIFY_NOW);
2025
2026                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2027                                 bond->dev->name, slave->dev->name);
2028
2029                         if (bond->params.mode == BOND_MODE_8023AD)
2030                                 bond_3ad_handle_link_change(slave,
2031                                                             BOND_LINK_DOWN);
2032
2033                         if (bond_is_lb(bond))
2034                                 bond_alb_handle_link_change(bond, slave,
2035                                                             BOND_LINK_DOWN);
2036
2037                         if (slave == bond->curr_active_slave)
2038                                 goto do_failover;
2039
2040                         continue;
2041
2042                 default:
2043                         pr_err("%s: invalid new link %d on slave %s\n",
2044                                bond->dev->name, slave->new_link,
2045                                slave->dev->name);
2046                         slave->new_link = BOND_LINK_NOCHANGE;
2047
2048                         continue;
2049                 }
2050
2051 do_failover:
2052                 ASSERT_RTNL();
2053                 block_netpoll_tx();
2054                 write_lock_bh(&bond->curr_slave_lock);
2055                 bond_select_active_slave(bond);
2056                 write_unlock_bh(&bond->curr_slave_lock);
2057                 unblock_netpoll_tx();
2058         }
2059
2060         bond_set_carrier(bond);
2061 }
2062
2063 /*
2064  * bond_mii_monitor
2065  *
2066  * Really a wrapper that splits the mii monitor into two phases: an
2067  * inspection, then (if inspection indicates something needs to be done)
2068  * an acquisition of appropriate locks followed by a commit phase to
2069  * implement whatever link state changes are indicated.
2070  */
2071 static void bond_mii_monitor(struct work_struct *work)
2072 {
2073         struct bonding *bond = container_of(work, struct bonding,
2074                                             mii_work.work);
2075         bool should_notify_peers = false;
2076         unsigned long delay;
2077
2078         delay = msecs_to_jiffies(bond->params.miimon);
2079
2080         if (!bond_has_slaves(bond))
2081                 goto re_arm;
2082
2083         rcu_read_lock();
2084
2085         should_notify_peers = bond_should_notify_peers(bond);
2086
2087         if (bond_miimon_inspect(bond)) {
2088                 rcu_read_unlock();
2089
2090                 /* Race avoidance with bond_close cancel of workqueue */
2091                 if (!rtnl_trylock()) {
2092                         delay = 1;
2093                         should_notify_peers = false;
2094                         goto re_arm;
2095                 }
2096
2097                 bond_miimon_commit(bond);
2098
2099                 rtnl_unlock();  /* might sleep, hold no other locks */
2100         } else
2101                 rcu_read_unlock();
2102
2103 re_arm:
2104         if (bond->params.miimon)
2105                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2106
2107         if (should_notify_peers) {
2108                 if (!rtnl_trylock())
2109                         return;
2110                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2111                 rtnl_unlock();
2112         }
2113 }
2114
2115 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2116 {
2117         struct net_device *upper;
2118         struct list_head *iter;
2119         bool ret = false;
2120
2121         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2122                 return true;
2123
2124         rcu_read_lock();
2125         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2126                 if (ip == bond_confirm_addr(upper, 0, ip)) {
2127                         ret = true;
2128                         break;
2129                 }
2130         }
2131         rcu_read_unlock();
2132
2133         return ret;
2134 }
2135
2136 /*
2137  * We go to the (large) trouble of VLAN tagging ARP frames because
2138  * switches in VLAN mode (especially if ports are configured as
2139  * "native" to a VLAN) might not pass non-tagged frames.
2140  */
2141 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2142 {
2143         struct sk_buff *skb;
2144
2145         pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2146                  slave_dev->name, &dest_ip, &src_ip, vlan_id);
2147
2148         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2149                          NULL, slave_dev->dev_addr, NULL);
2150
2151         if (!skb) {
2152                 pr_err("ARP packet allocation failed\n");
2153                 return;
2154         }
2155         if (vlan_id) {
2156                 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2157                 if (!skb) {
2158                         pr_err("failed to insert VLAN tag\n");
2159                         return;
2160                 }
2161         }
2162         arp_xmit(skb);
2163 }
2164
2165
2166 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2167 {
2168         struct net_device *upper, *vlan_upper;
2169         struct list_head *iter, *vlan_iter;
2170         struct rtable *rt;
2171         __be32 *targets = bond->params.arp_targets, addr;
2172         int i, vlan_id;
2173
2174         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2175                 pr_debug("basa: target %pI4\n", &targets[i]);
2176
2177                 /* Find out through which dev should the packet go */
2178                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2179                                      RTO_ONLINK, 0);
2180                 if (IS_ERR(rt)) {
2181                         pr_debug("%s: no route to arp_ip_target %pI4\n",
2182                                  bond->dev->name, &targets[i]);
2183                         continue;
2184                 }
2185
2186                 vlan_id = 0;
2187
2188                 /* bond device itself */
2189                 if (rt->dst.dev == bond->dev)
2190                         goto found;
2191
2192                 rcu_read_lock();
2193                 /* first we search only for vlan devices. for every vlan
2194                  * found we verify its upper dev list, searching for the
2195                  * rt->dst.dev. If found we save the tag of the vlan and
2196                  * proceed to send the packet.
2197                  *
2198                  * TODO: QinQ?
2199                  */
2200                 netdev_for_each_all_upper_dev_rcu(bond->dev, vlan_upper,
2201                                                   vlan_iter) {
2202                         if (!is_vlan_dev(vlan_upper))
2203                                 continue;
2204                         netdev_for_each_all_upper_dev_rcu(vlan_upper, upper,
2205                                                           iter) {
2206                                 if (upper == rt->dst.dev) {
2207                                         vlan_id = vlan_dev_vlan_id(vlan_upper);
2208                                         rcu_read_unlock();
2209                                         goto found;
2210                                 }
2211                         }
2212                 }
2213
2214                 /* if the device we're looking for is not on top of any of
2215                  * our upper vlans, then just search for any dev that
2216                  * matches, and in case it's a vlan - save the id
2217                  */
2218                 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2219                         if (upper == rt->dst.dev) {
2220                                 /* if it's a vlan - get its VID */
2221                                 if (is_vlan_dev(upper))
2222                                         vlan_id = vlan_dev_vlan_id(upper);
2223
2224                                 rcu_read_unlock();
2225                                 goto found;
2226                         }
2227                 }
2228                 rcu_read_unlock();
2229
2230                 /* Not our device - skip */
2231                 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2232                          bond->dev->name, &targets[i],
2233                          rt->dst.dev ? rt->dst.dev->name : "NULL");
2234
2235                 ip_rt_put(rt);
2236                 continue;
2237
2238 found:
2239                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2240                 ip_rt_put(rt);
2241                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2242                               addr, vlan_id);
2243         }
2244 }
2245
2246 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2247 {
2248         int i;
2249
2250         if (!sip || !bond_has_this_ip(bond, tip)) {
2251                 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2252                 return;
2253         }
2254
2255         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2256         if (i == -1) {
2257                 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2258                 return;
2259         }
2260         slave->last_arp_rx = jiffies;
2261         slave->target_last_arp_rx[i] = jiffies;
2262 }
2263
2264 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2265                  struct slave *slave)
2266 {
2267         struct arphdr *arp = (struct arphdr *)skb->data;
2268         unsigned char *arp_ptr;
2269         __be32 sip, tip;
2270         int alen;
2271
2272         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2273                 return RX_HANDLER_ANOTHER;
2274
2275         read_lock(&bond->lock);
2276
2277         if (!slave_do_arp_validate(bond, slave))
2278                 goto out_unlock;
2279
2280         alen = arp_hdr_len(bond->dev);
2281
2282         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2283                  bond->dev->name, skb->dev->name);
2284
2285         if (alen > skb_headlen(skb)) {
2286                 arp = kmalloc(alen, GFP_ATOMIC);
2287                 if (!arp)
2288                         goto out_unlock;
2289                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2290                         goto out_unlock;
2291         }
2292
2293         if (arp->ar_hln != bond->dev->addr_len ||
2294             skb->pkt_type == PACKET_OTHERHOST ||
2295             skb->pkt_type == PACKET_LOOPBACK ||
2296             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2297             arp->ar_pro != htons(ETH_P_IP) ||
2298             arp->ar_pln != 4)
2299                 goto out_unlock;
2300
2301         arp_ptr = (unsigned char *)(arp + 1);
2302         arp_ptr += bond->dev->addr_len;
2303         memcpy(&sip, arp_ptr, 4);
2304         arp_ptr += 4 + bond->dev->addr_len;
2305         memcpy(&tip, arp_ptr, 4);
2306
2307         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2308                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2309                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2310                  &sip, &tip);
2311
2312         /*
2313          * Backup slaves won't see the ARP reply, but do come through
2314          * here for each ARP probe (so we swap the sip/tip to validate
2315          * the probe).  In a "redundant switch, common router" type of
2316          * configuration, the ARP probe will (hopefully) travel from
2317          * the active, through one switch, the router, then the other
2318          * switch before reaching the backup.
2319          *
2320          * We 'trust' the arp requests if there is an active slave and
2321          * it received valid arp reply(s) after it became active. This
2322          * is done to avoid endless looping when we can't reach the
2323          * arp_ip_target and fool ourselves with our own arp requests.
2324          */
2325         if (bond_is_active_slave(slave))
2326                 bond_validate_arp(bond, slave, sip, tip);
2327         else if (bond->curr_active_slave &&
2328                  time_after(slave_last_rx(bond, bond->curr_active_slave),
2329                             bond->curr_active_slave->jiffies))
2330                 bond_validate_arp(bond, slave, tip, sip);
2331
2332 out_unlock:
2333         read_unlock(&bond->lock);
2334         if (arp != (struct arphdr *)skb->data)
2335                 kfree(arp);
2336         return RX_HANDLER_ANOTHER;
2337 }
2338
2339 /* function to verify if we're in the arp_interval timeslice, returns true if
2340  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2341  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2342  */
2343 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2344                                   int mod)
2345 {
2346         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2347
2348         return time_in_range(jiffies,
2349                              last_act - delta_in_ticks,
2350                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2351 }
2352
2353 /*
2354  * this function is called regularly to monitor each slave's link
2355  * ensuring that traffic is being sent and received when arp monitoring
2356  * is used in load-balancing mode. if the adapter has been dormant, then an
2357  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2358  * arp monitoring in active backup mode.
