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                 if (new_active)
 834                         bond_set_slave_active_flags(new_active);
 835         } else {
 836                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 837         }
 838
 839         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
 840                 if (old_active)
 841                         bond_set_slave_inactive_flags(old_active);
 842
 843                 if (new_active) {
 844                         bool should_notify_peers = false;
 845
 846                         bond_set_slave_active_flags(new_active);
 847
 848                         if (bond->params.fail_over_mac)
 849                                 bond_do_fail_over_mac(bond, new_active,
 850                                                       old_active);
 851
 852                         if (netif_running(bond->dev)) {
 853                                 bond->send_peer_notif =
 854                                         bond->params.num_peer_notif;
 855                                 should_notify_peers =
 856                                         bond_should_notify_peers(bond);
 857                         }
 858
 859                         write_unlock_bh(&bond->curr_slave_lock);
 860
 861                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 862                         if (should_notify_peers)
 863                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 864                                                          bond->dev);
 865
 866                         write_lock_bh(&bond->curr_slave_lock);
 867                 }
 868         }
 869
 870         /* resend IGMP joins since active slave has changed or
 871          * all were sent on curr_active_slave.
 872          * resend only if bond is brought up with the affected
 873          * bonding modes and the retransmission is enabled */
 874         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 875             ((USES_PRIMARY(bond->params.mode) && new_active) ||
 876              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
 877                 bond->igmp_retrans = bond->params.resend_igmp;
 878                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 879         }
 880 }
 881
 882 /**
 883  * bond_select_active_slave - select a new active slave, if needed
 884  * @bond: our bonding struct
 885  *
 886  * This functions should be called when one of the following occurs:
 887  * - The old curr_active_slave has been released or lost its link.
 888  * - The primary_slave has got its link back.
 889  * - A slave has got its link back and there's no old curr_active_slave.
 890  *
 891  * Caller must hold curr_slave_lock for write_bh.
 892  */
 893 void bond_select_active_slave(struct bonding *bond)
 894 {
 895         struct slave *best_slave;
 896         int rv;
 897
 898         best_slave = bond_find_best_slave(bond);
 899         if (best_slave != bond->curr_active_slave) {
 900                 bond_change_active_slave(bond, best_slave);
 901                 rv = bond_set_carrier(bond);
 902                 if (!rv)
 903                         return;
 904
 905                 if (netif_carrier_ok(bond->dev)) {
 906                         pr_info("%s: first active interface up!\n",
 907                                 bond->dev->name);
 908                 } else {
 909                         pr_info("%s: now running without any active interface !\n",
 910                                 bond->dev->name);
 911                 }
 912         }
 913 }
 914
 915 #ifdef CONFIG_NET_POLL_CONTROLLER
 916 static inline int slave_enable_netpoll(struct slave *slave)
 917 {
 918         struct netpoll *np;
 919         int err = 0;
 920
 921         np = kzalloc(sizeof(*np), GFP_ATOMIC);
 922         err = -ENOMEM;
 923         if (!np)
 924                 goto out;
 925
 926         err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
 927         if (err) {
 928                 kfree(np);
 929                 goto out;
 930         }
 931         slave->np = np;
 932 out:
 933         return err;
 934 }
 935 static inline void slave_disable_netpoll(struct slave *slave)
 936 {
 937         struct netpoll *np = slave->np;
 938
 939         if (!np)
 940                 return;
 941
 942         slave->np = NULL;
 943         __netpoll_free_async(np);
 944 }
 945 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
 946 {
 947         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
 948                 return false;
 949         if (!slave_dev->netdev_ops->ndo_poll_controller)
 950                 return false;
 951         return true;
 952 }
 953
 954 static void bond_poll_controller(struct net_device *bond_dev)
 955 {
 956 }
 957
 958 static void bond_netpoll_cleanup(struct net_device *bond_dev)
 959 {
 960         struct bonding *bond = netdev_priv(bond_dev);
 961         struct list_head *iter;
 962         struct slave *slave;
 963
 964         bond_for_each_slave(bond, slave, iter)
 965                 if (IS_UP(slave->dev))
 966                         slave_disable_netpoll(slave);
 967 }
 968
 969 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
 970 {
 971         struct bonding *bond = netdev_priv(dev);
 972         struct list_head *iter;
 973         struct slave *slave;
 974         int err = 0;
 975
 976         bond_for_each_slave(bond, slave, iter) {
 977                 err = slave_enable_netpoll(slave);
 978                 if (err) {
 979                         bond_netpoll_cleanup(dev);
 980                         break;
 981                 }
 982         }
 983         return err;
 984 }
 985 #else
 986 static inline int slave_enable_netpoll(struct slave *slave)
 987 {
 988         return 0;
 989 }
 990 static inline void slave_disable_netpoll(struct slave *slave)
 991 {
 992 }
 993 static void bond_netpoll_cleanup(struct net_device *bond_dev)
 994 {
 995 }
 996 #endif
 997
 998 /*---------------------------------- IOCTL ----------------------------------*/
 999
1000 static netdev_features_t bond_fix_features(struct net_device *dev,
1001                                            netdev_features_t features)
1002 {
1003         struct bonding *bond = netdev_priv(dev);
1004         struct list_head *iter;
1005         netdev_features_t mask;
1006         struct slave *slave;
1007
1008         if (!bond_has_slaves(bond)) {
1009                 /* Disable adding VLANs to empty bond. But why? --mq */
1010                 features |= NETIF_F_VLAN_CHALLENGED;
1011                 return features;
1012         }
1013
1014         mask = features;
1015         features &= ~NETIF_F_ONE_FOR_ALL;
1016         features |= NETIF_F_ALL_FOR_ALL;
1017
1018         bond_for_each_slave(bond, slave, iter) {
1019                 features = netdev_increment_features(features,
1020                                                      slave->dev->features,
1021                                                      mask);
1022         }
1023         features = netdev_add_tso_features(features, mask);
1024
1025         return features;
1026 }
1027
1028 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1029                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1030                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1031
1032 static void bond_compute_features(struct bonding *bond)
1033 {
1034         unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1035         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1036         struct net_device *bond_dev = bond->dev;
1037         struct list_head *iter;
1038         struct slave *slave;
1039         unsigned short max_hard_header_len = ETH_HLEN;
1040         unsigned int gso_max_size = GSO_MAX_SIZE;
1041         u16 gso_max_segs = GSO_MAX_SEGS;
1042
1043         if (!bond_has_slaves(bond))
1044                 goto done;
1045
1046         bond_for_each_slave(bond, slave, iter) {
1047                 vlan_features = netdev_increment_features(vlan_features,
1048                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1049
1050                 dst_release_flag &= slave->dev->priv_flags;
1051                 if (slave->dev->hard_header_len > max_hard_header_len)
1052                         max_hard_header_len = slave->dev->hard_header_len;
1053
1054                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1055                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1056         }
1057
1058 done:
1059         bond_dev->vlan_features = vlan_features;
1060         bond_dev->hard_header_len = max_hard_header_len;
1061         bond_dev->gso_max_segs = gso_max_segs;
1062         netif_set_gso_max_size(bond_dev, gso_max_size);
1063
1064         flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1065         bond_dev->priv_flags = flags | dst_release_flag;
1066
1067         netdev_change_features(bond_dev);
1068 }
1069
1070 static void bond_setup_by_slave(struct net_device *bond_dev,
1071                                 struct net_device *slave_dev)
1072 {
1073         bond_dev->header_ops        = slave_dev->header_ops;
1074
1075         bond_dev->type              = slave_dev->type;
1076         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1077         bond_dev->addr_len          = slave_dev->addr_len;
1078
1079         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1080                 slave_dev->addr_len);
1081 }
1082
1083 /* On bonding slaves other than the currently active slave, suppress
1084  * duplicates except for alb non-mcast/bcast.
1085  */
1086 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1087                                             struct slave *slave,
1088                                             struct bonding *bond)
1089 {
1090         if (bond_is_slave_inactive(slave)) {
1091                 if (bond->params.mode == BOND_MODE_ALB &&
1092                     skb->pkt_type != PACKET_BROADCAST &&
1093                     skb->pkt_type != PACKET_MULTICAST)
1094                         return false;
1095                 return true;
1096         }
1097         return false;
1098 }
1099
1100 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1101 {
1102         struct sk_buff *skb = *pskb;
1103         struct slave *slave;
1104         struct bonding *bond;
1105         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1106                           struct slave *);
1107         int ret = RX_HANDLER_ANOTHER;
1108
1109         skb = skb_share_check(skb, GFP_ATOMIC);
1110         if (unlikely(!skb))
1111                 return RX_HANDLER_CONSUMED;
1112
1113         *pskb = skb;
1114
1115         slave = bond_slave_get_rcu(skb->dev);
1116         bond = slave->bond;
1117
1118         if (bond->params.arp_interval)
1119                 slave->dev->last_rx = jiffies;
1120
1121         recv_probe = ACCESS_ONCE(bond->recv_probe);
1122         if (recv_probe) {
1123                 ret = recv_probe(skb, bond, slave);
1124                 if (ret == RX_HANDLER_CONSUMED) {
1125                         consume_skb(skb);
1126                         return ret;
1127                 }
1128         }
1129
1130         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1131                 return RX_HANDLER_EXACT;
1132         }
1133
1134         skb->dev = bond->dev;
1135
1136         if (bond->params.mode == BOND_MODE_ALB &&
1137             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1138             skb->pkt_type == PACKET_HOST) {
1139
1140                 if (unlikely(skb_cow_head(skb,
1141                                           skb->data - skb_mac_header(skb)))) {
1142                         kfree_skb(skb);
1143                         return RX_HANDLER_CONSUMED;
1144                 }
1145                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1146         }
1147
1148         return ret;
1149 }
1150
1151 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1152                                       struct net_device *slave_dev,
1153                                       struct slave *slave)
1154 {
1155         int err;
1156
1157         err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1158         if (err)
1159                 return err;
1160         slave_dev->flags |= IFF_SLAVE;
1161         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1162         return 0;
1163 }
1164
1165 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1166                                   struct net_device *slave_dev)
1167 {
1168         netdev_upper_dev_unlink(slave_dev, bond_dev);
1169         slave_dev->flags &= ~IFF_SLAVE;
1170         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1171 }
1172
1173 /* enslave device <slave> to bond device <master> */
1174 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1175 {
1176         struct bonding *bond = netdev_priv(bond_dev);
1177         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1178         struct slave *new_slave = NULL, *prev_slave;
1179         struct sockaddr addr;
1180         int link_reporting;
1181         int res = 0, i;
1182
1183         if (!bond->params.use_carrier &&
1184             slave_dev->ethtool_ops->get_link == NULL &&
1185             slave_ops->ndo_do_ioctl == NULL) {
1186                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1187                            bond_dev->name, slave_dev->name);
1188         }
1189
1190         /* already enslaved */
1191         if (slave_dev->flags & IFF_SLAVE) {
1192                 pr_debug("Error, Device was already enslaved\n");
1193                 return -EBUSY;
1194         }
1195
1196         /* vlan challenged mutual exclusion */
1197         /* no need to lock since we're protected by rtnl_lock */
1198         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1199                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1200                 if (vlan_uses_dev(bond_dev)) {
1201                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1202                                bond_dev->name, slave_dev->name, bond_dev->name);
1203                         return -EPERM;
1204                 } else {
1205                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1206                                    bond_dev->name, slave_dev->name,
1207                                    slave_dev->name, bond_dev->name);
1208                 }
1209         } else {
1210                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1211         }
1212
1213         /*
1214          * Old ifenslave binaries are no longer supported.  These can
1215          * be identified with moderate accuracy by the state of the slave:
1216          * the current ifenslave will set the interface down prior to
1217          * enslaving it; the old ifenslave will not.
1218          */
1219         if ((slave_dev->flags & IFF_UP)) {
1220                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1221                        slave_dev->name);
1222                 res = -EPERM;
1223                 goto err_undo_flags;
1224         }
1225
1226         /* set bonding device ether type by slave - bonding netdevices are
1227          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1228          * there is a need to override some of the type dependent attribs/funcs.
1229          *
1230          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1231          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1232          */
1233         if (!bond_has_slaves(bond)) {
1234                 if (bond_dev->type != slave_dev->type) {
1235                         pr_debug("%s: change device type from %d to %d\n",
1236                                  bond_dev->name,
1237                                  bond_dev->type, slave_dev->type);
1238
1239                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1240                                                        bond_dev);
1241                         res = notifier_to_errno(res);
1242                         if (res) {
1243                                 pr_err("%s: refused to change device type\n",
1244                                        bond_dev->name);
1245                                 res = -EBUSY;
1246                                 goto err_undo_flags;
1247                         }
1248
1249                         /* Flush unicast and multicast addresses */
1250                         dev_uc_flush(bond_dev);
1251                         dev_mc_flush(bond_dev);
1252
1253                         if (slave_dev->type != ARPHRD_ETHER)
1254                                 bond_setup_by_slave(bond_dev, slave_dev);
1255                         else {
1256                                 ether_setup(bond_dev);
1257                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1258                         }
1259
1260                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1261                                                  bond_dev);
1262                 }
1263         } else if (bond_dev->type != slave_dev->type) {
1264                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1265                        slave_dev->name,
1266                        slave_dev->type, bond_dev->type);
1267                 res = -EINVAL;
1268                 goto err_undo_flags;
1269         }
1270
1271         if (slave_ops->ndo_set_mac_address == NULL) {
1272                 if (!bond_has_slaves(bond)) {
1273                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1274                                    bond_dev->name);
1275                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1276                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1277                         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",
1278                                bond_dev->name);
1279                         res = -EOPNOTSUPP;
1280                         goto err_undo_flags;
1281                 }
1282         }
1283
1284         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1285
1286         /* If this is the first slave, then we need to set the master's hardware
1287          * address to be the same as the slave's. */
1288         if (!bond_has_slaves(bond) &&
1289             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1290                 bond_set_dev_addr(bond->dev, slave_dev);
1291
1292         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1293         if (!new_slave) {
1294                 res = -ENOMEM;
1295                 goto err_undo_flags;
1296         }
1297         /*
1298          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1299          * is set via sysfs or module option if desired.
1300          */
1301         new_slave->queue_id = 0;
1302
1303         /* Save slave's original mtu and then set it to match the bond */
1304         new_slave->original_mtu = slave_dev->mtu;
1305         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1306         if (res) {
1307                 pr_debug("Error %d calling dev_set_mtu\n", res);
1308                 goto err_free;
1309         }
1310
1311         /*
1312          * Save slave's original ("permanent") mac address for modes
1313          * that need it, and for restoring it upon release, and then
1314          * set it to the master's address
1315          */
1316         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1317
1318         if (!bond->params.fail_over_mac) {
1319                 /*
1320                  * Set slave to master's mac address.  The application already
1321                  * set the master's mac address to that of the first slave
1322                  */
1323                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1324                 addr.sa_family = slave_dev->type;
1325                 res = dev_set_mac_address(slave_dev, &addr);
1326                 if (res) {
1327                         pr_debug("Error %d calling set_mac_address\n", res);
1328                         goto err_restore_mtu;
1329                 }
1330         }
1331
1332         /* open the slave since the application closed it */
1333         res = dev_open(slave_dev);
1334         if (res) {
1335                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1336                 goto err_restore_mac;
1337         }
1338
1339         new_slave->bond = bond;
1340         new_slave->dev = slave_dev;
1341         slave_dev->priv_flags |= IFF_BONDING;
1342
1343         if (bond_is_lb(bond)) {
1344                 /* bond_alb_init_slave() must be called before all other stages since
1345                  * it might fail and we do not want to have to undo everything
1346                  */
1347                 res = bond_alb_init_slave(bond, new_slave);
1348                 if (res)
1349                         goto err_close;
1350         }
1351
1352         /* If the mode USES_PRIMARY, then the following is handled by
1353          * bond_change_active_slave().
1354          */
1355         if (!USES_PRIMARY(bond->params.mode)) {
1356                 /* set promiscuity level to new slave */
1357                 if (bond_dev->flags & IFF_PROMISC) {
1358                         res = dev_set_promiscuity(slave_dev, 1);
1359                         if (res)
1360                                 goto err_close;
1361                 }
1362
1363                 /* set allmulti level to new slave */
1364                 if (bond_dev->flags & IFF_ALLMULTI) {
1365                         res = dev_set_allmulti(slave_dev, 1);
1366                         if (res)
1367                                 goto err_close;
1368                 }
1369
1370                 netif_addr_lock_bh(bond_dev);
1371
1372                 dev_mc_sync_multiple(slave_dev, bond_dev);
1373                 dev_uc_sync_multiple(slave_dev, bond_dev);
1374
1375                 netif_addr_unlock_bh(bond_dev);
1376         }
1377
1378         if (bond->params.mode == BOND_MODE_8023AD) {
1379                 /* add lacpdu mc addr to mc list */
1380                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1381
1382                 dev_mc_add(slave_dev, lacpdu_multicast);
1383         }
1384
1385         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1386         if (res) {
1387                 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1388                        bond_dev->name, slave_dev->name);
1389                 goto err_close;
1390         }
1391
1392         prev_slave = bond_last_slave(bond);
1393
1394         new_slave->delay = 0;
1395         new_slave->link_failure_count = 0;
1396
1397         bond_update_speed_duplex(new_slave);
1398
1399         new_slave->last_arp_rx = jiffies -
1400                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1401         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1402                 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
1403
1404         if (bond->params.miimon && !bond->params.use_carrier) {
1405                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1406
1407                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1408                         /*
1409                          * miimon is set but a bonded network driver
1410                          * does not support ETHTOOL/MII and
1411                          * arp_interval is not set.  Note: if
1412                          * use_carrier is enabled, we will never go
1413                          * here (because netif_carrier is always
1414                          * supported); thus, we don't need to change
1415                          * the messages for netif_carrier.
1416                          */
1417                         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",
1418                                bond_dev->name, slave_dev->name);
1419                 } else if (link_reporting == -1) {
1420                         /* unable get link status using mii/ethtool */
1421                         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",
1422                                    bond_dev->name, slave_dev->name);
1423                 }
1424         }
1425
1426         /* check for initial state */
1427         if (bond->params.miimon) {
1428                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1429                         if (bond->params.updelay) {
1430                                 new_slave->link = BOND_LINK_BACK;
1431                                 new_slave->delay = bond->params.updelay;
1432                         } else {
1433                                 new_slave->link = BOND_LINK_UP;
1434                         }
1435                 } else {
1436                         new_slave->link = BOND_LINK_DOWN;
1437                 }
1438         } else if (bond->params.arp_interval) {
1439                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1440                         BOND_LINK_UP : BOND_LINK_DOWN);
1441         } else {
1442                 new_slave->link = BOND_LINK_UP;
1443         }
1444
1445         if (new_slave->link != BOND_LINK_DOWN)
1446                 new_slave->jiffies = jiffies;
1447         pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1448                 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1449                         (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1450
1451         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1452                 /* if there is a primary slave, remember it */
1453                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1454                         bond->primary_slave = new_slave;
1455                         bond->force_primary = true;
1456                 }
1457         }
1458
1459         switch (bond->params.mode) {
1460         case BOND_MODE_ACTIVEBACKUP:
1461                 bond_set_slave_inactive_flags(new_slave);
1462                 break;
1463         case BOND_MODE_8023AD:
1464                 /* in 802.3ad mode, the internal mechanism
1465                  * will activate the slaves in the selected
1466                  * aggregator
1467                  */
1468                 bond_set_slave_inactive_flags(new_slave);
1469                 /* if this is the first slave */
1470                 if (!prev_slave) {
1471                         SLAVE_AD_INFO(new_slave).id = 1;
1472                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1473                          * can be called only after the mac address of the bond is set
1474                          */
1475                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1476                 } else {
1477                         SLAVE_AD_INFO(new_slave).id =
1478                                 SLAVE_AD_INFO(prev_slave).id + 1;
1479                 }
1480
1481                 bond_3ad_bind_slave(new_slave);
1482                 break;
1483         case BOND_MODE_TLB:
1484         case BOND_MODE_ALB:
1485                 bond_set_active_slave(new_slave);
1486                 bond_set_slave_inactive_flags(new_slave);
1487                 break;
1488         default:
1489                 pr_debug("This slave is always active in trunk mode\n");
1490
1491                 /* always active in trunk mode */
1492                 bond_set_active_slave(new_slave);
1493
1494                 /* In trunking mode there is little meaning to curr_active_slave
1495                  * anyway (it holds no special properties of the bond device),
1496                  * so we can change it without calling change_active_interface()
1497                  */
1498                 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1499                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1500
1501                 break;
1502         } /* switch(bond_mode) */
1503
1504 #ifdef CONFIG_NET_POLL_CONTROLLER
1505         slave_dev->npinfo = bond->dev->npinfo;
1506         if (slave_dev->npinfo) {
1507                 if (slave_enable_netpoll(new_slave)) {
1508                         read_unlock(&bond->lock);
1509                         pr_info("Error, %s: master_dev is using netpoll, "
1510                                  "but new slave device does not support netpoll.\n",
1511                                  bond_dev->name);
1512                         res = -EBUSY;
1513                         goto err_detach;
1514                 }
1515         }
1516 #endif
1517
1518         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1519                                          new_slave);
1520         if (res) {
1521                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1522                 goto err_detach;
1523         }
1524
1525         res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1526         if (res) {
1527                 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1528                 goto err_unregister;
1529         }
1530
1531         res = bond_sysfs_slave_add(new_slave);
1532         if (res) {
1533                 pr_debug("Error %d calling bond_sysfs_slave_add\n", res);
1534                 goto err_upper_unlink;
1535         }
1536
1537         bond->slave_cnt++;
1538         bond_compute_features(bond);
1539         bond_set_carrier(bond);
1540
1541         if (USES_PRIMARY(bond->params.mode)) {
1542                 write_lock_bh(&bond->curr_slave_lock);
1543                 bond_select_active_slave(bond);
1544                 write_unlock_bh(&bond->curr_slave_lock);
1545         }
1546
1547         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1548                 bond_dev->name, slave_dev->name,
1549                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1550                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1551
1552         /* enslave is successful */
1553         return 0;
1554
1555 /* Undo stages on error */
1556 err_upper_unlink:
1557         bond_upper_dev_unlink(bond_dev, slave_dev);
1558
1559 err_unregister:
1560         netdev_rx_handler_unregister(slave_dev);
1561
1562 err_detach:
1563         if (!USES_PRIMARY(bond->params.mode))
1564                 bond_hw_addr_flush(bond_dev, slave_dev);
1565
1566         vlan_vids_del_by_dev(slave_dev, bond_dev);
1567         if (bond->primary_slave == new_slave)
1568                 bond->primary_slave = NULL;
1569         if (bond->curr_active_slave == new_slave) {
1570                 write_lock_bh(&bond->curr_slave_lock);
1571                 bond_change_active_slave(bond, NULL);
1572                 bond_select_active_slave(bond);
1573                 write_unlock_bh(&bond->curr_slave_lock);
1574         }
1575         slave_disable_netpoll(new_slave);
1576
1577 err_close:
1578         slave_dev->priv_flags &= ~IFF_BONDING;
1579         dev_close(slave_dev);
1580
1581 err_restore_mac:
1582         if (!bond->params.fail_over_mac) {
1583                 /* XXX TODO - fom follow mode needs to change master's
1584                  * MAC if this slave's MAC is in use by the bond, or at
1585                  * least print a warning.
1586                  */
1587                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1588                 addr.sa_family = slave_dev->type;
1589                 dev_set_mac_address(slave_dev, &addr);
1590         }
1591
1592 err_restore_mtu:
1593         dev_set_mtu(slave_dev, new_slave->original_mtu);
1594
1595 err_free:
1596         kfree(new_slave);
1597
1598 err_undo_flags:
1599         /* Enslave of first slave has failed and we need to fix master's mac */
1600         if (!bond_has_slaves(bond) &&
1601             ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
1602                 eth_hw_addr_random(bond_dev);
1603
1604         return res;
1605 }
1606
1607 /*
1608  * Try to release the slave device <slave> from the bond device <master>
1609  * It is legal to access curr_active_slave without a lock because all the function
1610  * is write-locked. If "all" is true it means that the function is being called
1611  * while destroying a bond interface and all slaves are being released.
1612  *
1613  * The rules for slave state should be:
1614  *   for Active/Backup:
1615  *     Active stays on all backups go down
1616  *   for Bonded connections:
1617  *     The first up interface should be left on and all others downed.
1618  */
1619 static int __bond_release_one(struct net_device *bond_dev,
1620                               struct net_device *slave_dev,
1621                               bool all)
1622 {
1623         struct bonding *bond = netdev_priv(bond_dev);
1624         struct slave *slave, *oldcurrent;
1625         struct sockaddr addr;
1626         int old_flags = bond_dev->flags;
1627         netdev_features_t old_features = bond_dev->features;
1628
1629         /* slave is not a slave or master is not master of this slave */
1630         if (!(slave_dev->flags & IFF_SLAVE) ||
1631             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1632                 pr_err("%s: Error: cannot release %s.\n",
1633                        bond_dev->name, slave_dev->name);
1634                 return -EINVAL;
1635         }
1636
1637         block_netpoll_tx();
1638
1639         slave = bond_get_slave_by_dev(bond, slave_dev);
1640         if (!slave) {
1641                 /* not a slave of this bond */
1642                 pr_info("%s: %s not enslaved\n",
1643                         bond_dev->name, slave_dev->name);
1644                 unblock_netpoll_tx();
1645                 return -EINVAL;
1646         }
1647
1648         /* release the slave from its bond */
1649         bond->slave_cnt--;
1650
1651         bond_sysfs_slave_del(slave);
1652
1653         bond_upper_dev_unlink(bond_dev, slave_dev);
1654         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1655          * for this slave anymore.