2359  */
2360 static void bond_loadbalance_arp_mon(struct work_struct *work)
2361 {
2362         struct bonding *bond = container_of(work, struct bonding,
2363                                             arp_work.work);
2364         struct slave *slave, *oldcurrent;
2365         struct list_head *iter;
2366         int do_failover = 0, slave_state_changed = 0;
2367
2368         if (!bond_has_slaves(bond))
2369                 goto re_arm;
2370
2371         rcu_read_lock();
2372
2373         oldcurrent = ACCESS_ONCE(bond->curr_active_slave);
2374         /* see if any of the previous devices are up now (i.e. they have
2375          * xmt and rcv traffic). the curr_active_slave does not come into
2376          * the picture unless it is null. also, slave->jiffies is not needed
2377          * here because we send an arp on each slave and give a slave as
2378          * long as it needs to get the tx/rx within the delta.
2379          * TODO: what about up/down delay in arp mode? it wasn't here before
2380          *       so it can wait
2381          */
2382         bond_for_each_slave_rcu(bond, slave, iter) {
2383                 unsigned long trans_start = dev_trans_start(slave->dev);
2384
2385                 if (slave->link != BOND_LINK_UP) {
2386                         if (bond_time_in_interval(bond, trans_start, 1) &&
2387                             bond_time_in_interval(bond, slave->dev->last_rx, 1)) {
2388
2389                                 slave->link  = BOND_LINK_UP;
2390                                 slave_state_changed = 1;
2391
2392                                 /* primary_slave has no meaning in round-robin
2393                                  * mode. the window of a slave being up and
2394                                  * curr_active_slave being null after enslaving
2395                                  * is closed.
2396                                  */
2397                                 if (!oldcurrent) {
2398                                         pr_info("%s: link status definitely up for interface %s, ",
2399                                                 bond->dev->name,
2400                                                 slave->dev->name);
2401                                         do_failover = 1;
2402                                 } else {
2403                                         pr_info("%s: interface %s is now up\n",
2404                                                 bond->dev->name,
2405                                                 slave->dev->name);
2406                                 }
2407                         }
2408                 } else {
2409                         /* slave->link == BOND_LINK_UP */
2410
2411                         /* not all switches will respond to an arp request
2412                          * when the source ip is 0, so don't take the link down
2413                          * if we don't know our ip yet
2414                          */
2415                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2416                             !bond_time_in_interval(bond, slave->dev->last_rx, 2)) {
2417
2418                                 slave->link  = BOND_LINK_DOWN;
2419                                 slave_state_changed = 1;
2420
2421                                 if (slave->link_failure_count < UINT_MAX)
2422                                         slave->link_failure_count++;
2423
2424                                 pr_info("%s: interface %s is now down.\n",
2425                                         bond->dev->name,
2426                                         slave->dev->name);
2427
2428                                 if (slave == oldcurrent)
2429                                         do_failover = 1;
2430                         }
2431                 }
2432
2433                 /* note: if switch is in round-robin mode, all links
2434                  * must tx arp to ensure all links rx an arp - otherwise
2435                  * links may oscillate or not come up at all; if switch is
2436                  * in something like xor mode, there is nothing we can
2437                  * do - all replies will be rx'ed on same link causing slaves
2438                  * to be unstable during low/no traffic periods
2439                  */
2440                 if (IS_UP(slave->dev))
2441                         bond_arp_send_all(bond, slave);
2442         }
2443
2444         rcu_read_unlock();
2445
2446         if (do_failover || slave_state_changed) {
2447                 if (!rtnl_trylock())
2448                         goto re_arm;
2449
2450                 if (slave_state_changed) {
2451                         bond_slave_state_change(bond);
2452                 } else if (do_failover) {
2453                         /* the bond_select_active_slave must hold RTNL
2454                          * and curr_slave_lock for write.
2455                          */
2456                         block_netpoll_tx();
2457                         write_lock_bh(&bond->curr_slave_lock);
2458
2459                         bond_select_active_slave(bond);
2460
2461                         write_unlock_bh(&bond->curr_slave_lock);
2462                         unblock_netpoll_tx();
2463                 }
2464                 rtnl_unlock();
2465         }
2466
2467 re_arm:
2468         if (bond->params.arp_interval)
2469                 queue_delayed_work(bond->wq, &bond->arp_work,
2470                                    msecs_to_jiffies(bond->params.arp_interval));
2471 }
2472
2473 /*
2474  * Called to inspect slaves for active-backup mode ARP monitor link state
2475  * changes.  Sets new_link in slaves to specify what action should take
2476  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2477  * to link states must be committed.
2478  *
2479  * Called with rcu_read_lock hold.
2480  */
2481 static int bond_ab_arp_inspect(struct bonding *bond)
2482 {
2483         unsigned long trans_start, last_rx;
2484         struct list_head *iter;
2485         struct slave *slave;
2486         int commit = 0;
2487
2488         bond_for_each_slave_rcu(bond, slave, iter) {
2489                 slave->new_link = BOND_LINK_NOCHANGE;
2490                 last_rx = slave_last_rx(bond, slave);
2491
2492                 if (slave->link != BOND_LINK_UP) {
2493                         if (bond_time_in_interval(bond, last_rx, 1)) {
2494                                 slave->new_link = BOND_LINK_UP;
2495                                 commit++;
2496                         }
2497                         continue;
2498                 }
2499
2500                 /*
2501                  * Give slaves 2*delta after being enslaved or made
2502                  * active.  This avoids bouncing, as the last receive
2503                  * times need a full ARP monitor cycle to be updated.
2504                  */
2505                 if (bond_time_in_interval(bond, slave->jiffies, 2))
2506                         continue;
2507
2508                 /*
2509                  * Backup slave is down if:
2510                  * - No current_arp_slave AND
2511                  * - more than 3*delta since last receive AND
2512                  * - the bond has an IP address
2513                  *
2514                  * Note: a non-null current_arp_slave indicates
2515                  * the curr_active_slave went down and we are
2516                  * searching for a new one; under this condition
2517                  * we only take the curr_active_slave down - this
2518                  * gives each slave a chance to tx/rx traffic
2519                  * before being taken out
2520                  */
2521                 if (!bond_is_active_slave(slave) &&
2522                     !bond->current_arp_slave &&
2523                     !bond_time_in_interval(bond, last_rx, 3)) {
2524                         slave->new_link = BOND_LINK_DOWN;
2525                         commit++;
2526                 }
2527
2528                 /*
2529                  * Active slave is down if:
2530                  * - more than 2*delta since transmitting OR
2531                  * - (more than 2*delta since receive AND
2532                  *    the bond has an IP address)
2533                  */
2534                 trans_start = dev_trans_start(slave->dev);
2535                 if (bond_is_active_slave(slave) &&
2536                     (!bond_time_in_interval(bond, trans_start, 2) ||
2537                      !bond_time_in_interval(bond, last_rx, 2))) {
2538                         slave->new_link = BOND_LINK_DOWN;
2539                         commit++;
2540                 }
2541         }
2542
2543         return commit;
2544 }
2545
2546 /*
2547  * Called to commit link state changes noted by inspection step of
2548  * active-backup mode ARP monitor.
2549  *
2550  * Called with RTNL hold.
2551  */
2552 static void bond_ab_arp_commit(struct bonding *bond)
2553 {
2554         unsigned long trans_start;
2555         struct list_head *iter;
2556         struct slave *slave;
2557
2558         bond_for_each_slave(bond, slave, iter) {
2559                 switch (slave->new_link) {
2560                 case BOND_LINK_NOCHANGE:
2561                         continue;
2562
2563                 case BOND_LINK_UP:
2564                         trans_start = dev_trans_start(slave->dev);
2565                         if (bond->curr_active_slave != slave ||
2566                             (!bond->curr_active_slave &&
2567                              bond_time_in_interval(bond, trans_start, 1))) {
2568                                 slave->link = BOND_LINK_UP;
2569                                 if (bond->current_arp_slave) {
2570                                         bond_set_slave_inactive_flags(
2571                                                 bond->current_arp_slave,
2572                                                 BOND_SLAVE_NOTIFY_NOW);
2573                                         bond->current_arp_slave = NULL;
2574                                 }
2575
2576                                 pr_info("%s: link status definitely up for interface %s.\n",
2577                                         bond->dev->name, slave->dev->name);
2578
2579                                 if (!bond->curr_active_slave ||
2580                                     (slave == bond->primary_slave))
2581                                         goto do_failover;
2582
2583                         }
2584
2585                         continue;
2586
2587                 case BOND_LINK_DOWN:
2588                         if (slave->link_failure_count < UINT_MAX)
2589                                 slave->link_failure_count++;
2590
2591                         slave->link = BOND_LINK_DOWN;
2592                         bond_set_slave_inactive_flags(slave,
2593                                                       BOND_SLAVE_NOTIFY_NOW);
2594
2595                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2596                                 bond->dev->name, slave->dev->name);
2597
2598                         if (slave == bond->curr_active_slave) {
2599                                 bond->current_arp_slave = NULL;
2600                                 goto do_failover;
2601                         }
2602
2603                         continue;
2604
2605                 default:
2606                         pr_err("%s: impossible: new_link %d on slave %s\n",
2607                                bond->dev->name, slave->new_link,
2608                                slave->dev->name);
2609                         continue;
2610                 }
2611
2612 do_failover:
2613                 ASSERT_RTNL();
2614                 block_netpoll_tx();
2615                 write_lock_bh(&bond->curr_slave_lock);
2616                 bond_select_active_slave(bond);
2617                 write_unlock_bh(&bond->curr_slave_lock);
2618                 unblock_netpoll_tx();
2619         }
2620
2621         bond_set_carrier(bond);
2622 }
2623
2624 /*
2625  * Send ARP probes for active-backup mode ARP monitor.
2626  *
2627  * Called with rcu_read_lock hold.