1656          */
1657         netdev_rx_handler_unregister(slave_dev);
1658         write_lock_bh(&bond->lock);
1659
1660         /* Inform AD package of unbinding of slave. */
1661         if (bond->params.mode == BOND_MODE_8023AD)
1662                 bond_3ad_unbind_slave(slave);
1663
1664         write_unlock_bh(&bond->lock);
1665
1666         pr_info("%s: releasing %s interface %s\n",
1667                 bond_dev->name,
1668                 bond_is_active_slave(slave) ? "active" : "backup",
1669                 slave_dev->name);
1670
1671         oldcurrent = bond->curr_active_slave;
1672
1673         bond->current_arp_slave = NULL;
1674
1675         if (!all && !bond->params.fail_over_mac) {
1676                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1677                     bond_has_slaves(bond))
1678                         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",
1679                                    bond_dev->name, slave_dev->name,
1680                                    slave->perm_hwaddr,
1681                                    bond_dev->name, slave_dev->name);
1682         }
1683
1684         if (bond->primary_slave == slave)
1685                 bond->primary_slave = NULL;
1686
1687         if (oldcurrent == slave) {
1688                 write_lock_bh(&bond->curr_slave_lock);
1689                 bond_change_active_slave(bond, NULL);
1690                 write_unlock_bh(&bond->curr_slave_lock);
1691         }
1692
1693         if (bond_is_lb(bond)) {
1694                 /* Must be called only after the slave has been
1695                  * detached from the list and the curr_active_slave
1696                  * has been cleared (if our_slave == old_current),
1697                  * but before a new active slave is selected.
1698                  */
1699                 bond_alb_deinit_slave(bond, slave);
1700         }
1701
1702         if (all) {
1703                 rcu_assign_pointer(bond->curr_active_slave, NULL);
1704         } else if (oldcurrent == slave) {
1705                 /*
1706                  * Note that we hold RTNL over this sequence, so there
1707                  * is no concern that another slave add/remove event
1708                  * will interfere.
1709                  */
1710                 write_lock_bh(&bond->curr_slave_lock);
1711
1712                 bond_select_active_slave(bond);
1713
1714                 write_unlock_bh(&bond->curr_slave_lock);
1715         }
1716
1717         if (!bond_has_slaves(bond)) {
1718                 bond_set_carrier(bond);
1719                 eth_hw_addr_random(bond_dev);
1720
1721                 if (vlan_uses_dev(bond_dev)) {
1722                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1723                                    bond_dev->name, bond_dev->name);
1724                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1725                                    bond_dev->name);
1726                 }
1727         }
1728
1729         unblock_netpoll_tx();
1730         synchronize_rcu();
1731
1732         if (!bond_has_slaves(bond)) {
1733                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1734                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1735         }
1736
1737         bond_compute_features(bond);
1738         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1739             (old_features & NETIF_F_VLAN_CHALLENGED))
1740                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1741                         bond_dev->name, slave_dev->name, bond_dev->name);
1742
1743         /* must do this from outside any spinlocks */
1744         vlan_vids_del_by_dev(slave_dev, bond_dev);
1745
1746         /* If the mode USES_PRIMARY, then this cases was handled above by
1747          * bond_change_active_slave(..., NULL)
1748          */
1749         if (!USES_PRIMARY(bond->params.mode)) {
1750                 /* unset promiscuity level from slave
1751                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1752                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1753                  * value of that flag before that change, as that was the value
1754                  * when this slave was attached, so we cache at the start of the
1755                  * function and use it here. Same goes for ALLMULTI below
1756                  */
1757                 if (old_flags & IFF_PROMISC)
1758                         dev_set_promiscuity(slave_dev, -1);
1759
1760                 /* unset allmulti level from slave */
1761                 if (old_flags & IFF_ALLMULTI)
1762                         dev_set_allmulti(slave_dev, -1);
1763
1764                 bond_hw_addr_flush(bond_dev, slave_dev);
1765         }
1766
1767         slave_disable_netpoll(slave);
1768
1769         /* close slave before restoring its mac address */
1770         dev_close(slave_dev);
1771
1772         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1773                 /* restore original ("permanent") mac address */
1774                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1775                 addr.sa_family = slave_dev->type;
1776                 dev_set_mac_address(slave_dev, &addr);
1777         }
1778
1779         dev_set_mtu(slave_dev, slave->original_mtu);
1780
1781         slave_dev->priv_flags &= ~IFF_BONDING;
1782
1783         kfree(slave);
1784
1785         return 0;  /* deletion OK */
1786 }
1787
1788 /* A wrapper used because of ndo_del_link */
1789 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1790 {
1791         return __bond_release_one(bond_dev, slave_dev, false);
1792 }
1793
1794 /*
1795 * First release a slave and then destroy the bond if no more slaves are left.
1796 * Must be under rtnl_lock when this function is called.
1797 */
1798 static int  bond_release_and_destroy(struct net_device *bond_dev,
1799                                      struct net_device *slave_dev)
1800 {
1801         struct bonding *bond = netdev_priv(bond_dev);
1802         int ret;
1803
1804         ret = bond_release(bond_dev, slave_dev);
1805         if (ret == 0 && !bond_has_slaves(bond)) {
1806                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1807                 pr_info("%s: destroying bond %s.\n",
1808                         bond_dev->name, bond_dev->name);
1809                 unregister_netdevice(bond_dev);
1810         }
1811         return ret;
1812 }
1813
1814 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1815 {
1816         struct bonding *bond = netdev_priv(bond_dev);
1817
1818         info->bond_mode = bond->params.mode;
1819         info->miimon = bond->params.miimon;
1820
1821         read_lock(&bond->lock);
1822         info->num_slaves = bond->slave_cnt;
1823         read_unlock(&bond->lock);
1824
1825         return 0;
1826 }
1827
1828 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1829 {
1830         struct bonding *bond = netdev_priv(bond_dev);
1831         struct list_head *iter;
1832         int i = 0, res = -ENODEV;
1833         struct slave *slave;
1834
1835         read_lock(&bond->lock);
1836         bond_for_each_slave(bond, slave, iter) {
1837                 if (i++ == (int)info->slave_id) {
1838                         res = 0;
1839                         strcpy(info->slave_name, slave->dev->name);
1840                         info->link = slave->link;
1841                         info->state = bond_slave_state(slave);
1842                         info->link_failure_count = slave->link_failure_count;
1843                         break;
1844                 }
1845         }
1846         read_unlock(&bond->lock);
1847
1848         return res;
1849 }
1850
1851 /*-------------------------------- Monitoring -------------------------------*/
1852
1853
1854 static int bond_miimon_inspect(struct bonding *bond)
1855 {
1856         int link_state, commit = 0;
1857         struct list_head *iter;
1858         struct slave *slave;
1859         bool ignore_updelay;
1860
1861         ignore_updelay = !bond->curr_active_slave ? true : false;
1862
1863         bond_for_each_slave_rcu(bond, slave, iter) {
1864                 slave->new_link = BOND_LINK_NOCHANGE;
1865
1866                 link_state = bond_check_dev_link(bond, slave->dev, 0);
1867
1868                 switch (slave->link) {
1869                 case BOND_LINK_UP:
1870                         if (link_state)
1871                                 continue;
1872
1873                         slave->link = BOND_LINK_FAIL;
1874                         slave->delay = bond->params.downdelay;
1875                         if (slave->delay) {
1876                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
1877                                         bond->dev->name,
1878                                         (bond->params.mode ==
1879                                          BOND_MODE_ACTIVEBACKUP) ?
1880                                         (bond_is_active_slave(slave) ?
1881                                          "active " : "backup ") : "",
1882                                         slave->dev->name,
1883                                         bond->params.downdelay * bond->params.miimon);
1884                         }
1885                         /*FALLTHRU*/
1886                 case BOND_LINK_FAIL:
1887                         if (link_state) {
1888                                 /*
1889                                  * recovered before downdelay expired
1890                                  */
1891                                 slave->link = BOND_LINK_UP;
1892                                 slave->jiffies = jiffies;
1893                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
1894                                         bond->dev->name,
1895                                         (bond->params.downdelay - slave->delay) *
1896                                         bond->params.miimon,
1897                                         slave->dev->name);
1898                                 continue;
1899                         }
1900
1901                         if (slave->delay <= 0) {
1902                                 slave->new_link = BOND_LINK_DOWN;
1903                                 commit++;
1904                                 continue;
1905                         }
1906
1907                         slave->delay--;
1908                         break;
1909
1910                 case BOND_LINK_DOWN:
1911                         if (!link_state)
1912                                 continue;
1913
1914                         slave->link = BOND_LINK_BACK;
1915                         slave->delay = bond->params.updelay;
1916
1917                         if (slave->delay) {
1918                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
1919                                         bond->dev->name, slave->dev->name,
1920                                         ignore_updelay ? 0 :
1921                                         bond->params.updelay *
1922                                         bond->params.miimon);
1923                         }
1924                         /*FALLTHRU*/
1925                 case BOND_LINK_BACK:
1926                         if (!link_state) {
1927                                 slave->link = BOND_LINK_DOWN;
1928                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
1929                                         bond->dev->name,
1930                                         (bond->params.updelay - slave->delay) *
1931                                         bond->params.miimon,
1932                                         slave->dev->name);
1933
1934                                 continue;
1935                         }
1936
1937                         if (ignore_updelay)
1938                                 slave->delay = 0;
1939
1940                         if (slave->delay <= 0) {
1941                                 slave->new_link = BOND_LINK_UP;
1942                                 commit++;
1943                                 ignore_updelay = false;
1944                                 continue;
1945                         }
1946
1947                         slave->delay--;
1948                         break;
1949                 }
1950         }
1951
1952         return commit;
1953 }
1954
1955 static void bond_miimon_commit(struct bonding *bond)
1956 {
1957         struct list_head *iter;
1958         struct slave *slave;
1959
1960         bond_for_each_slave(bond, slave, iter) {
1961                 switch (slave->new_link) {
1962                 case BOND_LINK_NOCHANGE:
1963                         continue;
1964
1965                 case BOND_LINK_UP:
1966                         slave->link = BOND_LINK_UP;
1967                         slave->jiffies = jiffies;
1968
1969                         if (bond->params.mode == BOND_MODE_8023AD) {
1970                                 /* prevent it from being the active one */
1971                                 bond_set_backup_slave(slave);
1972                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1973                                 /* make it immediately active */
1974                                 bond_set_active_slave(slave);
1975                         } else if (slave != bond->primary_slave) {
1976                                 /* prevent it from being the active one */
1977                                 bond_set_backup_slave(slave);
1978                         }
1979
1980                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
1981                                 bond->dev->name, slave->dev->name,
1982                                 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
1983                                 slave->duplex ? "full" : "half");
1984
1985                         /* notify ad that the link status has changed */
1986                         if (bond->params.mode == BOND_MODE_8023AD)
1987                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
1988
1989                         if (bond_is_lb(bond))
1990                                 bond_alb_handle_link_change(bond, slave,
1991                                                             BOND_LINK_UP);
1992
1993                         if (!bond->curr_active_slave ||
1994                             (slave == bond->primary_slave))
1995                                 goto do_failover;
1996
1997                         continue;
1998
1999                 case BOND_LINK_DOWN:
2000                         if (slave->link_failure_count < UINT_MAX)
2001                                 slave->link_failure_count++;
2002
2003                         slave->link = BOND_LINK_DOWN;
2004
2005                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2006                             bond->params.mode == BOND_MODE_8023AD)
2007                                 bond_set_slave_inactive_flags(slave);
2008
2009                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2010                                 bond->dev->name, slave->dev->name);
2011
2012                         if (bond->params.mode == BOND_MODE_8023AD)
2013                                 bond_3ad_handle_link_change(slave,
2014                                                             BOND_LINK_DOWN);
2015
2016                         if (bond_is_lb(bond))
2017                                 bond_alb_handle_link_change(bond, slave,
2018                                                             BOND_LINK_DOWN);
2019
2020                         if (slave == bond->curr_active_slave)
2021                                 goto do_failover;
2022
2023                         continue;
2024
2025                 default:
2026                         pr_err("%s: invalid new link %d on slave %s\n",
2027                                bond->dev->name, slave->new_link,
2028                                slave->dev->name);
2029                         slave->new_link = BOND_LINK_NOCHANGE;
2030
2031                         continue;
2032                 }
2033
2034 do_failover:
2035                 ASSERT_RTNL();
2036                 block_netpoll_tx();
2037                 write_lock_bh(&bond->curr_slave_lock);
2038                 bond_select_active_slave(bond);
2039                 write_unlock_bh(&bond->curr_slave_lock);
2040                 unblock_netpoll_tx();
2041         }
2042
2043         bond_set_carrier(bond);
2044 }
2045
2046 /*
2047  * bond_mii_monitor
2048  *
2049  * Really a wrapper that splits the mii monitor into two phases: an
2050  * inspection, then (if inspection indicates something needs to be done)
2051  * an acquisition of appropriate locks followed by a commit phase to
2052  * implement whatever link state changes are indicated.