2628  */
2629 static bool bond_ab_arp_probe(struct bonding *bond)
2630 {
2631         struct slave *slave, *before = NULL, *new_slave = NULL,
2632                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2633                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2634         struct list_head *iter;
2635         bool found = false;
2636         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2637
2638         if (curr_arp_slave && curr_active_slave)
2639                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2640                         curr_arp_slave->dev->name,
2641                         curr_active_slave->dev->name);
2642
2643         if (curr_active_slave) {
2644                 bond_arp_send_all(bond, curr_active_slave);
2645                 return should_notify_rtnl;
2646         }
2647
2648         /* if we don't have a curr_active_slave, search for the next available
2649          * backup slave from the current_arp_slave and make it the candidate
2650          * for becoming the curr_active_slave
2651          */
2652
2653         if (!curr_arp_slave) {
2654                 curr_arp_slave = bond_first_slave_rcu(bond);
2655                 if (!curr_arp_slave)
2656                         return should_notify_rtnl;
2657         }
2658
2659         bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2660
2661         bond_for_each_slave_rcu(bond, slave, iter) {
2662                 if (!found && !before && IS_UP(slave->dev))
2663                         before = slave;
2664
2665                 if (found && !new_slave && IS_UP(slave->dev))
2666                         new_slave = slave;
2667                 /* if the link state is up at this point, we
2668                  * mark it down - this can happen if we have
2669                  * simultaneous link failures and
2670                  * reselect_active_interface doesn't make this
2671                  * one the current slave so it is still marked
2672                  * up when it is actually down
2673                  */
2674                 if (!IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
2675                         slave->link = BOND_LINK_DOWN;
2676                         if (slave->link_failure_count < UINT_MAX)
2677                                 slave->link_failure_count++;
2678
2679                         bond_set_slave_inactive_flags(slave,
2680                                                       BOND_SLAVE_NOTIFY_LATER);
2681
2682                         pr_info("%s: backup interface %s is now down.\n",
2683                                 bond->dev->name, slave->dev->name);
2684                 }
2685                 if (slave == curr_arp_slave)
2686                         found = true;
2687         }
2688
2689         if (!new_slave && before)
2690                 new_slave = before;
2691
2692         if (!new_slave)
2693                 goto check_state;
2694
2695         new_slave->link = BOND_LINK_BACK;
2696         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2697         bond_arp_send_all(bond, new_slave);
2698         new_slave->jiffies = jiffies;
2699         rcu_assign_pointer(bond->current_arp_slave, new_slave);
2700
2701 check_state:
2702         bond_for_each_slave_rcu(bond, slave, iter) {
2703                 if (slave->should_notify) {
2704                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2705                         break;
2706                 }
2707         }
2708         return should_notify_rtnl;
2709 }
2710
2711 static void bond_activebackup_arp_mon(struct work_struct *work)
2712 {
2713         struct bonding *bond = container_of(work, struct bonding,
2714                                             arp_work.work);
2715         bool should_notify_peers = false;
2716         bool should_notify_rtnl = false;
2717         int delta_in_ticks;
2718
2719         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2720
2721         if (!bond_has_slaves(bond))
2722                 goto re_arm;
2723
2724         rcu_read_lock();
2725
2726         should_notify_peers = bond_should_notify_peers(bond);
2727
2728         if (bond_ab_arp_inspect(bond)) {
2729                 rcu_read_unlock();
2730
2731                 /* Race avoidance with bond_close flush of workqueue */
2732                 if (!rtnl_trylock()) {
2733                         delta_in_ticks = 1;
2734                         should_notify_peers = false;
2735                         goto re_arm;
2736                 }
2737
2738                 bond_ab_arp_commit(bond);
2739
2740                 rtnl_unlock();
2741                 rcu_read_lock();
2742         }
2743
2744         should_notify_rtnl = bond_ab_arp_probe(bond);
2745         rcu_read_unlock();
2746
2747 re_arm:
2748         if (bond->params.arp_interval)
2749                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2750
2751         if (should_notify_peers || should_notify_rtnl) {
2752                 if (!rtnl_trylock())
2753                         return;
2754
2755                 if (should_notify_peers)
2756                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2757                                                  bond->dev);
2758                 if (should_notify_rtnl)
2759                         bond_slave_state_notify(bond);
2760
2761                 rtnl_unlock();
2762         }
2763 }
2764
2765 /*-------------------------- netdev event handling --------------------------*/
2766
2767 /*
2768  * Change device name
2769  */
2770 static int bond_event_changename(struct bonding *bond)
2771 {
2772         bond_remove_proc_entry(bond);
2773         bond_create_proc_entry(bond);
2774
2775         bond_debug_reregister(bond);
2776
2777         return NOTIFY_DONE;
2778 }
2779
2780 static int bond_master_netdev_event(unsigned long event,
2781                                     struct net_device *bond_dev)
2782 {
2783         struct bonding *event_bond = netdev_priv(bond_dev);
2784
2785         switch (event) {
2786         case NETDEV_CHANGENAME:
2787                 return bond_event_changename(event_bond);
2788         case NETDEV_UNREGISTER:
2789                 bond_remove_proc_entry(event_bond);
2790                 break;
2791         case NETDEV_REGISTER:
2792                 bond_create_proc_entry(event_bond);
2793                 break;
2794         case NETDEV_NOTIFY_PEERS:
2795                 if (event_bond->send_peer_notif)
2796                         event_bond->send_peer_notif--;
2797                 break;
2798         default:
2799                 break;
2800         }
2801
2802         return NOTIFY_DONE;
2803 }
2804
2805 static int bond_slave_netdev_event(unsigned long event,
2806                                    struct net_device *slave_dev)
2807 {
2808         struct slave *slave = bond_slave_get_rtnl(slave_dev);
2809         struct bonding *bond;
2810         struct net_device *bond_dev;
2811         u32 old_speed;
2812         u8 old_duplex;
2813
2814         /* A netdev event can be generated while enslaving a device
2815          * before netdev_rx_handler_register is called in which case
2816          * slave will be NULL
2817          */
2818         if (!slave)
2819                 return NOTIFY_DONE;
2820         bond_dev = slave->bond->dev;
2821         bond = slave->bond;
2822
2823         switch (event) {
2824         case NETDEV_UNREGISTER:
2825                 if (bond_dev->type != ARPHRD_ETHER)
2826                         bond_release_and_destroy(bond_dev, slave_dev);
2827                 else
2828                         bond_release(bond_dev, slave_dev);
2829                 break;
2830         case NETDEV_UP:
2831         case NETDEV_CHANGE:
2832                 old_speed = slave->speed;
2833                 old_duplex = slave->duplex;
2834
2835                 bond_update_speed_duplex(slave);
2836
2837                 if (bond->params.mode == BOND_MODE_8023AD) {
2838                         if (old_speed != slave->speed)
2839                                 bond_3ad_adapter_speed_changed(slave);
2840                         if (old_duplex != slave->duplex)
2841                                 bond_3ad_adapter_duplex_changed(slave);
2842                 }
2843                 break;
2844         case NETDEV_DOWN:
2845                 /*
2846                  * ... Or is it this?
2847                  */
2848                 break;
2849         case NETDEV_CHANGEMTU:
2850                 /*
2851                  * TODO: Should slaves be allowed to
2852                  * independently alter their MTU?  For
2853                  * an active-backup bond, slaves need
2854                  * not be the same type of device, so
2855                  * MTUs may vary.  For other modes,
2856                  * slaves arguably should have the
2857                  * same MTUs. To do this, we'd need to
2858                  * take over the slave's change_mtu
2859                  * function for the duration of their
2860                  * servitude.
2861                  */
2862                 break;
2863         case NETDEV_CHANGENAME:
2864                 /* we don't care if we don't have primary set */
2865                 if (!USES_PRIMARY(bond->params.mode) ||
2866                     !bond->params.primary[0])
2867                         break;
2868
2869                 if (slave == bond->primary_slave) {
2870                         /* slave's name changed - he's no longer primary */
2871                         bond->primary_slave = NULL;
2872                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2873                         /* we have a new primary slave */
2874                         bond->primary_slave = slave;
2875                 } else { /* we didn't change primary - exit */
2876                         break;
2877                 }
2878
2879                 pr_info("%s: Primary slave changed to %s, reselecting active slave.\n",
2880                         bond->dev->name, bond->primary_slave ? slave_dev->name :
2881                                                                "none");
2882
2883                 block_netpoll_tx();
2884                 write_lock_bh(&bond->curr_slave_lock);
2885                 bond_select_active_slave(bond);
2886                 write_unlock_bh(&bond->curr_slave_lock);
2887                 unblock_netpoll_tx();
2888                 break;
2889         case NETDEV_FEAT_CHANGE:
2890                 bond_compute_features(bond);
2891                 break;
2892         case NETDEV_RESEND_IGMP:
2893                 /* Propagate to master device */
2894                 call_netdevice_notifiers(event, slave->bond->dev);
2895                 break;
2896         default:
2897                 break;
2898         }
2899
2900         return NOTIFY_DONE;
2901 }
2902
2903 /*
2904  * bond_netdev_event: handle netdev notifier chain events.
2905  *
2906  * This function receives events for the netdev chain.  The caller (an
2907  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2908  * locks for us to safely manipulate the slave devices (RTNL lock,
2909  * dev_probe_lock).
2910  */
2911 static int bond_netdev_event(struct notifier_block *this,
2912                              unsigned long event, void *ptr)
2913 {
2914         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2915
2916         pr_debug("event_dev: %s, event: %lx\n",
2917                  event_dev ? event_dev->name : "None",
2918                  event);
2919
2920         if (!(event_dev->priv_flags & IFF_BONDING))
2921                 return NOTIFY_DONE;
2922
2923         if (event_dev->flags & IFF_MASTER) {
2924                 pr_debug("IFF_MASTER\n");
2925                 return bond_master_netdev_event(event, event_dev);
2926         }
2927
2928         if (event_dev->flags & IFF_SLAVE) {
2929                 pr_debug("IFF_SLAVE\n");
2930                 return bond_slave_netdev_event(event, event_dev);
2931         }
2932
2933         return NOTIFY_DONE;
2934 }
2935
2936 static struct notifier_block bond_netdev_notifier = {
2937         .notifier_call = bond_netdev_event,
2938 };
2939
2940 /*---------------------------- Hashing Policies -----------------------------*/
2941
2942 /* L2 hash helper */
2943 static inline u32 bond_eth_hash(struct sk_buff *skb)
2944 {
2945         struct ethhdr *data = (struct ethhdr *)skb->data;
2946
2947         if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
2948                 return data->h_dest[5] ^ data->h_source[5];
2949
2950         return 0;
2951 }
2952
2953 /* Extract the appropriate headers based on bond's xmit policy */
2954 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
2955                               struct flow_keys *fk)
2956 {
2957         const struct ipv6hdr *iph6;
2958         const struct iphdr *iph;
2959         int noff, proto = -1;
2960
2961         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
2962                 return skb_flow_dissect(skb, fk);
2963
2964         fk->ports = 0;
2965         noff = skb_network_offset(skb);
2966         if (skb->protocol == htons(ETH_P_IP)) {
2967                 if (!pskb_may_pull(skb, noff + sizeof(*iph)))
2968                         return false;
2969                 iph = ip_hdr(skb);
2970                 fk->src = iph->saddr;
2971                 fk->dst = iph->daddr;
2972                 noff += iph->ihl << 2;
2973                 if (!ip_is_fragment(iph))
2974                         proto = iph->protocol;
2975         } else if (skb->protocol == htons(ETH_P_IPV6)) {
2976                 if (!pskb_may_pull(skb, noff + sizeof(*iph6)))
2977                         return false;
2978                 iph6 = ipv6_hdr(skb);
2979                 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
2980                 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
2981                 noff += sizeof(*iph6);
2982                 proto = iph6->nexthdr;
2983         } else {
2984                 return false;
2985         }
2986         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
2987                 fk->ports = skb_flow_get_ports(skb, noff, proto);
2988
2989         return true;
2990 }
2991
2992 /**
2993  * bond_xmit_hash - generate a hash value based on the xmit policy
2994  * @bond: bonding device
2995  * @skb: buffer to use for headers
2996  * @count: modulo value
2997  *
2998  * This function will extract the necessary headers from the skb buffer and use
2999  * them to generate a hash based on the xmit_policy set in the bonding device
3000  * which will be reduced modulo count before returning.
3001  */
3002 int bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, int count)
3003 {
3004         struct flow_keys flow;
3005         u32 hash;
3006
3007         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3008             !bond_flow_dissect(bond, skb, &flow))
3009                 return bond_eth_hash(skb) % count;
3010
3011         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3012             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3013                 hash = bond_eth_hash(skb);
3014         else
3015                 hash = (__force u32)flow.ports;
3016         hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
3017         hash ^= (hash >> 16);
3018         hash ^= (hash >> 8);
3019
3020         return hash % count;
3021 }
3022
3023 /*-------------------------- Device entry points ----------------------------*/
3024
3025 static void bond_work_init_all(struct bonding *bond)
3026 {
3027         INIT_DELAYED_WORK(&bond->mcast_work,
3028                           bond_resend_igmp_join_requests_delayed);
3029         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3030         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3031         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3032                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3033         else
3034                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3035         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3036 }
3037
3038 static void bond_work_cancel_all(struct bonding *bond)
3039 {
3040         cancel_delayed_work_sync(&bond->mii_work);
3041         cancel_delayed_work_sync(&bond->arp_work);
3042         cancel_delayed_work_sync(&bond->alb_work);
3043         cancel_delayed_work_sync(&bond->ad_work);
3044         cancel_delayed_work_sync(&bond->mcast_work);
3045 }
3046
3047 static int bond_open(struct net_device *bond_dev)
3048 {
3049         struct bonding *bond = netdev_priv(bond_dev);
3050         struct list_head *iter;
3051         struct slave *slave;
3052
3053         /* reset slave->backup and slave->inactive */
3054         read_lock(&bond->lock);
3055         if (bond_has_slaves(bond)) {
3056                 read_lock(&bond->curr_slave_lock);
3057                 bond_for_each_slave(bond, slave, iter) {
3058                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3059                                 && (slave != bond->curr_active_slave)) {
3060                                 bond_set_slave_inactive_flags(slave,
3061                                                               BOND_SLAVE_NOTIFY_NOW);
3062                         } else {
3063                                 bond_set_slave_active_flags(slave,
3064                                                             BOND_SLAVE_NOTIFY_NOW);
3065                         }
3066                 }
3067                 read_unlock(&bond->curr_slave_lock);
3068         }
3069         read_unlock(&bond->lock);
3070
3071         bond_work_init_all(bond);
3072
3073         if (bond_is_lb(bond)) {
3074                 /* bond_alb_initialize must be called before the timer
3075                  * is started.
3076                  */
3077                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3078                         return -ENOMEM;
3079                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3080         }
3081
3082         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3083                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3084
3085         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3086                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3087                 if (bond->params.arp_validate)
3088                         bond->recv_probe = bond_arp_rcv;
3089         }
3090
3091         if (bond->params.mode == BOND_MODE_8023AD) {
3092                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3093                 /* register to receive LACPDUs */
3094                 bond->recv_probe = bond_3ad_lacpdu_recv;
3095                 bond_3ad_initiate_agg_selection(bond, 1);
3096         }
3097
3098         return 0;
3099 }
3100
3101 static int bond_close(struct net_device *bond_dev)
3102 {
3103         struct bonding *bond = netdev_priv(bond_dev);
3104
3105         bond_work_cancel_all(bond);
3106         bond->send_peer_notif = 0;
3107         if (bond_is_lb(bond))
3108                 bond_alb_deinitialize(bond);
3109         bond->recv_probe = NULL;
3110
3111         return 0;
3112 }
3113
3114 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3115                                                 struct rtnl_link_stats64 *stats)
3116 {
3117         struct bonding *bond = netdev_priv(bond_dev);
3118         struct rtnl_link_stats64 temp;
3119         struct list_head *iter;
3120         struct slave *slave;
3121
3122         memset(stats, 0, sizeof(*stats));
3123
3124         read_lock_bh(&bond->lock);
3125         bond_for_each_slave(bond, slave, iter) {
3126                 const struct rtnl_link_stats64 *sstats =
3127                         dev_get_stats(slave->dev, &temp);
3128
3129                 stats->rx_packets += sstats->rx_packets;
3130                 stats->rx_bytes += sstats->rx_bytes;
3131                 stats->rx_errors += sstats->rx_errors;
3132                 stats->rx_dropped += sstats->rx_dropped;
3133
3134                 stats->tx_packets += sstats->tx_packets;
3135                 stats->tx_bytes += sstats->tx_bytes;
3136                 stats->tx_errors += sstats->tx_errors;
3137                 stats->tx_dropped += sstats->tx_dropped;
3138
3139                 stats->multicast += sstats->multicast;
3140                 stats->collisions += sstats->collisions;
3141
3142                 stats->rx_length_errors += sstats->rx_length_errors;
3143                 stats->rx_over_errors += sstats->rx_over_errors;
3144                 stats->rx_crc_errors += sstats->rx_crc_errors;
3145                 stats->rx_frame_errors += sstats->rx_frame_errors;
3146                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3147                 stats->rx_missed_errors += sstats->rx_missed_errors;
3148
3149                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3150                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3151                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3152                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3153                 stats->tx_window_errors += sstats->tx_window_errors;
3154         }
3155         read_unlock_bh(&bond->lock);
3156
3157         return stats;
3158 }
3159
3160 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3161 {
3162         struct bonding *bond = netdev_priv(bond_dev);
3163         struct net_device *slave_dev = NULL;
3164         struct ifbond k_binfo;
3165         struct ifbond __user *u_binfo = NULL;
3166         struct ifslave k_sinfo;
3167         struct ifslave __user *u_sinfo = NULL;
3168         struct mii_ioctl_data *mii = NULL;
3169         struct bond_opt_value newval;
3170         struct net *net;
3171         int res = 0;
3172
3173         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3174
3175         switch (cmd) {
3176         case SIOCGMIIPHY:
3177                 mii = if_mii(ifr);
3178                 if (!mii)
3179                         return -EINVAL;
3180
3181                 mii->phy_id = 0;
3182                 /* Fall Through */
3183         case SIOCGMIIREG:
3184                 /*
3185                  * We do this again just in case we were called by SIOCGMIIREG
3186                  * instead of SIOCGMIIPHY.
3187                  */
3188                 mii = if_mii(ifr);
3189                 if (!mii)
3190                         return -EINVAL;
3191
3192
3193                 if (mii->reg_num == 1) {
3194                         mii->val_out = 0;
3195                         read_lock(&bond->lock);
3196                         read_lock(&bond->curr_slave_lock);
3197                         if (netif_carrier_ok(bond->dev))
3198                                 mii->val_out = BMSR_LSTATUS;
3199
3200                         read_unlock(&bond->curr_slave_lock);
3201                         read_unlock(&bond->lock);
3202                 }
3203
3204                 return 0;
3205         case BOND_INFO_QUERY_OLD:
3206         case SIOCBONDINFOQUERY:
3207                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3208
3209                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3210                         return -EFAULT;
3211
3212                 res = bond_info_query(bond_dev, &k_binfo);
3213                 if (res == 0 &&
3214                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3215                         return -EFAULT;
3216
3217                 return res;
3218         case BOND_SLAVE_INFO_QUERY_OLD:
3219         case SIOCBONDSLAVEINFOQUERY:
3220                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3221
3222                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3223                         return -EFAULT;
3224
3225                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3226                 if (res == 0 &&
3227                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3228                         return -EFAULT;
3229
3230                 return res;
3231         default:
3232                 /* Go on */
3233                 break;
3234         }
3235
3236         net = dev_net(bond_dev);
3237
3238         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3239                 return -EPERM;
3240
3241         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3242
3243         pr_debug("slave_dev=%p:\n", slave_dev);
3244
3245         if (!slave_dev)
3246                 return -ENODEV;
3247
3248         pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3249         switch (cmd) {
3250         case BOND_ENSLAVE_OLD:
3251         case SIOCBONDENSLAVE:
3252                 res = bond_enslave(bond_dev, slave_dev);
3253                 break;
3254         case BOND_RELEASE_OLD:
3255         case SIOCBONDRELEASE:
3256                 res = bond_release(bond_dev, slave_dev);
3257                 break;
3258         case BOND_SETHWADDR_OLD:
3259         case SIOCBONDSETHWADDR:
3260                 bond_set_dev_addr(bond_dev, slave_dev);
3261                 res = 0;
3262                 break;
3263         case BOND_CHANGE_ACTIVE_OLD:
3264         case SIOCBONDCHANGEACTIVE:
3265                 bond_opt_initstr(&newval, slave_dev->name);
3266                 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3267                 break;
3268         default:
3269                 res = -EOPNOTSUPP;
3270         }
3271
3272         return res;
3273 }
3274
3275 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3276 {
3277         struct bonding *bond = netdev_priv(bond_dev);
3278
3279         if (change & IFF_PROMISC)
3280                 bond_set_promiscuity(bond,
3281                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3282
3283         if (change & IFF_ALLMULTI)
3284                 bond_set_allmulti(bond,
3285                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3286 }
3287
3288 static void bond_set_rx_mode(struct net_device *bond_dev)
3289 {
3290         struct bonding *bond = netdev_priv(bond_dev);
3291         struct list_head *iter;
3292         struct slave *slave;
3293
3294
3295         rcu_read_lock();
3296         if (USES_PRIMARY(bond->params.mode)) {
3297                 slave = rcu_dereference(bond->curr_active_slave);
3298                 if (slave) {
3299                         dev_uc_sync(slave->dev, bond_dev);
3300                         dev_mc_sync(slave->dev, bond_dev);
3301                 }
3302         } else {
3303                 bond_for_each_slave_rcu(bond, slave, iter) {
3304                         dev_uc_sync_multiple(slave->dev, bond_dev);
3305                         dev_mc_sync_multiple(slave->dev, bond_dev);
3306                 }
3307         }
3308         rcu_read_unlock();
3309 }
3310
3311 static int bond_neigh_init(struct neighbour *n)
3312 {
3313         struct bonding *bond = netdev_priv(n->dev);
3314         const struct net_device_ops *slave_ops;
3315         struct neigh_parms parms;
3316         struct slave *slave;
3317         int ret;
3318
3319         slave = bond_first_slave(bond);
3320         if (!slave)
3321                 return 0;
3322         slave_ops = slave->dev->netdev_ops;
3323         if (!slave_ops->ndo_neigh_setup)
3324                 return 0;
3325
3326         parms.neigh_setup = NULL;
3327         parms.neigh_cleanup = NULL;
3328         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3329         if (ret)
3330                 return ret;
3331
3332         /*
3333          * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3334          * after the last slave has been detached.  Assumes that all slaves
3335          * utilize the same neigh_cleanup (true at this writing as only user
3336          * is ipoib).