2053  */
2054 static void bond_mii_monitor(struct work_struct *work)
2055 {
2056         struct bonding *bond = container_of(work, struct bonding,
2057                                             mii_work.work);
2058         bool should_notify_peers = false;
2059         unsigned long delay;
2060
2061         delay = msecs_to_jiffies(bond->params.miimon);
2062
2063         if (!bond_has_slaves(bond))
2064                 goto re_arm;
2065
2066         rcu_read_lock();
2067
2068         should_notify_peers = bond_should_notify_peers(bond);
2069
2070         if (bond_miimon_inspect(bond)) {
2071                 rcu_read_unlock();
2072
2073                 /* Race avoidance with bond_close cancel of workqueue */
2074                 if (!rtnl_trylock()) {
2075                         delay = 1;
2076                         should_notify_peers = false;
2077                         goto re_arm;
2078                 }
2079
2080                 bond_miimon_commit(bond);
2081
2082                 rtnl_unlock();  /* might sleep, hold no other locks */
2083         } else
2084                 rcu_read_unlock();
2085
2086 re_arm:
2087         if (bond->params.miimon)
2088                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2089
2090         if (should_notify_peers) {
2091                 if (!rtnl_trylock())
2092                         return;
2093                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2094                 rtnl_unlock();
2095         }
2096 }
2097
2098 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2099 {
2100         struct net_device *upper;
2101         struct list_head *iter;
2102         bool ret = false;
2103
2104         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2105                 return true;
2106
2107         rcu_read_lock();
2108         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2109                 if (ip == bond_confirm_addr(upper, 0, ip)) {
2110                         ret = true;
2111                         break;
2112                 }
2113         }
2114         rcu_read_unlock();
2115
2116         return ret;
2117 }
2118
2119 /*
2120  * We go to the (large) trouble of VLAN tagging ARP frames because
2121  * switches in VLAN mode (especially if ports are configured as
2122  * "native" to a VLAN) might not pass non-tagged frames.
2123  */
2124 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2125 {
2126         struct sk_buff *skb;
2127
2128         pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2129                  slave_dev->name, &dest_ip, &src_ip, vlan_id);
2130
2131         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2132                          NULL, slave_dev->dev_addr, NULL);
2133
2134         if (!skb) {
2135                 pr_err("ARP packet allocation failed\n");
2136                 return;
2137         }
2138         if (vlan_id) {
2139                 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2140                 if (!skb) {
2141                         pr_err("failed to insert VLAN tag\n");
2142                         return;
2143                 }
2144         }
2145         arp_xmit(skb);
2146 }
2147
2148
2149 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2150 {
2151         struct net_device *upper, *vlan_upper;
2152         struct list_head *iter, *vlan_iter;
2153         struct rtable *rt;
2154         __be32 *targets = bond->params.arp_targets, addr;
2155         int i, vlan_id;
2156
2157         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2158                 pr_debug("basa: target %pI4\n", &targets[i]);
2159
2160                 /* Find out through which dev should the packet go */
2161                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2162                                      RTO_ONLINK, 0);
2163                 if (IS_ERR(rt)) {
2164                         pr_debug("%s: no route to arp_ip_target %pI4\n",
2165                                  bond->dev->name, &targets[i]);
2166                         continue;
2167                 }
2168
2169                 vlan_id = 0;
2170
2171                 /* bond device itself */
2172                 if (rt->dst.dev == bond->dev)
2173                         goto found;
2174
2175                 rcu_read_lock();
2176                 /* first we search only for vlan devices. for every vlan
2177                  * found we verify its upper dev list, searching for the
2178                  * rt->dst.dev. If found we save the tag of the vlan and
2179                  * proceed to send the packet.
2180                  *
2181                  * TODO: QinQ?
2182                  */
2183                 netdev_for_each_all_upper_dev_rcu(bond->dev, vlan_upper,
2184                                                   vlan_iter) {
2185                         if (!is_vlan_dev(vlan_upper))
2186                                 continue;
2187                         netdev_for_each_all_upper_dev_rcu(vlan_upper, upper,
2188                                                           iter) {
2189                                 if (upper == rt->dst.dev) {
2190                                         vlan_id = vlan_dev_vlan_id(vlan_upper);
2191                                         rcu_read_unlock();
2192                                         goto found;
2193                                 }
2194                         }
2195                 }
2196
2197                 /* if the device we're looking for is not on top of any of
2198                  * our upper vlans, then just search for any dev that
2199                  * matches, and in case it's a vlan - save the id
2200                  */
2201                 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2202                         if (upper == rt->dst.dev) {
2203                                 /* if it's a vlan - get its VID */
2204                                 if (is_vlan_dev(upper))
2205                                         vlan_id = vlan_dev_vlan_id(upper);
2206
2207                                 rcu_read_unlock();
2208                                 goto found;
2209                         }
2210                 }
2211                 rcu_read_unlock();
2212
2213                 /* Not our device - skip */
2214                 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2215                          bond->dev->name, &targets[i],
2216                          rt->dst.dev ? rt->dst.dev->name : "NULL");
2217
2218                 ip_rt_put(rt);
2219                 continue;
2220
2221 found:
2222                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2223                 ip_rt_put(rt);
2224                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2225                               addr, vlan_id);
2226         }
2227 }
2228
2229 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2230 {
2231         int i;
2232
2233         if (!sip || !bond_has_this_ip(bond, tip)) {
2234                 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2235                 return;
2236         }
2237
2238         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2239         if (i == -1) {
2240                 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2241                 return;
2242         }
2243         slave->last_arp_rx = jiffies;
2244         slave->target_last_arp_rx[i] = jiffies;
2245 }
2246
2247 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2248                  struct slave *slave)
2249 {
2250         struct arphdr *arp = (struct arphdr *)skb->data;
2251         unsigned char *arp_ptr;
2252         __be32 sip, tip;
2253         int alen;
2254
2255         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2256                 return RX_HANDLER_ANOTHER;
2257
2258         read_lock(&bond->lock);
2259
2260         if (!slave_do_arp_validate(bond, slave))
2261                 goto out_unlock;
2262
2263         alen = arp_hdr_len(bond->dev);
2264
2265         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2266                  bond->dev->name, skb->dev->name);
2267
2268         if (alen > skb_headlen(skb)) {
2269                 arp = kmalloc(alen, GFP_ATOMIC);
2270                 if (!arp)
2271                         goto out_unlock;
2272                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2273                         goto out_unlock;
2274         }
2275
2276         if (arp->ar_hln != bond->dev->addr_len ||
2277             skb->pkt_type == PACKET_OTHERHOST ||
2278             skb->pkt_type == PACKET_LOOPBACK ||
2279             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2280             arp->ar_pro != htons(ETH_P_IP) ||
2281             arp->ar_pln != 4)
2282                 goto out_unlock;
2283
2284         arp_ptr = (unsigned char *)(arp + 1);
2285         arp_ptr += bond->dev->addr_len;
2286         memcpy(&sip, arp_ptr, 4);
2287         arp_ptr += 4 + bond->dev->addr_len;
2288         memcpy(&tip, arp_ptr, 4);
2289
2290         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2291                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2292                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2293                  &sip, &tip);
2294
2295         /*
2296          * Backup slaves won't see the ARP reply, but do come through
2297          * here for each ARP probe (so we swap the sip/tip to validate
2298          * the probe).  In a "redundant switch, common router" type of
2299          * configuration, the ARP probe will (hopefully) travel from
2300          * the active, through one switch, the router, then the other
2301          * switch before reaching the backup.
2302          *
2303          * We 'trust' the arp requests if there is an active slave and
2304          * it received valid arp reply(s) after it became active. This
2305          * is done to avoid endless looping when we can't reach the
2306          * arp_ip_target and fool ourselves with our own arp requests.
2307          */
2308         if (bond_is_active_slave(slave))
2309                 bond_validate_arp(bond, slave, sip, tip);
2310         else if (bond->curr_active_slave &&
2311                  time_after(slave_last_rx(bond, bond->curr_active_slave),
2312                             bond->curr_active_slave->jiffies))
2313                 bond_validate_arp(bond, slave, tip, sip);
2314
2315 out_unlock:
2316         read_unlock(&bond->lock);
2317         if (arp != (struct arphdr *)skb->data)
2318                 kfree(arp);
2319         return RX_HANDLER_ANOTHER;
2320 }
2321
2322 /* function to verify if we're in the arp_interval timeslice, returns true if
2323  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2324  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2325  */
2326 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2327                                   int mod)
2328 {
2329         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2330
2331         return time_in_range(jiffies,
2332                              last_act - delta_in_ticks,
2333                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2334 }
2335
2336 /*
2337  * this function is called regularly to monitor each slave's link
2338  * ensuring that traffic is being sent and received when arp monitoring
2339  * is used in load-balancing mode. if the adapter has been dormant, then an
2340  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2341  * arp monitoring in active backup mode.
2342  */
2343 static void bond_loadbalance_arp_mon(struct work_struct *work)
2344 {
2345         struct bonding *bond = container_of(work, struct bonding,
2346                                             arp_work.work);
2347         struct slave *slave, *oldcurrent;
2348         struct list_head *iter;
2349         int do_failover = 0;
2350
2351         if (!bond_has_slaves(bond))
2352                 goto re_arm;
2353
2354         rcu_read_lock();
2355
2356         oldcurrent = ACCESS_ONCE(bond->curr_active_slave);
2357         /* see if any of the previous devices are up now (i.e. they have
2358          * xmt and rcv traffic). the curr_active_slave does not come into
2359          * the picture unless it is null. also, slave->jiffies is not needed
2360          * here because we send an arp on each slave and give a slave as
2361          * long as it needs to get the tx/rx within the delta.
2362          * TODO: what about up/down delay in arp mode? it wasn't here before
2363          *       so it can wait
2364          */
2365         bond_for_each_slave_rcu(bond, slave, iter) {
2366                 unsigned long trans_start = dev_trans_start(slave->dev);
2367
2368                 if (slave->link != BOND_LINK_UP) {
2369                         if (bond_time_in_interval(bond, trans_start, 1) &&
2370                             bond_time_in_interval(bond, slave->dev->last_rx, 1)) {
2371
2372                                 slave->link  = BOND_LINK_UP;
2373                                 bond_set_active_slave(slave);
2374
2375                                 /* primary_slave has no meaning in round-robin
2376                                  * mode. the window of a slave being up and
2377                                  * curr_active_slave being null after enslaving
2378                                  * is closed.
2379                                  */
2380                                 if (!oldcurrent) {
2381                                         pr_info("%s: link status definitely up for interface %s, ",
2382                                                 bond->dev->name,
2383                                                 slave->dev->name);
2384                                         do_failover = 1;
2385                                 } else {
2386                                         pr_info("%s: interface %s is now up\n",
2387                                                 bond->dev->name,
2388                                                 slave->dev->name);
2389                                 }
2390                         }
2391                 } else {
2392                         /* slave->link == BOND_LINK_UP */
2393
2394                         /* not all switches will respond to an arp request
2395                          * when the source ip is 0, so don't take the link down
2396                          * if we don't know our ip yet
2397                          */
2398                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2399                             !bond_time_in_interval(bond, slave->dev->last_rx, 2)) {
2400
2401                                 slave->link  = BOND_LINK_DOWN;
2402                                 bond_set_backup_slave(slave);
2403
2404                                 if (slave->link_failure_count < UINT_MAX)
2405                                         slave->link_failure_count++;
2406
2407                                 pr_info("%s: interface %s is now down.\n",
2408                                         bond->dev->name,
2409                                         slave->dev->name);
2410
2411                                 if (slave == oldcurrent)
2412                                         do_failover = 1;
2413                         }
2414                 }
2415
2416                 /* note: if switch is in round-robin mode, all links
2417                  * must tx arp to ensure all links rx an arp - otherwise
2418                  * links may oscillate or not come up at all; if switch is
2419                  * in something like xor mode, there is nothing we can
2420                  * do - all replies will be rx'ed on same link causing slaves
2421                  * to be unstable during low/no traffic periods
2422                  */
2423                 if (IS_UP(slave->dev))
2424                         bond_arp_send_all(bond, slave);
2425         }
2426
2427         rcu_read_unlock();
2428
2429         if (do_failover) {
2430                 /* the bond_select_active_slave must hold RTNL
2431                  * and curr_slave_lock for write.
2432                  */
2433                 if (!rtnl_trylock())
2434                         goto re_arm;
2435                 block_netpoll_tx();
2436                 write_lock_bh(&bond->curr_slave_lock);
2437
2438                 bond_select_active_slave(bond);
2439
2440                 write_unlock_bh(&bond->curr_slave_lock);
2441                 unblock_netpoll_tx();
2442                 rtnl_unlock();
2443         }
2444
2445 re_arm:
2446         if (bond->params.arp_interval)
2447                 queue_delayed_work(bond->wq, &bond->arp_work,
2448                                    msecs_to_jiffies(bond->params.arp_interval));
2449 }
2450
2451 /*
2452  * Called to inspect slaves for active-backup mode ARP monitor link state
2453  * changes.  Sets new_link in slaves to specify what action should take
2454  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2455  * to link states must be committed.
2456  *
2457  * Called with rcu_read_lock hold.
2458  */
2459 static int bond_ab_arp_inspect(struct bonding *bond)
2460 {
2461         unsigned long trans_start, last_rx;
2462         struct list_head *iter;
2463         struct slave *slave;
2464         int commit = 0;
2465
2466         bond_for_each_slave_rcu(bond, slave, iter) {
2467                 slave->new_link = BOND_LINK_NOCHANGE;
2468                 last_rx = slave_last_rx(bond, slave);
2469
2470                 if (slave->link != BOND_LINK_UP) {
2471                         if (bond_time_in_interval(bond, last_rx, 1)) {
2472                                 slave->new_link = BOND_LINK_UP;
2473                                 commit++;
2474                         }
2475                         continue;
2476                 }
2477
2478                 /*
2479                  * Give slaves 2*delta after being enslaved or made
2480                  * active.  This avoids bouncing, as the last receive
2481                  * times need a full ARP monitor cycle to be updated.