3337          */
3338         n->parms->neigh_cleanup = parms.neigh_cleanup;
3339
3340         if (!parms.neigh_setup)
3341                 return 0;
3342
3343         return parms.neigh_setup(n);
3344 }
3345
3346 /*
3347  * The bonding ndo_neigh_setup is called at init time beofre any
3348  * slave exists. So we must declare proxy setup function which will
3349  * be used at run time to resolve the actual slave neigh param setup.
3350  *
3351  * It's also called by master devices (such as vlans) to setup their
3352  * underlying devices. In that case - do nothing, we're already set up from
3353  * our init.
3354  */
3355 static int bond_neigh_setup(struct net_device *dev,
3356                             struct neigh_parms *parms)
3357 {
3358         /* modify only our neigh_parms */
3359         if (parms->dev == dev)
3360                 parms->neigh_setup = bond_neigh_init;
3361
3362         return 0;
3363 }
3364
3365 /*
3366  * Change the MTU of all of a master's slaves to match the master
3367  */
3368 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3369 {
3370         struct bonding *bond = netdev_priv(bond_dev);
3371         struct slave *slave, *rollback_slave;
3372         struct list_head *iter;
3373         int res = 0;
3374
3375         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3376                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3377
3378         /* Can't hold bond->lock with bh disabled here since
3379          * some base drivers panic. On the other hand we can't
3380          * hold bond->lock without bh disabled because we'll
3381          * deadlock. The only solution is to rely on the fact
3382          * that we're under rtnl_lock here, and the slaves
3383          * list won't change. This doesn't solve the problem
3384          * of setting the slave's MTU while it is
3385          * transmitting, but the assumption is that the base
3386          * driver can handle that.
3387          *
3388          * TODO: figure out a way to safely iterate the slaves
3389          * list, but without holding a lock around the actual
3390          * call to the base driver.
3391          */
3392
3393         bond_for_each_slave(bond, slave, iter) {
3394                 pr_debug("s %p c_m %p\n",
3395                          slave,
3396                          slave->dev->netdev_ops->ndo_change_mtu);
3397
3398                 res = dev_set_mtu(slave->dev, new_mtu);
3399
3400                 if (res) {
3401                         /* If we failed to set the slave's mtu to the new value
3402                          * we must abort the operation even in ACTIVE_BACKUP
3403                          * mode, because if we allow the backup slaves to have
3404                          * different mtu values than the active slave we'll
3405                          * need to change their mtu when doing a failover. That
3406                          * means changing their mtu from timer context, which
3407                          * is probably not a good idea.
3408                          */
3409                         pr_debug("err %d %s\n", res, slave->dev->name);
3410                         goto unwind;
3411                 }
3412         }
3413
3414         bond_dev->mtu = new_mtu;
3415
3416         return 0;
3417
3418 unwind:
3419         /* unwind from head to the slave that failed */
3420         bond_for_each_slave(bond, rollback_slave, iter) {
3421                 int tmp_res;
3422
3423                 if (rollback_slave == slave)
3424                         break;
3425
3426                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3427                 if (tmp_res) {
3428                         pr_debug("unwind err %d dev %s\n",
3429                                  tmp_res, rollback_slave->dev->name);
3430                 }
3431         }
3432
3433         return res;
3434 }
3435
3436 /*
3437  * Change HW address
3438  *
3439  * Note that many devices must be down to change the HW address, and
3440  * downing the master releases all slaves.  We can make bonds full of
3441  * bonding devices to test this, however.
3442  */
3443 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3444 {
3445         struct bonding *bond = netdev_priv(bond_dev);
3446         struct slave *slave, *rollback_slave;
3447         struct sockaddr *sa = addr, tmp_sa;
3448         struct list_head *iter;
3449         int res = 0;
3450
3451         if (bond->params.mode == BOND_MODE_ALB)
3452                 return bond_alb_set_mac_address(bond_dev, addr);
3453
3454
3455         pr_debug("bond=%p, name=%s\n",
3456                  bond, bond_dev ? bond_dev->name : "None");
3457
3458         /* If fail_over_mac is enabled, do nothing and return success.
3459          * Returning an error causes ifenslave to fail.
3460          */
3461         if (bond->params.fail_over_mac &&
3462             bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3463                 return 0;
3464
3465         if (!is_valid_ether_addr(sa->sa_data))
3466                 return -EADDRNOTAVAIL;
3467
3468         /* Can't hold bond->lock with bh disabled here since
3469          * some base drivers panic. On the other hand we can't
3470          * hold bond->lock without bh disabled because we'll
3471          * deadlock. The only solution is to rely on the fact
3472          * that we're under rtnl_lock here, and the slaves
3473          * list won't change. This doesn't solve the problem
3474          * of setting the slave's hw address while it is
3475          * transmitting, but the assumption is that the base
3476          * driver can handle that.
3477          *
3478          * TODO: figure out a way to safely iterate the slaves
3479          * list, but without holding a lock around the actual
3480          * call to the base driver.
3481          */
3482
3483         bond_for_each_slave(bond, slave, iter) {
3484                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3485                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3486
3487                 if (slave_ops->ndo_set_mac_address == NULL) {
3488                         res = -EOPNOTSUPP;
3489                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3490                         goto unwind;
3491                 }
3492
3493                 res = dev_set_mac_address(slave->dev, addr);
3494                 if (res) {
3495                         /* TODO: consider downing the slave
3496                          * and retry ?
3497                          * User should expect communications
3498                          * breakage anyway until ARP finish
3499                          * updating, so...
3500                          */
3501                         pr_debug("err %d %s\n", res, slave->dev->name);
3502                         goto unwind;
3503                 }
3504         }
3505
3506         /* success */
3507         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3508         return 0;
3509
3510 unwind:
3511         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3512         tmp_sa.sa_family = bond_dev->type;
3513
3514         /* unwind from head to the slave that failed */
3515         bond_for_each_slave(bond, rollback_slave, iter) {
3516                 int tmp_res;
3517
3518                 if (rollback_slave == slave)
3519                         break;
3520
3521                 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3522                 if (tmp_res) {
3523                         pr_debug("unwind err %d dev %s\n",
3524                                  tmp_res, rollback_slave->dev->name);
3525                 }
3526         }
3527
3528         return res;
3529 }
3530
3531 /**
3532  * bond_xmit_slave_id - transmit skb through slave with slave_id
3533  * @bond: bonding device that is transmitting
3534  * @skb: buffer to transmit
3535  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3536  *
3537  * This function tries to transmit through slave with slave_id but in case
3538  * it fails, it tries to find the first available slave for transmission.
3539  * The skb is consumed in all cases, thus the function is void.
3540  */
3541 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3542 {
3543         struct list_head *iter;
3544         struct slave *slave;
3545         int i = slave_id;
3546
3547         /* Here we start from the slave with slave_id */
3548         bond_for_each_slave_rcu(bond, slave, iter) {
3549                 if (--i < 0) {
3550                         if (slave_can_tx(slave)) {
3551                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3552                                 return;
3553                         }
3554                 }
3555         }
3556
3557         /* Here we start from the first slave up to slave_id */
3558         i = slave_id;
3559         bond_for_each_slave_rcu(bond, slave, iter) {
3560                 if (--i < 0)
3561                         break;
3562                 if (slave_can_tx(slave)) {
3563                         bond_dev_queue_xmit(bond, skb, slave->dev);
3564                         return;
3565                 }
3566         }
3567         /* no slave that can tx has been found */
3568         kfree_skb(skb);
3569 }
3570
3571 /**
3572  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3573  * @bond: bonding device to use
3574  *
3575  * Based on the value of the bonding device's packets_per_slave parameter
3576  * this function generates a slave id, which is usually used as the next
3577  * slave to transmit through.
3578  */
3579 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3580 {
3581         u32 slave_id;
3582         struct reciprocal_value reciprocal_packets_per_slave;
3583         int packets_per_slave = bond->params.packets_per_slave;
3584
3585         switch (packets_per_slave) {
3586         case 0:
3587                 slave_id = prandom_u32();
3588                 break;
3589         case 1:
3590                 slave_id = bond->rr_tx_counter;
3591                 break;
3592         default:
3593                 reciprocal_packets_per_slave =
3594                         bond->params.reciprocal_packets_per_slave;
3595                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3596                                              reciprocal_packets_per_slave);
3597                 break;
3598         }
3599         bond->rr_tx_counter++;
3600
3601         return slave_id;
3602 }
3603
3604 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3605 {
3606         struct bonding *bond = netdev_priv(bond_dev);
3607         struct iphdr *iph = ip_hdr(skb);
3608         struct slave *slave;
3609         u32 slave_id;
3610
3611         /* Start with the curr_active_slave that joined the bond as the
3612          * default for sending IGMP traffic.  For failover purposes one
3613          * needs to maintain some consistency for the interface that will
3614          * send the join/membership reports.  The curr_active_slave found
3615          * will send all of this type of traffic.
3616          */
3617         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3618                 slave = rcu_dereference(bond->curr_active_slave);
3619                 if (slave && slave_can_tx(slave))
3620                         bond_dev_queue_xmit(bond, skb, slave->dev);
3621                 else
3622                         bond_xmit_slave_id(bond, skb, 0);
3623         } else {
3624                 slave_id = bond_rr_gen_slave_id(bond);
3625                 bond_xmit_slave_id(bond, skb, slave_id % bond->slave_cnt);
3626         }
3627
3628         return NETDEV_TX_OK;
3629 }
3630
3631 /*
3632  * in active-backup mode, we know that bond->curr_active_slave is always valid if
3633  * the bond has a usable interface.
3634  */
3635 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3636 {
3637         struct bonding *bond = netdev_priv(bond_dev);
3638         struct slave *slave;
3639
3640         slave = rcu_dereference(bond->curr_active_slave);
3641         if (slave)
3642                 bond_dev_queue_xmit(bond, skb, slave->dev);
3643         else
3644                 kfree_skb(skb);
3645
3646         return NETDEV_TX_OK;
3647 }
3648
3649 /* In bond_xmit_xor() , we determine the output device by using a pre-
3650  * determined xmit_hash_policy(), If the selected device is not enabled,
3651  * find the next active slave.
3652  */
3653 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3654 {
3655         struct bonding *bond = netdev_priv(bond_dev);
3656
3657         bond_xmit_slave_id(bond, skb, bond_xmit_hash(bond, skb, bond->slave_cnt));
3658
3659         return NETDEV_TX_OK;
3660 }
3661
3662 /* in broadcast mode, we send everything to all usable interfaces. */
3663 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3664 {
3665         struct bonding *bond = netdev_priv(bond_dev);
3666         struct slave *slave = NULL;
3667         struct list_head *iter;
3668
3669         bond_for_each_slave_rcu(bond, slave, iter) {
3670                 if (bond_is_last_slave(bond, slave))
3671                         break;
3672                 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3673                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3674
3675                         if (!skb2) {
3676                                 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3677                                        bond_dev->name);
3678                                 continue;
3679                         }
3680                         /* bond_dev_queue_xmit always returns 0 */
3681                         bond_dev_queue_xmit(bond, skb2, slave->dev);
3682                 }
3683         }
3684         if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3685                 bond_dev_queue_xmit(bond, skb, slave->dev);
3686         else
3687                 kfree_skb(skb);
3688
3689         return NETDEV_TX_OK;
3690 }
3691
3692 /*------------------------- Device initialization ---------------------------*/
3693
3694 /*
3695  * Lookup the slave that corresponds to a qid
3696  */
3697 static inline int bond_slave_override(struct bonding *bond,
3698                                       struct sk_buff *skb)
3699 {
3700         struct slave *slave = NULL;
3701         struct list_head *iter;
3702
3703         if (!skb->queue_mapping)
3704                 return 1;
3705
3706         /* Find out if any slaves have the same mapping as this skb. */
3707         bond_for_each_slave_rcu(bond, slave, iter) {
3708                 if (slave->queue_id == skb->queue_mapping) {
3709                         if (slave_can_tx(slave)) {
3710                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3711                                 return 0;
3712                         }
3713                         /* If the slave isn't UP, use default transmit policy. */
3714                         break;
3715                 }
3716         }
3717
3718         return 1;
3719 }
3720
3721
3722 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3723                              void *accel_priv, select_queue_fallback_t fallback)
3724 {
3725         /*
3726          * This helper function exists to help dev_pick_tx get the correct
3727          * destination queue.  Using a helper function skips a call to
3728          * skb_tx_hash and will put the skbs in the queue we expect on their
3729          * way down to the bonding driver.
3730          */
3731         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3732
3733         /*
3734          * Save the original txq to restore before passing to the driver
3735          */
3736         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3737
3738         if (unlikely(txq >= dev->real_num_tx_queues)) {
3739                 do {
3740                         txq -= dev->real_num_tx_queues;
3741                 } while (txq >= dev->real_num_tx_queues);
3742         }
3743         return txq;
3744 }
3745
3746 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3747 {
3748         struct bonding *bond = netdev_priv(dev);
3749
3750         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
3751                 if (!bond_slave_override(bond, skb))
3752                         return NETDEV_TX_OK;
3753         }
3754
3755         switch (bond->params.mode) {
3756         case BOND_MODE_ROUNDROBIN:
3757                 return bond_xmit_roundrobin(skb, dev);
3758         case BOND_MODE_ACTIVEBACKUP:
3759                 return bond_xmit_activebackup(skb, dev);
3760         case BOND_MODE_XOR:
3761                 return bond_xmit_xor(skb, dev);
3762         case BOND_MODE_BROADCAST:
3763                 return bond_xmit_broadcast(skb, dev);
3764         case BOND_MODE_8023AD:
3765                 return bond_3ad_xmit_xor(skb, dev);
3766         case BOND_MODE_ALB:
3767         case BOND_MODE_TLB:
3768                 return bond_alb_xmit(skb, dev);
3769         default:
3770                 /* Should never happen, mode already checked */
3771                 pr_err("%s: Error: Unknown bonding mode %d\n",
3772                        dev->name, bond->params.mode);
3773                 WARN_ON_ONCE(1);
3774                 kfree_skb(skb);
3775                 return NETDEV_TX_OK;
3776         }
3777 }
3778
3779 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3780 {
3781         struct bonding *bond = netdev_priv(dev);
3782         netdev_tx_t ret = NETDEV_TX_OK;
3783
3784         /*
3785          * If we risk deadlock from transmitting this in the
3786          * netpoll path, tell netpoll to queue the frame for later tx
3787          */
3788         if (is_netpoll_tx_blocked(dev))
3789                 return NETDEV_TX_BUSY;
3790
3791         rcu_read_lock();
3792         if (bond_has_slaves(bond))
3793                 ret = __bond_start_xmit(skb, dev);
3794         else
3795                 kfree_skb(skb);
3796         rcu_read_unlock();
3797
3798         return ret;
3799 }
3800
3801 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3802                                      struct ethtool_cmd *ecmd)
3803 {
3804         struct bonding *bond = netdev_priv(bond_dev);
3805         unsigned long speed = 0;
3806         struct list_head *iter;
3807         struct slave *slave;
3808
3809         ecmd->duplex = DUPLEX_UNKNOWN;
3810         ecmd->port = PORT_OTHER;
3811
3812         /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3813          * do not need to check mode.  Though link speed might not represent
3814          * the true receive or transmit bandwidth (not all modes are symmetric)
3815          * this is an accurate maximum.
3816          */
3817         read_lock(&bond->lock);
3818         bond_for_each_slave(bond, slave, iter) {
3819                 if (SLAVE_IS_OK(slave)) {
3820                         if (slave->speed != SPEED_UNKNOWN)
3821                                 speed += slave->speed;
3822                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
3823                             slave->duplex != DUPLEX_UNKNOWN)
3824                                 ecmd->duplex = slave->duplex;
3825                 }
3826         }
3827         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3828         read_unlock(&bond->lock);
3829
3830         return 0;
3831 }
3832
3833 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3834                                      struct ethtool_drvinfo *drvinfo)
3835 {
3836         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3837         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3838         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3839                  BOND_ABI_VERSION);
3840 }
3841
3842 static const struct ethtool_ops bond_ethtool_ops = {
3843         .get_drvinfo            = bond_ethtool_get_drvinfo,
3844         .get_settings           = bond_ethtool_get_settings,
3845         .get_link               = ethtool_op_get_link,
3846 };
3847
3848 static const struct net_device_ops bond_netdev_ops = {
3849         .ndo_init               = bond_init,
3850         .ndo_uninit             = bond_uninit,
3851         .ndo_open               = bond_open,
3852         .ndo_stop               = bond_close,
3853         .ndo_start_xmit         = bond_start_xmit,
3854         .ndo_select_queue       = bond_select_queue,
3855         .ndo_get_stats64        = bond_get_stats,
3856         .ndo_do_ioctl           = bond_do_ioctl,
3857         .ndo_change_rx_flags    = bond_change_rx_flags,
3858         .ndo_set_rx_mode        = bond_set_rx_mode,
3859         .ndo_change_mtu         = bond_change_mtu,
3860         .ndo_set_mac_address    = bond_set_mac_address,
3861         .ndo_neigh_setup        = bond_neigh_setup,
3862         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
3863         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
3864 #ifdef CONFIG_NET_POLL_CONTROLLER
3865         .ndo_netpoll_setup      = bond_netpoll_setup,
3866         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
3867         .ndo_poll_controller    = bond_poll_controller,
3868 #endif
3869         .ndo_add_slave          = bond_enslave,
3870         .ndo_del_slave          = bond_release,
3871         .ndo_fix_features       = bond_fix_features,
3872 };
3873
3874 static const struct device_type bond_type = {
3875         .name = "bond",
3876 };
3877
3878 static void bond_destructor(struct net_device *bond_dev)
3879 {
3880         struct bonding *bond = netdev_priv(bond_dev);
3881         if (bond->wq)
3882                 destroy_workqueue(bond->wq);
3883         free_netdev(bond_dev);
3884 }
3885
3886 void bond_setup(struct net_device *bond_dev)
3887 {
3888         struct bonding *bond = netdev_priv(bond_dev);
3889
3890         /* initialize rwlocks */
3891         rwlock_init(&bond->lock);
3892         rwlock_init(&bond->curr_slave_lock);
3893         bond->params = bonding_defaults;
3894
3895         /* Initialize pointers */
3896         bond->dev = bond_dev;
3897
3898         /* Initialize the device entry points */
3899         ether_setup(bond_dev);
3900         bond_dev->netdev_ops = &bond_netdev_ops;
3901         bond_dev->ethtool_ops = &bond_ethtool_ops;
3902
3903         bond_dev->destructor = bond_destructor;
3904
3905         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
3906
3907         /* Initialize the device options */
3908         bond_dev->tx_queue_len = 0;
3909         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
3910         bond_dev->priv_flags |= IFF_BONDING;
3911         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
3912
3913         /* At first, we block adding VLANs. That's the only way to
3914          * prevent problems that occur when adding VLANs over an
3915          * empty bond. The block will be removed once non-challenged
3916          * slaves are enslaved.