2482                  */
2483                 if (bond_time_in_interval(bond, slave->jiffies, 2))
2484                         continue;
2485
2486                 /*
2487                  * Backup slave is down if:
2488                  * - No current_arp_slave AND
2489                  * - more than 3*delta since last receive AND
2490                  * - the bond has an IP address
2491                  *
2492                  * Note: a non-null current_arp_slave indicates
2493                  * the curr_active_slave went down and we are
2494                  * searching for a new one; under this condition
2495                  * we only take the curr_active_slave down - this
2496                  * gives each slave a chance to tx/rx traffic
2497                  * before being taken out
2498                  */
2499                 if (!bond_is_active_slave(slave) &&
2500                     !bond->current_arp_slave &&
2501                     !bond_time_in_interval(bond, last_rx, 3)) {
2502                         slave->new_link = BOND_LINK_DOWN;
2503                         commit++;
2504                 }
2505
2506                 /*
2507                  * Active slave is down if:
2508                  * - more than 2*delta since transmitting OR
2509                  * - (more than 2*delta since receive AND
2510                  *    the bond has an IP address)
2511                  */
2512                 trans_start = dev_trans_start(slave->dev);
2513                 if (bond_is_active_slave(slave) &&
2514                     (!bond_time_in_interval(bond, trans_start, 2) ||
2515                      !bond_time_in_interval(bond, last_rx, 2))) {
2516                         slave->new_link = BOND_LINK_DOWN;
2517                         commit++;
2518                 }
2519         }
2520
2521         return commit;
2522 }
2523
2524 /*
2525  * Called to commit link state changes noted by inspection step of
2526  * active-backup mode ARP monitor.
2527  *
2528  * Called with RTNL hold.
2529  */
2530 static void bond_ab_arp_commit(struct bonding *bond)
2531 {
2532         unsigned long trans_start;
2533         struct list_head *iter;
2534         struct slave *slave;
2535
2536         bond_for_each_slave(bond, slave, iter) {
2537                 switch (slave->new_link) {
2538                 case BOND_LINK_NOCHANGE:
2539                         continue;
2540
2541                 case BOND_LINK_UP:
2542                         trans_start = dev_trans_start(slave->dev);
2543                         if (bond->curr_active_slave != slave ||
2544                             (!bond->curr_active_slave &&
2545                              bond_time_in_interval(bond, trans_start, 1))) {
2546                                 slave->link = BOND_LINK_UP;
2547                                 if (bond->current_arp_slave) {
2548                                         bond_set_slave_inactive_flags(
2549                                                 bond->current_arp_slave);
2550                                         bond->current_arp_slave = NULL;
2551                                 }
2552
2553                                 pr_info("%s: link status definitely up for interface %s.\n",
2554                                         bond->dev->name, slave->dev->name);
2555
2556                                 if (!bond->curr_active_slave ||
2557                                     (slave == bond->primary_slave))
2558                                         goto do_failover;
2559
2560                         }
2561
2562                         continue;
2563
2564                 case BOND_LINK_DOWN:
2565                         if (slave->link_failure_count < UINT_MAX)
2566                                 slave->link_failure_count++;
2567
2568                         slave->link = BOND_LINK_DOWN;
2569                         bond_set_slave_inactive_flags(slave);
2570
2571                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2572                                 bond->dev->name, slave->dev->name);
2573
2574                         if (slave == bond->curr_active_slave) {
2575                                 bond->current_arp_slave = NULL;
2576                                 goto do_failover;
2577                         }
2578
2579                         continue;
2580
2581                 default:
2582                         pr_err("%s: impossible: new_link %d on slave %s\n",
2583                                bond->dev->name, slave->new_link,
2584                                slave->dev->name);
2585                         continue;
2586                 }
2587
2588 do_failover:
2589                 ASSERT_RTNL();
2590                 block_netpoll_tx();
2591                 write_lock_bh(&bond->curr_slave_lock);
2592                 bond_select_active_slave(bond);
2593                 write_unlock_bh(&bond->curr_slave_lock);
2594                 unblock_netpoll_tx();
2595         }
2596
2597         bond_set_carrier(bond);
2598 }
2599
2600 /*
2601  * Send ARP probes for active-backup mode ARP monitor.
2602  *
2603  * Called with rcu_read_lock hold.
2604  */
2605 static void bond_ab_arp_probe(struct bonding *bond)
2606 {
2607         struct slave *slave, *before = NULL, *new_slave = NULL,
2608                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2609         struct list_head *iter;
2610         bool found = false;
2611
2612         read_lock(&bond->curr_slave_lock);
2613
2614         if (curr_arp_slave && bond->curr_active_slave)
2615                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2616                         curr_arp_slave->dev->name,
2617                         bond->curr_active_slave->dev->name);
2618
2619         if (bond->curr_active_slave) {
2620                 bond_arp_send_all(bond, bond->curr_active_slave);
2621                 read_unlock(&bond->curr_slave_lock);
2622                 return;
2623         }
2624
2625         read_unlock(&bond->curr_slave_lock);
2626
2627         /* if we don't have a curr_active_slave, search for the next available
2628          * backup slave from the current_arp_slave and make it the candidate
2629          * for becoming the curr_active_slave
2630          */
2631
2632         if (!curr_arp_slave) {
2633                 curr_arp_slave = bond_first_slave_rcu(bond);
2634                 if (!curr_arp_slave)
2635                         return;
2636         }
2637
2638         bond_set_slave_inactive_flags(curr_arp_slave);
2639
2640         bond_for_each_slave_rcu(bond, slave, iter) {
2641                 if (!found && !before && IS_UP(slave->dev))
2642                         before = slave;
2643
2644                 if (found && !new_slave && IS_UP(slave->dev))
2645                         new_slave = slave;
2646                 /* if the link state is up at this point, we
2647                  * mark it down - this can happen if we have
2648                  * simultaneous link failures and
2649                  * reselect_active_interface doesn't make this
2650                  * one the current slave so it is still marked
2651                  * up when it is actually down
2652                  */
2653                 if (!IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
2654                         slave->link = BOND_LINK_DOWN;
2655                         if (slave->link_failure_count < UINT_MAX)
2656                                 slave->link_failure_count++;
2657
2658                         bond_set_slave_inactive_flags(slave);
2659
2660                         pr_info("%s: backup interface %s is now down.\n",
2661                                 bond->dev->name, slave->dev->name);
2662                 }
2663                 if (slave == curr_arp_slave)
2664                         found = true;
2665         }
2666
2667         if (!new_slave && before)
2668                 new_slave = before;
2669
2670         if (!new_slave)
2671                 return;
2672
2673         new_slave->link = BOND_LINK_BACK;
2674         bond_set_slave_active_flags(new_slave);
2675         bond_arp_send_all(bond, new_slave);
2676         new_slave->jiffies = jiffies;
2677         rcu_assign_pointer(bond->current_arp_slave, new_slave);
2678 }
2679
2680 static void bond_activebackup_arp_mon(struct work_struct *work)
2681 {
2682         struct bonding *bond = container_of(work, struct bonding,
2683                                             arp_work.work);
2684         bool should_notify_peers = false;
2685         int delta_in_ticks;
2686
2687         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2688
2689         if (!bond_has_slaves(bond))
2690                 goto re_arm;
2691
2692         rcu_read_lock();
2693
2694         should_notify_peers = bond_should_notify_peers(bond);
2695
2696         if (bond_ab_arp_inspect(bond)) {
2697                 rcu_read_unlock();
2698
2699                 /* Race avoidance with bond_close flush of workqueue */
2700                 if (!rtnl_trylock()) {
2701                         delta_in_ticks = 1;
2702                         should_notify_peers = false;
2703                         goto re_arm;
2704                 }
2705
2706                 bond_ab_arp_commit(bond);
2707
2708                 rtnl_unlock();
2709                 rcu_read_lock();
2710         }
2711
2712         bond_ab_arp_probe(bond);
2713         rcu_read_unlock();
2714
2715 re_arm:
2716         if (bond->params.arp_interval)
2717                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2718
2719         if (should_notify_peers) {
2720                 if (!rtnl_trylock())
2721                         return;
2722                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2723                 rtnl_unlock();
2724         }
2725 }
2726
2727 /*-------------------------- netdev event handling --------------------------*/
2728
2729 /*
2730  * Change device name
2731  */
2732 static int bond_event_changename(struct bonding *bond)
2733 {
2734         bond_remove_proc_entry(bond);
2735         bond_create_proc_entry(bond);
2736
2737         bond_debug_reregister(bond);
2738
2739         return NOTIFY_DONE;
2740 }
2741
2742 static int bond_master_netdev_event(unsigned long event,
2743                                     struct net_device *bond_dev)
2744 {
2745         struct bonding *event_bond = netdev_priv(bond_dev);
2746
2747         switch (event) {
2748         case NETDEV_CHANGENAME:
2749                 return bond_event_changename(event_bond);
2750         case NETDEV_UNREGISTER:
2751                 bond_remove_proc_entry(event_bond);
2752                 break;
2753         case NETDEV_REGISTER:
2754                 bond_create_proc_entry(event_bond);
2755                 break;
2756         case NETDEV_NOTIFY_PEERS:
2757                 if (event_bond->send_peer_notif)
2758                         event_bond->send_peer_notif--;
2759                 break;
2760         default:
2761                 break;
2762         }
2763
2764         return NOTIFY_DONE;
2765 }
2766
2767 static int bond_slave_netdev_event(unsigned long event,
2768                                    struct net_device *slave_dev)
2769 {
2770         struct slave *slave = bond_slave_get_rtnl(slave_dev);
2771         struct bonding *bond;
2772         struct net_device *bond_dev;
2773         u32 old_speed;
2774         u8 old_duplex;
2775
2776         /* A netdev event can be generated while enslaving a device
2777          * before netdev_rx_handler_register is called in which case
2778          * slave will be NULL
2779          */
2780         if (!slave)
2781                 return NOTIFY_DONE;
2782         bond_dev = slave->bond->dev;
2783         bond = slave->bond;
2784
2785         switch (event) {
2786         case NETDEV_UNREGISTER:
2787                 if (bond_dev->type != ARPHRD_ETHER)
2788                         bond_release_and_destroy(bond_dev, slave_dev);
2789                 else
2790                         bond_release(bond_dev, slave_dev);
2791                 break;
2792         case NETDEV_UP:
2793         case NETDEV_CHANGE:
2794                 old_speed = slave->speed;
2795                 old_duplex = slave->duplex;
2796
2797                 bond_update_speed_duplex(slave);
2798
2799                 if (bond->params.mode == BOND_MODE_8023AD) {
2800                         if (old_speed != slave->speed)
2801                                 bond_3ad_adapter_speed_changed(slave);
2802                         if (old_duplex != slave->duplex)
2803                                 bond_3ad_adapter_duplex_changed(slave);
2804                 }
2805                 break;
2806         case NETDEV_DOWN:
2807                 /*
2808                  * ... Or is it this?
2809                  */
2810                 break;
2811         case NETDEV_CHANGEMTU:
2812                 /*
2813                  * TODO: Should slaves be allowed to
2814                  * independently alter their MTU?  For
2815                  * an active-backup bond, slaves need
2816                  * not be the same type of device, so
2817                  * MTUs may vary.  For other modes,
2818                  * slaves arguably should have the
2819                  * same MTUs. To do this, we'd need to
2820                  * take over the slave's change_mtu
2821                  * function for the duration of their
2822                  * servitude.
2823                  */
2824                 break;
2825         case NETDEV_CHANGENAME:
2826                 /* we don't care if we don't have primary set */
2827                 if (!USES_PRIMARY(bond->params.mode) ||
2828                     !bond->params.primary[0])
2829                         break;
2830
2831                 if (slave == bond->primary_slave) {
2832                         /* slave's name changed - he's no longer primary */
2833                         bond->primary_slave = NULL;
2834                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2835                         /* we have a new primary slave */
2836                         bond->primary_slave = slave;
2837                 } else { /* we didn't change primary - exit */
2838                         break;
2839                 }
2840
2841                 pr_info("%s: Primary slave changed to %s, reselecting active slave.\n",
2842                         bond->dev->name, bond->primary_slave ? slave_dev->name :
2843                                                                "none");
2844                 write_lock_bh(&bond->curr_slave_lock);
2845                 bond_select_active_slave(bond);
2846                 write_unlock_bh(&bond->curr_slave_lock);
2847                 break;
2848         case NETDEV_FEAT_CHANGE:
2849                 bond_compute_features(bond);
2850                 break;
2851         case NETDEV_RESEND_IGMP:
2852                 /* Propagate to master device */
2853                 call_netdevice_notifiers(event, slave->bond->dev);
2854                 break;
2855         default:
2856                 break;
2857         }
2858
2859         return NOTIFY_DONE;
2860 }
2861
2862 /*
2863  * bond_netdev_event: handle netdev notifier chain events.
2864  *
2865  * This function receives events for the netdev chain.  The caller (an
2866  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2867  * locks for us to safely manipulate the slave devices (RTNL lock,
2868  * dev_probe_lock).
2869  */
2870 static int bond_netdev_event(struct notifier_block *this,
2871                              unsigned long event, void *ptr)
2872 {
2873         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2874
2875         pr_debug("event_dev: %s, event: %lx\n",
2876                  event_dev ? event_dev->name : "None",
2877                  event);
2878
2879         if (!(event_dev->priv_flags & IFF_BONDING))
2880                 return NOTIFY_DONE;
2881
2882         if (event_dev->flags & IFF_MASTER) {
2883                 pr_debug("IFF_MASTER\n");
2884                 return bond_master_netdev_event(event, event_dev);
2885         }
2886
2887         if (event_dev->flags & IFF_SLAVE) {
2888                 pr_debug("IFF_SLAVE\n");
2889                 return bond_slave_netdev_event(event, event_dev);
2890         }
2891
2892         return NOTIFY_DONE;
2893 }
2894
2895 static struct notifier_block bond_netdev_notifier = {
2896         .notifier_call = bond_netdev_event,
2897 };
2898
2899 /*---------------------------- Hashing Policies -----------------------------*/
2900
2901 /* L2 hash helper */
2902 static inline u32 bond_eth_hash(struct sk_buff *skb)
2903 {
2904         struct ethhdr *data = (struct ethhdr *)skb->data;
2905
2906         if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
2907                 return data->h_dest[5] ^ data->h_source[5];
2908
2909         return 0;
2910 }
2911
2912 /* Extract the appropriate headers based on bond's xmit policy */
2913 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
2914                               struct flow_keys *fk)
2915 {
2916         const struct ipv6hdr *iph6;
2917         const struct iphdr *iph;
2918         int noff, proto = -1;
2919
2920         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
2921                 return skb_flow_dissect(skb, fk);
2922
2923         fk->ports = 0;
2924         noff = skb_network_offset(skb);
2925         if (skb->protocol == htons(ETH_P_IP)) {
2926                 if (!pskb_may_pull(skb, noff + sizeof(*iph)))
2927                         return false;
2928                 iph = ip_hdr(skb);
2929                 fk->src = iph->saddr;
2930                 fk->dst = iph->daddr;
2931                 noff += iph->ihl << 2;
2932                 if (!ip_is_fragment(iph))
2933                         proto = iph->protocol;
2934         } else if (skb->protocol == htons(ETH_P_IPV6)) {
2935                 if (!pskb_may_pull(skb, noff + sizeof(*iph6)))
2936                         return false;
2937                 iph6 = ipv6_hdr(skb);
2938                 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
2939                 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
2940                 noff += sizeof(*iph6);
2941                 proto = iph6->nexthdr;
2942         } else {
2943                 return false;
2944         }
2945         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
2946                 fk->ports = skb_flow_get_ports(skb, noff, proto);
2947
2948         return true;
2949 }
2950
2951 /**
2952  * bond_xmit_hash - generate a hash value based on the xmit policy
2953  * @bond: bonding device
2954  * @skb: buffer to use for headers
2955  * @count: modulo value
2956  *
2957  * This function will extract the necessary headers from the skb buffer and use
2958  * them to generate a hash based on the xmit_policy set in the bonding device
2959  * which will be reduced modulo count before returning.
2960  */
2961 int bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, int count)
2962 {
2963         struct flow_keys flow;
2964         u32 hash;
2965
2966         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
2967             !bond_flow_dissect(bond, skb, &flow))
2968                 return bond_eth_hash(skb) % count;
2969
2970         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
2971             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
2972                 hash = bond_eth_hash(skb);
2973         else
2974                 hash = (__force u32)flow.ports;
2975         hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
2976         hash ^= (hash >> 16);
2977         hash ^= (hash >> 8);
2978
2979         return hash % count;
2980 }
2981
2982 /*-------------------------- Device entry points ----------------------------*/
2983
2984 static void bond_work_init_all(struct bonding *bond)
2985 {
2986         INIT_DELAYED_WORK(&bond->mcast_work,
2987                           bond_resend_igmp_join_requests_delayed);
2988         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
2989         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
2990         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2991                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
2992         else
2993                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
2994         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
2995 }
2996
2997 static void bond_work_cancel_all(struct bonding *bond)
2998 {
2999         cancel_delayed_work_sync(&bond->mii_work);
3000         cancel_delayed_work_sync(&bond->arp_work);
3001         cancel_delayed_work_sync(&bond->alb_work);
3002         cancel_delayed_work_sync(&bond->ad_work);
3003         cancel_delayed_work_sync(&bond->mcast_work);
3004 }
3005
3006 static int bond_open(struct net_device *bond_dev)
3007 {
3008         struct bonding *bond = netdev_priv(bond_dev);
3009         struct list_head *iter;
3010         struct slave *slave;
3011
3012         /* reset slave->backup and slave->inactive */
3013         read_lock(&bond->lock);
3014         if (bond_has_slaves(bond)) {
3015                 read_lock(&bond->curr_slave_lock);
3016                 bond_for_each_slave(bond, slave, iter) {
3017                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3018                                 && (slave != bond->curr_active_slave)) {
3019                                 bond_set_slave_inactive_flags(slave);
3020                         } else {
3021                                 bond_set_slave_active_flags(slave);
3022                         }
3023                 }
3024                 read_unlock(&bond->curr_slave_lock);
3025         }
3026         read_unlock(&bond->lock);
3027
3028         bond_work_init_all(bond);
3029
3030         if (bond_is_lb(bond)) {
3031                 /* bond_alb_initialize must be called before the timer
3032                  * is started.
3033                  */
3034                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3035                         return -ENOMEM;
3036                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3037         }
3038
3039         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3040                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3041
3042         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3043                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3044                 if (bond->params.arp_validate)
3045                         bond->recv_probe = bond_arp_rcv;
3046         }
3047
3048         if (bond->params.mode == BOND_MODE_8023AD) {
3049                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3050                 /* register to receive LACPDUs */
3051                 bond->recv_probe = bond_3ad_lacpdu_recv;
3052                 bond_3ad_initiate_agg_selection(bond, 1);
3053         }
3054
3055         return 0;
3056 }
3057
3058 static int bond_close(struct net_device *bond_dev)
3059 {
3060         struct bonding *bond = netdev_priv(bond_dev);
3061
3062         bond_work_cancel_all(bond);
3063         bond->send_peer_notif = 0;
3064         if (bond_is_lb(bond))
3065                 bond_alb_deinitialize(bond);
3066         bond->recv_probe = NULL;
3067
3068         return 0;
3069 }
3070
3071 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3072                                                 struct rtnl_link_stats64 *stats)
3073 {
3074         struct bonding *bond = netdev_priv(bond_dev);
3075         struct rtnl_link_stats64 temp;
3076         struct list_head *iter;
3077         struct slave *slave;
3078
3079         memset(stats, 0, sizeof(*stats));
3080
3081         read_lock_bh(&bond->lock);
3082         bond_for_each_slave(bond, slave, iter) {
3083                 const struct rtnl_link_stats64 *sstats =
3084                         dev_get_stats(slave->dev, &temp);
3085
3086                 stats->rx_packets += sstats->rx_packets;
3087                 stats->rx_bytes += sstats->rx_bytes;
3088                 stats->rx_errors += sstats->rx_errors;
3089                 stats->rx_dropped += sstats->rx_dropped;
3090
3091                 stats->tx_packets += sstats->tx_packets;
3092                 stats->tx_bytes += sstats->tx_bytes;
3093                 stats->tx_errors += sstats->tx_errors;
3094                 stats->tx_dropped += sstats->tx_dropped;
3095
3096                 stats->multicast += sstats->multicast;
3097                 stats->collisions += sstats->collisions;
3098
3099                 stats->rx_length_errors += sstats->rx_length_errors;
3100                 stats->rx_over_errors += sstats->rx_over_errors;
3101                 stats->rx_crc_errors += sstats->rx_crc_errors;
3102                 stats->rx_frame_errors += sstats->rx_frame_errors;
3103                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3104                 stats->rx_missed_errors += sstats->rx_missed_errors;
3105
3106                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3107                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3108                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3109                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3110                 stats->tx_window_errors += sstats->tx_window_errors;
3111         }
3112         read_unlock_bh(&bond->lock);
3113
3114         return stats;
3115 }
3116
3117 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3118 {
3119         struct bonding *bond = netdev_priv(bond_dev);
3120         struct net_device *slave_dev = NULL;
3121         struct ifbond k_binfo;
3122         struct ifbond __user *u_binfo = NULL;
3123         struct ifslave k_sinfo;
3124         struct ifslave __user *u_sinfo = NULL;
3125         struct mii_ioctl_data *mii = NULL;
3126         struct bond_opt_value newval;
3127         struct net *net;
3128         int res = 0;
3129
3130         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3131
3132         switch (cmd) {
3133         case SIOCGMIIPHY:
3134                 mii = if_mii(ifr);
3135                 if (!mii)
3136                         return -EINVAL;
3137
3138                 mii->phy_id = 0;
3139                 /* Fall Through */
3140         case SIOCGMIIREG:
3141                 /*
3142                  * We do this again just in case we were called by SIOCGMIIREG
3143                  * instead of SIOCGMIIPHY.
3144                  */
3145                 mii = if_mii(ifr);
3146                 if (!mii)
3147                         return -EINVAL;
3148
3149
3150                 if (mii->reg_num == 1) {
3151                         mii->val_out = 0;
3152                         read_lock(&bond->lock);
3153                         read_lock(&bond->curr_slave_lock);
3154                         if (netif_carrier_ok(bond->dev))
3155                                 mii->val_out = BMSR_LSTATUS;
3156
3157                         read_unlock(&bond->curr_slave_lock);
3158                         read_unlock(&bond->lock);
3159                 }
3160
3161                 return 0;
3162         case BOND_INFO_QUERY_OLD:
3163         case SIOCBONDINFOQUERY:
3164                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3165
3166                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3167                         return -EFAULT;
3168
3169                 res = bond_info_query(bond_dev, &k_binfo);
3170                 if (res == 0 &&
3171                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3172                         return -EFAULT;
3173
3174                 return res;
3175         case BOND_SLAVE_INFO_QUERY_OLD:
3176         case SIOCBONDSLAVEINFOQUERY:
3177                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3178
3179                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3180                         return -EFAULT;
3181
3182                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3183                 if (res == 0 &&
3184                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3185                         return -EFAULT;
3186
3187                 return res;
3188         default:
3189                 /* Go on */
3190                 break;
3191         }
3192
3193         net = dev_net(bond_dev);
3194
3195         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3196                 return -EPERM;
3197
3198         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3199
3200         pr_debug("slave_dev=%p:\n", slave_dev);
3201
3202         if (!slave_dev)
3203                 return -ENODEV;
3204
3205         pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3206         switch (cmd) {
3207         case BOND_ENSLAVE_OLD:
3208         case SIOCBONDENSLAVE:
3209                 res = bond_enslave(bond_dev, slave_dev);
3210                 break;
3211         case BOND_RELEASE_OLD:
3212         case SIOCBONDRELEASE:
3213                 res = bond_release(bond_dev, slave_dev);
3214                 break;
3215         case BOND_SETHWADDR_OLD:
3216         case SIOCBONDSETHWADDR:
3217                 bond_set_dev_addr(bond_dev, slave_dev);
3218                 res = 0;
3219                 break;
3220         case BOND_CHANGE_ACTIVE_OLD:
3221         case SIOCBONDCHANGEACTIVE:
3222                 bond_opt_initstr(&newval, slave_dev->name);
3223                 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3224                 break;
3225         default:
3226                 res = -EOPNOTSUPP;
3227         }
3228
3229         return res;
3230 }
3231
3232 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3233 {
3234         struct bonding *bond = netdev_priv(bond_dev);
3235
3236         if (change & IFF_PROMISC)
3237                 bond_set_promiscuity(bond,
3238                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3239
3240         if (change & IFF_ALLMULTI)
3241                 bond_set_allmulti(bond,
3242                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3243 }
3244
3245 static void bond_set_rx_mode(struct net_device *bond_dev)
3246 {
3247         struct bonding *bond = netdev_priv(bond_dev);
3248         struct list_head *iter;
3249         struct slave *slave;
3250
3251
3252         rcu_read_lock();
3253         if (USES_PRIMARY(bond->params.mode)) {
3254                 slave = rcu_dereference(bond->curr_active_slave);
3255                 if (slave) {
3256                         dev_uc_sync(slave->dev, bond_dev);
3257                         dev_mc_sync(slave->dev, bond_dev);
3258                 }
3259         } else {
3260                 bond_for_each_slave_rcu(bond, slave, iter) {
3261                         dev_uc_sync_multiple(slave->dev, bond_dev);
3262                         dev_mc_sync_multiple(slave->dev, bond_dev);
3263                 }
3264         }
3265         rcu_read_unlock();
3266 }
3267
3268 static int bond_neigh_init(struct neighbour *n)
3269 {
3270         struct bonding *bond = netdev_priv(n->dev);
3271         const struct net_device_ops *slave_ops;
3272         struct neigh_parms parms;
3273         struct slave *slave;
3274         int ret;
3275
3276         slave = bond_first_slave(bond);
3277         if (!slave)
3278                 return 0;
3279         slave_ops = slave->dev->netdev_ops;
3280         if (!slave_ops->ndo_neigh_setup)
3281                 return 0;
3282
3283         parms.neigh_setup = NULL;
3284         parms.neigh_cleanup = NULL;
3285         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3286         if (ret)
3287                 return ret;
3288
3289         /*
3290          * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3291          * after the last slave has been detached.  Assumes that all slaves
3292          * utilize the same neigh_cleanup (true at this writing as only user
3293          * is ipoib).