3917          */
3918         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
3919
3920         /* don't acquire bond device's netif_tx_lock when
3921          * transmitting */
3922         bond_dev->features |= NETIF_F_LLTX;
3923
3924         /* By default, we declare the bond to be fully
3925          * VLAN hardware accelerated capable. Special
3926          * care is taken in the various xmit functions
3927          * when there are slaves that are not hw accel
3928          * capable
3929          */
3930
3931         /* Don't allow bond devices to change network namespaces. */
3932         bond_dev->features |= NETIF_F_NETNS_LOCAL;
3933
3934         bond_dev->hw_features = BOND_VLAN_FEATURES |
3935                                 NETIF_F_HW_VLAN_CTAG_TX |
3936                                 NETIF_F_HW_VLAN_CTAG_RX |
3937                                 NETIF_F_HW_VLAN_CTAG_FILTER;
3938
3939         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
3940         bond_dev->features |= bond_dev->hw_features;
3941 }
3942
3943 /*
3944 * Destroy a bonding device.
3945 * Must be under rtnl_lock when this function is called.
3946 */
3947 static void bond_uninit(struct net_device *bond_dev)
3948 {
3949         struct bonding *bond = netdev_priv(bond_dev);
3950         struct list_head *iter;
3951         struct slave *slave;
3952
3953         bond_netpoll_cleanup(bond_dev);
3954
3955         /* Release the bonded slaves */
3956         bond_for_each_slave(bond, slave, iter)
3957                 __bond_release_one(bond_dev, slave->dev, true);
3958         pr_info("%s: released all slaves\n", bond_dev->name);
3959
3960         list_del(&bond->bond_list);
3961
3962         bond_debug_unregister(bond);
3963 }
3964
3965 /*------------------------- Module initialization ---------------------------*/
3966
3967 int bond_parm_tbl_lookup(int mode, const struct bond_parm_tbl *tbl)
3968 {
3969         int i;
3970
3971         for (i = 0; tbl[i].modename; i++)
3972                 if (mode == tbl[i].mode)
3973                         return tbl[i].mode;
3974
3975         return -1;
3976 }
3977
3978 static int bond_parm_tbl_lookup_name(const char *modename,
3979                                      const struct bond_parm_tbl *tbl)
3980 {
3981         int i;
3982
3983         for (i = 0; tbl[i].modename; i++)
3984                 if (strcmp(modename, tbl[i].modename) == 0)
3985                         return tbl[i].mode;
3986
3987         return -1;
3988 }
3989
3990 /*
3991  * Convert string input module parms.  Accept either the
3992  * number of the mode or its string name.  A bit complicated because
3993  * some mode names are substrings of other names, and calls from sysfs
3994  * may have whitespace in the name (trailing newlines, for example).
3995  */
3996 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
3997 {
3998         int modeint;
3999         char *p, modestr[BOND_MAX_MODENAME_LEN + 1];
4000
4001         for (p = (char *)buf; *p; p++)
4002                 if (!(isdigit(*p) || isspace(*p)))
4003                         break;
4004
4005         if (*p && sscanf(buf, "%20s", modestr) != 0)
4006                 return bond_parm_tbl_lookup_name(modestr, tbl);
4007         else if (sscanf(buf, "%d", &modeint) != 0)
4008                 return bond_parm_tbl_lookup(modeint, tbl);
4009
4010         return -1;
4011 }
4012
4013 static int bond_check_params(struct bond_params *params)
4014 {
4015         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4016         struct bond_opt_value newval, *valptr;
4017         int arp_all_targets_value;
4018
4019         /*
4020          * Convert string parameters.
4021          */
4022         if (mode) {
4023                 bond_opt_initstr(&newval, mode);
4024                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4025                 if (!valptr) {
4026                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4027                         return -EINVAL;
4028                 }
4029                 bond_mode = valptr->value;
4030         }
4031
4032         if (xmit_hash_policy) {
4033                 if ((bond_mode != BOND_MODE_XOR) &&
4034                     (bond_mode != BOND_MODE_8023AD)) {
4035                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4036                                bond_mode_name(bond_mode));
4037                 } else {
4038                         bond_opt_initstr(&newval, xmit_hash_policy);
4039                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4040                                                 &newval);
4041                         if (!valptr) {
4042                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4043                                        xmit_hash_policy);
4044                                 return -EINVAL;
4045                         }
4046                         xmit_hashtype = valptr->value;
4047                 }
4048         }
4049
4050         if (lacp_rate) {
4051                 if (bond_mode != BOND_MODE_8023AD) {
4052                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4053                                 bond_mode_name(bond_mode));
4054                 } else {
4055                         bond_opt_initstr(&newval, lacp_rate);
4056                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4057                                                 &newval);
4058                         if (!valptr) {
4059                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4060                                        lacp_rate);
4061                                 return -EINVAL;
4062                         }
4063                         lacp_fast = valptr->value;
4064                 }
4065         }
4066
4067         if (ad_select) {
4068                 bond_opt_initstr(&newval, lacp_rate);
4069                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4070                                         &newval);
4071                 if (!valptr) {
4072                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4073                         return -EINVAL;
4074                 }
4075                 params->ad_select = valptr->value;
4076                 if (bond_mode != BOND_MODE_8023AD)
4077                         pr_warning("ad_select param only affects 802.3ad mode\n");
4078         } else {
4079                 params->ad_select = BOND_AD_STABLE;
4080         }
4081
4082         if (max_bonds < 0) {
4083                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4084                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4085                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4086         }
4087
4088         if (miimon < 0) {
4089                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4090                            miimon, INT_MAX);
4091                 miimon = 0;
4092         }
4093
4094         if (updelay < 0) {
4095                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4096                            updelay, INT_MAX);
4097                 updelay = 0;
4098         }
4099
4100         if (downdelay < 0) {
4101                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4102                            downdelay, INT_MAX);
4103                 downdelay = 0;
4104         }
4105
4106         if ((use_carrier != 0) && (use_carrier != 1)) {
4107                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4108                            use_carrier);
4109                 use_carrier = 1;
4110         }
4111
4112         if (num_peer_notif < 0 || num_peer_notif > 255) {
4113                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4114                            num_peer_notif);
4115                 num_peer_notif = 1;
4116         }
4117
4118         /* reset values for 802.3ad/TLB/ALB */
4119         if (BOND_NO_USES_ARP(bond_mode)) {
4120                 if (!miimon) {
4121                         pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4122                         pr_warning("Forcing miimon to 100msec\n");
4123                         miimon = BOND_DEFAULT_MIIMON;
4124                 }
4125         }
4126
4127         if (tx_queues < 1 || tx_queues > 255) {
4128                 pr_warning("Warning: tx_queues (%d) should be between "
4129                            "1 and 255, resetting to %d\n",
4130                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4131                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4132         }
4133
4134         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4135                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4136                            "not of valid value (0/1), so it was set to "
4137                            "0\n", all_slaves_active);
4138                 all_slaves_active = 0;
4139         }
4140
4141         if (resend_igmp < 0 || resend_igmp > 255) {
4142                 pr_warning("Warning: resend_igmp (%d) should be between "
4143                            "0 and 255, resetting to %d\n",
4144                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4145                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4146         }
4147
4148         bond_opt_initval(&newval, packets_per_slave);
4149         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4150                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4151                         packets_per_slave, USHRT_MAX);
4152                 packets_per_slave = 1;
4153         }
4154
4155         if (bond_mode == BOND_MODE_ALB) {
4156                 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",
4157                           updelay);
4158         }
4159
4160         if (!miimon) {
4161                 if (updelay || downdelay) {
4162                         /* just warn the user the up/down delay will have
4163                          * no effect since miimon is zero...