3294          */
3295         n->parms->neigh_cleanup = parms.neigh_cleanup;
3296
3297         if (!parms.neigh_setup)
3298                 return 0;
3299
3300         return parms.neigh_setup(n);
3301 }
3302
3303 /*
3304  * The bonding ndo_neigh_setup is called at init time beofre any
3305  * slave exists. So we must declare proxy setup function which will
3306  * be used at run time to resolve the actual slave neigh param setup.
3307  *
3308  * It's also called by master devices (such as vlans) to setup their
3309  * underlying devices. In that case - do nothing, we're already set up from
3310  * our init.
3311  */
3312 static int bond_neigh_setup(struct net_device *dev,
3313                             struct neigh_parms *parms)
3314 {
3315         /* modify only our neigh_parms */
3316         if (parms->dev == dev)
3317                 parms->neigh_setup = bond_neigh_init;
3318
3319         return 0;
3320 }
3321
3322 /*
3323  * Change the MTU of all of a master's slaves to match the master
3324  */
3325 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3326 {
3327         struct bonding *bond = netdev_priv(bond_dev);
3328         struct slave *slave, *rollback_slave;
3329         struct list_head *iter;
3330         int res = 0;
3331
3332         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3333                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3334
3335         /* Can't hold bond->lock with bh disabled here since
3336          * some base drivers panic. On the other hand we can't
3337          * hold bond->lock without bh disabled because we'll
3338          * deadlock. The only solution is to rely on the fact
3339          * that we're under rtnl_lock here, and the slaves
3340          * list won't change. This doesn't solve the problem
3341          * of setting the slave's MTU while it is
3342          * transmitting, but the assumption is that the base
3343          * driver can handle that.
3344          *
3345          * TODO: figure out a way to safely iterate the slaves
3346          * list, but without holding a lock around the actual
3347          * call to the base driver.
3348          */
3349
3350         bond_for_each_slave(bond, slave, iter) {
3351                 pr_debug("s %p c_m %p\n",
3352                          slave,
3353                          slave->dev->netdev_ops->ndo_change_mtu);
3354
3355                 res = dev_set_mtu(slave->dev, new_mtu);
3356
3357                 if (res) {
3358                         /* If we failed to set the slave's mtu to the new value
3359                          * we must abort the operation even in ACTIVE_BACKUP
3360                          * mode, because if we allow the backup slaves to have
3361                          * different mtu values than the active slave we'll
3362                          * need to change their mtu when doing a failover. That
3363                          * means changing their mtu from timer context, which
3364                          * is probably not a good idea.
3365                          */
3366                         pr_debug("err %d %s\n", res, slave->dev->name);
3367                         goto unwind;
3368                 }
3369         }
3370
3371         bond_dev->mtu = new_mtu;
3372
3373         return 0;
3374
3375 unwind:
3376         /* unwind from head to the slave that failed */
3377         bond_for_each_slave(bond, rollback_slave, iter) {
3378                 int tmp_res;
3379
3380                 if (rollback_slave == slave)
3381                         break;
3382
3383                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3384                 if (tmp_res) {
3385                         pr_debug("unwind err %d dev %s\n",
3386                                  tmp_res, rollback_slave->dev->name);
3387                 }
3388         }
3389
3390         return res;
3391 }
3392
3393 /*
3394  * Change HW address
3395  *
3396  * Note that many devices must be down to change the HW address, and
3397  * downing the master releases all slaves.  We can make bonds full of
3398  * bonding devices to test this, however.
3399  */
3400 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3401 {
3402         struct bonding *bond = netdev_priv(bond_dev);
3403         struct slave *slave, *rollback_slave;
3404         struct sockaddr *sa = addr, tmp_sa;
3405         struct list_head *iter;
3406         int res = 0;
3407
3408         if (bond->params.mode == BOND_MODE_ALB)
3409                 return bond_alb_set_mac_address(bond_dev, addr);
3410
3411
3412         pr_debug("bond=%p, name=%s\n",
3413                  bond, bond_dev ? bond_dev->name : "None");
3414
3415         /* If fail_over_mac is enabled, do nothing and return success.
3416          * Returning an error causes ifenslave to fail.
3417          */
3418         if (bond->params.fail_over_mac)
3419                 return 0;
3420
3421         if (!is_valid_ether_addr(sa->sa_data))
3422                 return -EADDRNOTAVAIL;
3423
3424         /* Can't hold bond->lock with bh disabled here since
3425          * some base drivers panic. On the other hand we can't
3426          * hold bond->lock without bh disabled because we'll
3427          * deadlock. The only solution is to rely on the fact
3428          * that we're under rtnl_lock here, and the slaves
3429          * list won't change. This doesn't solve the problem
3430          * of setting the slave's hw address while it is
3431          * transmitting, but the assumption is that the base
3432          * driver can handle that.
3433          *
3434          * TODO: figure out a way to safely iterate the slaves
3435          * list, but without holding a lock around the actual
3436          * call to the base driver.
3437          */
3438
3439         bond_for_each_slave(bond, slave, iter) {
3440                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3441                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3442
3443                 if (slave_ops->ndo_set_mac_address == NULL) {
3444                         res = -EOPNOTSUPP;
3445                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3446                         goto unwind;
3447                 }
3448
3449                 res = dev_set_mac_address(slave->dev, addr);
3450                 if (res) {
3451                         /* TODO: consider downing the slave
3452                          * and retry ?
3453                          * User should expect communications
3454                          * breakage anyway until ARP finish
3455                          * updating, so...
3456                          */
3457                         pr_debug("err %d %s\n", res, slave->dev->name);
3458                         goto unwind;
3459                 }
3460         }
3461
3462         /* success */
3463         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3464         return 0;
3465
3466 unwind:
3467         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3468         tmp_sa.sa_family = bond_dev->type;
3469
3470         /* unwind from head to the slave that failed */
3471         bond_for_each_slave(bond, rollback_slave, iter) {
3472                 int tmp_res;
3473
3474                 if (rollback_slave == slave)
3475                         break;
3476
3477                 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3478                 if (tmp_res) {
3479                         pr_debug("unwind err %d dev %s\n",
3480                                  tmp_res, rollback_slave->dev->name);
3481                 }
3482         }
3483
3484         return res;
3485 }
3486
3487 /**
3488  * bond_xmit_slave_id - transmit skb through slave with slave_id
3489  * @bond: bonding device that is transmitting
3490  * @skb: buffer to transmit
3491  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3492  *
3493  * This function tries to transmit through slave with slave_id but in case
3494  * it fails, it tries to find the first available slave for transmission.
3495  * The skb is consumed in all cases, thus the function is void.
3496  */
3497 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3498 {
3499         struct list_head *iter;
3500         struct slave *slave;
3501         int i = slave_id;
3502
3503         /* Here we start from the slave with slave_id */
3504         bond_for_each_slave_rcu(bond, slave, iter) {
3505                 if (--i < 0) {
3506                         if (slave_can_tx(slave)) {
3507                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3508                                 return;
3509                         }
3510                 }
3511         }
3512
3513         /* Here we start from the first slave up to slave_id */
3514         i = slave_id;
3515         bond_for_each_slave_rcu(bond, slave, iter) {
3516                 if (--i < 0)
3517                         break;
3518                 if (slave_can_tx(slave)) {
3519                         bond_dev_queue_xmit(bond, skb, slave->dev);
3520                         return;
3521                 }
3522         }
3523         /* no slave that can tx has been found */
3524         kfree_skb(skb);
3525 }
3526
3527 /**
3528  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3529  * @bond: bonding device to use
3530  *
3531  * Based on the value of the bonding device's packets_per_slave parameter
3532  * this function generates a slave id, which is usually used as the next
3533  * slave to transmit through.
3534  */
3535 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3536 {
3537         u32 slave_id;
3538         struct reciprocal_value reciprocal_packets_per_slave;
3539         int packets_per_slave = bond->params.packets_per_slave;
3540
3541         switch (packets_per_slave) {
3542         case 0:
3543                 slave_id = prandom_u32();
3544                 break;
3545         case 1:
3546                 slave_id = bond->rr_tx_counter;
3547                 break;
3548         default:
3549                 reciprocal_packets_per_slave =
3550                         bond->params.reciprocal_packets_per_slave;
3551                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3552                                              reciprocal_packets_per_slave);
3553                 break;
3554         }
3555         bond->rr_tx_counter++;
3556
3557         return slave_id;
3558 }
3559
3560 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3561 {
3562         struct bonding *bond = netdev_priv(bond_dev);
3563         struct iphdr *iph = ip_hdr(skb);
3564         struct slave *slave;
3565         u32 slave_id;
3566
3567         /* Start with the curr_active_slave that joined the bond as the
3568          * default for sending IGMP traffic.  For failover purposes one
3569          * needs to maintain some consistency for the interface that will
3570          * send the join/membership reports.  The curr_active_slave found
3571          * will send all of this type of traffic.
3572          */
3573         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3574                 slave = rcu_dereference(bond->curr_active_slave);
3575                 if (slave && slave_can_tx(slave))
3576                         bond_dev_queue_xmit(bond, skb, slave->dev);
3577                 else
3578                         bond_xmit_slave_id(bond, skb, 0);
3579         } else {
3580                 slave_id = bond_rr_gen_slave_id(bond);
3581                 bond_xmit_slave_id(bond, skb, slave_id % bond->slave_cnt);
3582         }
3583
3584         return NETDEV_TX_OK;
3585 }
3586
3587 /*
3588  * in active-backup mode, we know that bond->curr_active_slave is always valid if
3589  * the bond has a usable interface.
3590  */
3591 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3592 {
3593         struct bonding *bond = netdev_priv(bond_dev);
3594         struct slave *slave;
3595
3596         slave = rcu_dereference(bond->curr_active_slave);
3597         if (slave)
3598                 bond_dev_queue_xmit(bond, skb, slave->dev);
3599         else
3600                 kfree_skb(skb);
3601
3602         return NETDEV_TX_OK;
3603 }
3604
3605 /* In bond_xmit_xor() , we determine the output device by using a pre-
3606  * determined xmit_hash_policy(), If the selected device is not enabled,
3607  * find the next active slave.
3608  */
3609 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3610 {
3611         struct bonding *bond = netdev_priv(bond_dev);
3612
3613         bond_xmit_slave_id(bond, skb, bond_xmit_hash(bond, skb, bond->slave_cnt));
3614
3615         return NETDEV_TX_OK;
3616 }
3617
3618 /* in broadcast mode, we send everything to all usable interfaces. */
3619 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3620 {
3621         struct bonding *bond = netdev_priv(bond_dev);
3622         struct slave *slave = NULL;
3623         struct list_head *iter;
3624
3625         bond_for_each_slave_rcu(bond, slave, iter) {
3626                 if (bond_is_last_slave(bond, slave))
3627                         break;
3628                 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3629                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3630
3631                         if (!skb2) {
3632                                 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3633                                        bond_dev->name);
3634                                 continue;
3635                         }
3636                         /* bond_dev_queue_xmit always returns 0 */
3637                         bond_dev_queue_xmit(bond, skb2, slave->dev);
3638                 }
3639         }
3640         if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3641                 bond_dev_queue_xmit(bond, skb, slave->dev);
3642         else
3643                 kfree_skb(skb);
3644
3645         return NETDEV_TX_OK;
3646 }
3647
3648 /*------------------------- Device initialization ---------------------------*/
3649
3650 /*
3651  * Lookup the slave that corresponds to a qid
3652  */
3653 static inline int bond_slave_override(struct bonding *bond,
3654                                       struct sk_buff *skb)
3655 {
3656         struct slave *slave = NULL;
3657         struct list_head *iter;
3658
3659         if (!skb->queue_mapping)
3660                 return 1;
3661
3662         /* Find out if any slaves have the same mapping as this skb. */
3663         bond_for_each_slave_rcu(bond, slave, iter) {
3664                 if (slave->queue_id == skb->queue_mapping) {
3665                         if (slave_can_tx(slave)) {
3666                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3667                                 return 0;
3668                         }
3669                         /* If the slave isn't UP, use default transmit policy. */
3670                         break;
3671                 }
3672         }
3673
3674         return 1;
3675 }
3676
3677
3678 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3679                              void *accel_priv)
3680 {
3681         /*
3682          * This helper function exists to help dev_pick_tx get the correct
3683          * destination queue.  Using a helper function skips a call to
3684          * skb_tx_hash and will put the skbs in the queue we expect on their
3685          * way down to the bonding driver.