4164                          */
4165                         pr_warning("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4166                                    updelay, downdelay);
4167                 }
4168         } else {
4169                 /* don't allow arp monitoring */
4170                 if (arp_interval) {
4171                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4172                                    miimon, arp_interval);
4173                         arp_interval = 0;
4174                 }
4175
4176                 if ((updelay % miimon) != 0) {
4177                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4178                                    updelay, miimon,
4179                                    (updelay / miimon) * miimon);
4180                 }
4181
4182                 updelay /= miimon;
4183
4184                 if ((downdelay % miimon) != 0) {
4185                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4186                                    downdelay, miimon,
4187                                    (downdelay / miimon) * miimon);
4188                 }
4189
4190                 downdelay /= miimon;
4191         }
4192
4193         if (arp_interval < 0) {
4194                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to 0\n",
4195                            arp_interval, INT_MAX);
4196                 arp_interval = 0;
4197         }
4198
4199         for (arp_ip_count = 0, i = 0;
4200              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4201                 /* not complete check, but should be good enough to
4202                    catch mistakes */
4203                 __be32 ip;
4204                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4205                     IS_IP_TARGET_UNUSABLE_ADDRESS(ip)) {
4206                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4207                                    arp_ip_target[i]);
4208                         arp_interval = 0;
4209                 } else {
4210                         if (bond_get_targets_ip(arp_target, ip) == -1)
4211                                 arp_target[arp_ip_count++] = ip;
4212                         else
4213                                 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4214                                            &ip);
4215                 }
4216         }
4217
4218         if (arp_interval && !arp_ip_count) {
4219                 /* don't allow arping if no arp_ip_target given... */
4220                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4221                            arp_interval);
4222                 arp_interval = 0;
4223         }
4224
4225         if (arp_validate) {
4226                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4227                         pr_err("arp_validate only supported in active-backup mode\n");
4228                         return -EINVAL;
4229                 }
4230                 if (!arp_interval) {
4231                         pr_err("arp_validate requires arp_interval\n");
4232                         return -EINVAL;
4233                 }
4234
4235                 bond_opt_initstr(&newval, arp_validate);
4236                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4237                                         &newval);
4238                 if (!valptr) {
4239                         pr_err("Error: invalid arp_validate \"%s\"\n",
4240                                arp_validate);
4241                         return -EINVAL;
4242                 }
4243                 arp_validate_value = valptr->value;
4244         } else {
4245                 arp_validate_value = 0;
4246         }
4247
4248         arp_all_targets_value = 0;
4249         if (arp_all_targets) {
4250                 bond_opt_initstr(&newval, arp_all_targets);
4251                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4252                                         &newval);
4253                 if (!valptr) {
4254                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4255                                arp_all_targets);
4256                         arp_all_targets_value = 0;
4257                 } else {
4258                         arp_all_targets_value = valptr->value;
4259                 }
4260         }
4261
4262         if (miimon) {
4263                 pr_info("MII link monitoring set to %d ms\n", miimon);
4264         } else if (arp_interval) {
4265                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4266                                           arp_validate_value);
4267                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4268                         arp_interval, valptr->string, arp_ip_count);
4269
4270                 for (i = 0; i < arp_ip_count; i++)
4271                         pr_info(" %s", arp_ip_target[i]);
4272
4273                 pr_info("\n");
4274
4275         } else if (max_bonds) {
4276                 /* miimon and arp_interval not set, we need one so things
4277                  * work as expected, see bonding.txt for details
4278                  */
4279                 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");
4280         }
4281
4282         if (primary && !USES_PRIMARY(bond_mode)) {
4283                 /* currently, using a primary only makes sense
4284                  * in active backup, TLB or ALB modes
4285                  */
4286                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4287                            primary, bond_mode_name(bond_mode));
4288                 primary = NULL;
4289         }
4290
4291         if (primary && primary_reselect) {
4292                 bond_opt_initstr(&newval, primary_reselect);
4293                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4294                                         &newval);
4295                 if (!valptr) {
4296                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4297                                primary_reselect);
4298                         return -EINVAL;
4299                 }
4300                 primary_reselect_value = valptr->value;
4301         } else {
4302                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4303         }
4304
4305         if (fail_over_mac) {
4306                 bond_opt_initstr(&newval, fail_over_mac);
4307                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4308                                         &newval);
4309                 if (!valptr) {
4310                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4311                                fail_over_mac);
4312                         return -EINVAL;
4313                 }
4314                 fail_over_mac_value = valptr->value;
4315                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4316                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4317         } else {
4318                 fail_over_mac_value = BOND_FOM_NONE;
4319         }
4320
4321         if (lp_interval == 0) {
4322                 pr_warning("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4323                            INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4324                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4325         }
4326
4327         /* fill params struct with the proper values */
4328         params->mode = bond_mode;
4329         params->xmit_policy = xmit_hashtype;
4330         params->miimon = miimon;
4331         params->num_peer_notif = num_peer_notif;
4332         params->arp_interval = arp_interval;
4333         params->arp_validate = arp_validate_value;
4334         params->arp_all_targets = arp_all_targets_value;
4335         params->updelay = updelay;
4336         params->downdelay = downdelay;
4337         params->use_carrier = use_carrier;
4338         params->lacp_fast = lacp_fast;
4339         params->primary[0] = 0;
4340         params->primary_reselect = primary_reselect_value;
4341         params->fail_over_mac = fail_over_mac_value;
4342         params->tx_queues = tx_queues;
4343         params->all_slaves_active = all_slaves_active;
4344         params->resend_igmp = resend_igmp;
4345         params->min_links = min_links;
4346         params->lp_interval = lp_interval;
4347         params->packets_per_slave = packets_per_slave;
4348         if (packets_per_slave > 0) {
4349                 params->reciprocal_packets_per_slave =
4350                         reciprocal_value(packets_per_slave);
4351         } else {
4352                 /* reciprocal_packets_per_slave is unused if
4353                  * packets_per_slave is 0 or 1, just initialize it
4354                  */
4355                 params->reciprocal_packets_per_slave =
4356                         (struct reciprocal_value) { 0 };
4357         }
4358
4359         if (primary) {
4360                 strncpy(params->primary, primary, IFNAMSIZ);
4361                 params->primary[IFNAMSIZ - 1] = 0;
4362         }
4363
4364         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4365
4366         return 0;
4367 }
4368
4369 static struct lock_class_key bonding_netdev_xmit_lock_key;
4370 static struct lock_class_key bonding_netdev_addr_lock_key;
4371 static struct lock_class_key bonding_tx_busylock_key;
4372
4373 static void bond_set_lockdep_class_one(struct net_device *dev,
4374                                        struct netdev_queue *txq,
4375                                        void *_unused)
4376 {
4377         lockdep_set_class(&txq->_xmit_lock,
4378                           &bonding_netdev_xmit_lock_key);
4379 }
4380
4381 static void bond_set_lockdep_class(struct net_device *dev)
4382 {
4383         lockdep_set_class(&dev->addr_list_lock,
4384                           &bonding_netdev_addr_lock_key);
4385         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4386         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4387 }
4388
4389 /*
4390  * Called from registration process
4391  */
4392 static int bond_init(struct net_device *bond_dev)
4393 {
4394         struct bonding *bond = netdev_priv(bond_dev);
4395         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4396         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4397
4398         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4399
4400         /*
4401          * Initialize locks that may be required during
4402          * en/deslave operations.  All of the bond_open work
4403          * (of which this is part) should really be moved to
4404          * a phase prior to dev_open
4405          */
4406         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4407         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4408
4409         bond->wq = create_singlethread_workqueue(bond_dev->name);
4410         if (!bond->wq)
4411                 return -ENOMEM;
4412
4413         bond_set_lockdep_class(bond_dev);
4414
4415         list_add_tail(&bond->bond_list, &bn->dev_list);
4416
4417         bond_prepare_sysfs_group(bond);
4418
4419         bond_debug_register(bond);
4420
4421         /* Ensure valid dev_addr */
4422         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4423             bond_dev->addr_assign_type == NET_ADDR_PERM)
4424                 eth_hw_addr_random(bond_dev);
4425
4426         return 0;
4427 }
4428
4429 unsigned int bond_get_num_tx_queues(void)
4430 {
4431         return tx_queues;
4432 }
4433
4434 /* Create a new bond based on the specified name and bonding parameters.
4435  * If name is NULL, obtain a suitable "bond%d" name for us.
4436  * Caller must NOT hold rtnl_lock; we need to release it here before we
4437  * set up our sysfs entries.
4438  */
4439 int bond_create(struct net *net, const char *name)
4440 {
4441         struct net_device *bond_dev;
4442         int res;
4443
4444         rtnl_lock();
4445
4446         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4447                                    name ? name : "bond%d",
4448                                    bond_setup, tx_queues);
4449         if (!bond_dev) {
4450                 pr_err("%s: eek! can't alloc netdev!\n", name);
4451                 rtnl_unlock();
4452                 return -ENOMEM;
4453         }
4454
4455         dev_net_set(bond_dev, net);
4456         bond_dev->rtnl_link_ops = &bond_link_ops;
4457
4458         res = register_netdevice(bond_dev);
4459
4460         netif_carrier_off(bond_dev);
4461
4462         rtnl_unlock();
4463         if (res < 0)
4464                 bond_destructor(bond_dev);
4465         return res;
4466 }
4467
4468 static int __net_init bond_net_init(struct net *net)
4469 {
4470         struct bond_net *bn = net_generic(net, bond_net_id);
4471
4472         bn->net = net;
4473         INIT_LIST_HEAD(&bn->dev_list);
4474
4475         bond_create_proc_dir(bn);
4476         bond_create_sysfs(bn);
4477
4478         return 0;
4479 }
4480
4481 static void __net_exit bond_net_exit(struct net *net)
4482 {
4483         struct bond_net *bn = net_generic(net, bond_net_id);
4484         struct bonding *bond, *tmp_bond;
4485         LIST_HEAD(list);
4486
4487         bond_destroy_sysfs(bn);
4488         bond_destroy_proc_dir(bn);
4489
4490         /* Kill off any bonds created after unregistering bond rtnl ops */
4491         rtnl_lock();
4492         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4493                 unregister_netdevice_queue(bond->dev, &list);
4494         unregister_netdevice_many(&list);
4495         rtnl_unlock();
4496 }
4497
4498 static struct pernet_operations bond_net_ops = {
4499         .init = bond_net_init,
4500         .exit = bond_net_exit,
4501         .id   = &bond_net_id,
4502         .size = sizeof(struct bond_net),
4503 };
4504
4505 static int __init bonding_init(void)
4506 {
4507         int i;
4508         int res;
4509
4510         pr_info("%s", bond_version);
4511
4512         res = bond_check_params(&bonding_defaults);
4513         if (res)
4514                 goto out;
4515
4516         res = register_pernet_subsys(&bond_net_ops);
4517         if (res)
4518                 goto out;
4519
4520         res = bond_netlink_init();
4521         if (res)
4522                 goto err_link;
4523
4524         bond_create_debugfs();
4525
4526         for (i = 0; i < max_bonds; i++) {
4527                 res = bond_create(&init_net, NULL);
4528                 if (res)
4529                         goto err;
4530         }
4531
4532         register_netdevice_notifier(&bond_netdev_notifier);
4533 out:
4534         return res;
4535 err:
4536         bond_netlink_fini();
4537 err_link:
4538         unregister_pernet_subsys(&bond_net_ops);
4539         goto out;
4540
4541 }
4542
4543 static void __exit bonding_exit(void)
4544 {
4545         unregister_netdevice_notifier(&bond_netdev_notifier);
4546
4547         bond_destroy_debugfs();
4548
4549         bond_netlink_fini();
4550         unregister_pernet_subsys(&bond_net_ops);
4551
4552 #ifdef CONFIG_NET_POLL_CONTROLLER
4553         /*
4554          * Make sure we don't have an imbalance on our netpoll blocking
4555          */
4556         WARN_ON(atomic_read(&netpoll_block_tx));
4557 #endif
4558 }
4559
4560 module_init(bonding_init);
4561 module_exit(bonding_exit);
4562 MODULE_LICENSE("GPL");
4563 MODULE_VERSION(DRV_VERSION);
4564 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4565 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");