3686          */
3687         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3688
3689         /*
3690          * Save the original txq to restore before passing to the driver
3691          */
3692         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3693
3694         if (unlikely(txq >= dev->real_num_tx_queues)) {
3695                 do {
3696                         txq -= dev->real_num_tx_queues;
3697                 } while (txq >= dev->real_num_tx_queues);
3698         }
3699         return txq;
3700 }
3701
3702 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3703 {
3704         struct bonding *bond = netdev_priv(dev);
3705
3706         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
3707                 if (!bond_slave_override(bond, skb))
3708                         return NETDEV_TX_OK;
3709         }
3710
3711         switch (bond->params.mode) {
3712         case BOND_MODE_ROUNDROBIN:
3713                 return bond_xmit_roundrobin(skb, dev);
3714         case BOND_MODE_ACTIVEBACKUP:
3715                 return bond_xmit_activebackup(skb, dev);
3716         case BOND_MODE_XOR:
3717                 return bond_xmit_xor(skb, dev);
3718         case BOND_MODE_BROADCAST:
3719                 return bond_xmit_broadcast(skb, dev);
3720         case BOND_MODE_8023AD:
3721                 return bond_3ad_xmit_xor(skb, dev);
3722         case BOND_MODE_ALB:
3723         case BOND_MODE_TLB:
3724                 return bond_alb_xmit(skb, dev);
3725         default:
3726                 /* Should never happen, mode already checked */
3727                 pr_err("%s: Error: Unknown bonding mode %d\n",
3728                        dev->name, bond->params.mode);
3729                 WARN_ON_ONCE(1);
3730                 kfree_skb(skb);
3731                 return NETDEV_TX_OK;
3732         }
3733 }
3734
3735 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3736 {
3737         struct bonding *bond = netdev_priv(dev);
3738         netdev_tx_t ret = NETDEV_TX_OK;
3739
3740         /*
3741          * If we risk deadlock from transmitting this in the
3742          * netpoll path, tell netpoll to queue the frame for later tx
3743          */
3744         if (is_netpoll_tx_blocked(dev))
3745                 return NETDEV_TX_BUSY;
3746
3747         rcu_read_lock();
3748         if (bond_has_slaves(bond))
3749                 ret = __bond_start_xmit(skb, dev);
3750         else
3751                 kfree_skb(skb);
3752         rcu_read_unlock();
3753
3754         return ret;
3755 }
3756
3757 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3758                                      struct ethtool_cmd *ecmd)
3759 {
3760         struct bonding *bond = netdev_priv(bond_dev);
3761         unsigned long speed = 0;
3762         struct list_head *iter;
3763         struct slave *slave;
3764
3765         ecmd->duplex = DUPLEX_UNKNOWN;
3766         ecmd->port = PORT_OTHER;
3767
3768         /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3769          * do not need to check mode.  Though link speed might not represent
3770          * the true receive or transmit bandwidth (not all modes are symmetric)
3771          * this is an accurate maximum.
3772          */
3773         read_lock(&bond->lock);
3774         bond_for_each_slave(bond, slave, iter) {
3775                 if (SLAVE_IS_OK(slave)) {
3776                         if (slave->speed != SPEED_UNKNOWN)
3777                                 speed += slave->speed;
3778                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
3779                             slave->duplex != DUPLEX_UNKNOWN)
3780                                 ecmd->duplex = slave->duplex;
3781                 }
3782         }
3783         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3784         read_unlock(&bond->lock);
3785
3786         return 0;
3787 }
3788
3789 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3790                                      struct ethtool_drvinfo *drvinfo)
3791 {
3792         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3793         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3794         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3795                  BOND_ABI_VERSION);
3796 }
3797
3798 static const struct ethtool_ops bond_ethtool_ops = {
3799         .get_drvinfo            = bond_ethtool_get_drvinfo,
3800         .get_settings           = bond_ethtool_get_settings,
3801         .get_link               = ethtool_op_get_link,
3802 };
3803
3804 static const struct net_device_ops bond_netdev_ops = {
3805         .ndo_init               = bond_init,
3806         .ndo_uninit             = bond_uninit,
3807         .ndo_open               = bond_open,
3808         .ndo_stop               = bond_close,
3809         .ndo_start_xmit         = bond_start_xmit,
3810         .ndo_select_queue       = bond_select_queue,
3811         .ndo_get_stats64        = bond_get_stats,
3812         .ndo_do_ioctl           = bond_do_ioctl,
3813         .ndo_change_rx_flags    = bond_change_rx_flags,
3814         .ndo_set_rx_mode        = bond_set_rx_mode,
3815         .ndo_change_mtu         = bond_change_mtu,
3816         .ndo_set_mac_address    = bond_set_mac_address,
3817         .ndo_neigh_setup        = bond_neigh_setup,
3818         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
3819         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
3820 #ifdef CONFIG_NET_POLL_CONTROLLER
3821         .ndo_netpoll_setup      = bond_netpoll_setup,
3822         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
3823         .ndo_poll_controller    = bond_poll_controller,
3824 #endif
3825         .ndo_add_slave          = bond_enslave,
3826         .ndo_del_slave          = bond_release,
3827         .ndo_fix_features       = bond_fix_features,
3828 };
3829
3830 static const struct device_type bond_type = {
3831         .name = "bond",
3832 };
3833
3834 static void bond_destructor(struct net_device *bond_dev)
3835 {
3836         struct bonding *bond = netdev_priv(bond_dev);
3837         if (bond->wq)
3838                 destroy_workqueue(bond->wq);
3839         free_netdev(bond_dev);
3840 }
3841
3842 void bond_setup(struct net_device *bond_dev)
3843 {
3844         struct bonding *bond = netdev_priv(bond_dev);
3845
3846         /* initialize rwlocks */
3847         rwlock_init(&bond->lock);
3848         rwlock_init(&bond->curr_slave_lock);
3849         bond->params = bonding_defaults;
3850
3851         /* Initialize pointers */
3852         bond->dev = bond_dev;
3853
3854         /* Initialize the device entry points */
3855         ether_setup(bond_dev);
3856         bond_dev->netdev_ops = &bond_netdev_ops;
3857         bond_dev->ethtool_ops = &bond_ethtool_ops;
3858
3859         bond_dev->destructor = bond_destructor;
3860
3861         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
3862
3863         /* Initialize the device options */
3864         bond_dev->tx_queue_len = 0;
3865         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
3866         bond_dev->priv_flags |= IFF_BONDING;
3867         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
3868
3869         /* At first, we block adding VLANs. That's the only way to
3870          * prevent problems that occur when adding VLANs over an
3871          * empty bond. The block will be removed once non-challenged
3872          * slaves are enslaved.
3873          */
3874         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
3875
3876         /* don't acquire bond device's netif_tx_lock when
3877          * transmitting */
3878         bond_dev->features |= NETIF_F_LLTX;
3879
3880         /* By default, we declare the bond to be fully
3881          * VLAN hardware accelerated capable. Special
3882          * care is taken in the various xmit functions
3883          * when there are slaves that are not hw accel
3884          * capable
3885          */
3886
3887         /* Don't allow bond devices to change network namespaces. */
3888         bond_dev->features |= NETIF_F_NETNS_LOCAL;
3889
3890         bond_dev->hw_features = BOND_VLAN_FEATURES |
3891                                 NETIF_F_HW_VLAN_CTAG_TX |
3892                                 NETIF_F_HW_VLAN_CTAG_RX |
3893                                 NETIF_F_HW_VLAN_CTAG_FILTER;
3894
3895         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
3896         bond_dev->features |= bond_dev->hw_features;
3897 }
3898
3899 /*
3900 * Destroy a bonding device.
3901 * Must be under rtnl_lock when this function is called.
3902 */
3903 static void bond_uninit(struct net_device *bond_dev)
3904 {
3905         struct bonding *bond = netdev_priv(bond_dev);
3906         struct list_head *iter;
3907         struct slave *slave;
3908
3909         bond_netpoll_cleanup(bond_dev);
3910
3911         /* Release the bonded slaves */
3912         bond_for_each_slave(bond, slave, iter)
3913                 __bond_release_one(bond_dev, slave->dev, true);
3914         pr_info("%s: released all slaves\n", bond_dev->name);
3915
3916         list_del(&bond->bond_list);
3917
3918         bond_debug_unregister(bond);
3919 }
3920
3921 /*------------------------- Module initialization ---------------------------*/
3922
3923 int bond_parm_tbl_lookup(int mode, const struct bond_parm_tbl *tbl)
3924 {
3925         int i;
3926
3927         for (i = 0; tbl[i].modename; i++)
3928                 if (mode == tbl[i].mode)
3929                         return tbl[i].mode;
3930
3931         return -1;
3932 }
3933
3934 static int bond_parm_tbl_lookup_name(const char *modename,
3935                                      const struct bond_parm_tbl *tbl)
3936 {
3937         int i;
3938
3939         for (i = 0; tbl[i].modename; i++)
3940                 if (strcmp(modename, tbl[i].modename) == 0)
3941                         return tbl[i].mode;
3942
3943         return -1;
3944 }
3945
3946 /*
3947  * Convert string input module parms.  Accept either the
3948  * number of the mode or its string name.  A bit complicated because
3949  * some mode names are substrings of other names, and calls from sysfs
3950  * may have whitespace in the name (trailing newlines, for example).
3951  */
3952 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
3953 {
3954         int modeint;
3955         char *p, modestr[BOND_MAX_MODENAME_LEN + 1];
3956
3957         for (p = (char *)buf; *p; p++)
3958                 if (!(isdigit(*p) || isspace(*p)))
3959                         break;
3960
3961         if (*p && sscanf(buf, "%20s", modestr) != 0)
3962                 return bond_parm_tbl_lookup_name(modestr, tbl);
3963         else if (sscanf(buf, "%d", &modeint) != 0)
3964                 return bond_parm_tbl_lookup(modeint, tbl);
3965
3966         return -1;
3967 }
3968
3969 static int bond_check_params(struct bond_params *params)
3970 {
3971         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
3972         struct bond_opt_value newval, *valptr;
3973         int arp_all_targets_value;
3974
3975         /*
3976          * Convert string parameters.
3977          */
3978         if (mode) {
3979                 bond_opt_initstr(&newval, mode);
3980                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
3981                 if (!valptr) {
3982                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
3983                         return -EINVAL;
3984                 }
3985                 bond_mode = valptr->value;
3986         }
3987
3988         if (xmit_hash_policy) {
3989                 if ((bond_mode != BOND_MODE_XOR) &&
3990                     (bond_mode != BOND_MODE_8023AD)) {
3991                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
3992                                bond_mode_name(bond_mode));
3993                 } else {
3994                         bond_opt_initstr(&newval, xmit_hash_policy);
3995                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
3996                                                 &newval);
3997                         if (!valptr) {
3998                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
3999                                        xmit_hash_policy);
4000                                 return -EINVAL;
4001                         }
4002                         xmit_hashtype = valptr->value;
4003                 }
4004         }
4005
4006         if (lacp_rate) {
4007                 if (bond_mode != BOND_MODE_8023AD) {
4008                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4009                                 bond_mode_name(bond_mode));
4010                 } else {
4011                         bond_opt_initstr(&newval, lacp_rate);
4012                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4013                                                 &newval);
4014                         if (!valptr) {
4015                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4016                                        lacp_rate);
4017                                 return -EINVAL;
4018                         }
4019                         lacp_fast = valptr->value;
4020                 }
4021         }
4022
4023         if (ad_select) {
4024                 bond_opt_initstr(&newval, lacp_rate);
4025                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4026                                         &newval);
4027                 if (!valptr) {
4028                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4029                         return -EINVAL;
4030                 }
4031                 params->ad_select = valptr->value;
4032                 if (bond_mode != BOND_MODE_8023AD)
4033                         pr_warning("ad_select param only affects 802.3ad mode\n");
4034         } else {
4035                 params->ad_select = BOND_AD_STABLE;
4036         }
4037
4038         if (max_bonds < 0) {
4039                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4040                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4041                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4042         }
4043
4044         if (miimon < 0) {
4045                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4046                            miimon, INT_MAX);
4047                 miimon = 0;
4048         }
4049
4050         if (updelay < 0) {
4051                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4052                            updelay, INT_MAX);
4053                 updelay = 0;
4054         }
4055
4056         if (downdelay < 0) {
4057                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4058                            downdelay, INT_MAX);
4059                 downdelay = 0;
4060         }
4061
4062         if ((use_carrier != 0) && (use_carrier != 1)) {
4063                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4064                            use_carrier);
4065                 use_carrier = 1;
4066         }
4067
4068         if (num_peer_notif < 0 || num_peer_notif > 255) {
4069                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4070                            num_peer_notif);
4071                 num_peer_notif = 1;
4072         }
4073
4074         /* reset values for 802.3ad/TLB/ALB */
4075         if (BOND_NO_USES_ARP(bond_mode)) {
4076                 if (!miimon) {
4077                         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");
4078                         pr_warning("Forcing miimon to 100msec\n");
4079                         miimon = BOND_DEFAULT_MIIMON;
4080                 }
4081         }
4082
4083         if (tx_queues < 1 || tx_queues > 255) {
4084                 pr_warning("Warning: tx_queues (%d) should be between "
4085                            "1 and 255, resetting to %d\n",
4086                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4087                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4088         }
4089
4090         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4091                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4092                            "not of valid value (0/1), so it was set to "
4093                            "0\n", all_slaves_active);
4094                 all_slaves_active = 0;
4095         }
4096
4097         if (resend_igmp < 0 || resend_igmp > 255) {
4098                 pr_warning("Warning: resend_igmp (%d) should be between "
4099                            "0 and 255, resetting to %d\n",
4100                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4101                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4102         }
4103
4104         bond_opt_initval(&newval, packets_per_slave);
4105         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4106                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4107                         packets_per_slave, USHRT_MAX);
4108                 packets_per_slave = 1;
4109         }
4110
4111         if (bond_mode == BOND_MODE_ALB) {
4112                 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",
4113                           updelay);
4114         }
4115
4116         if (!miimon) {
4117                 if (updelay || downdelay) {
4118                         /* just warn the user the up/down delay will have
4119                          * no effect since miimon is zero...
4120                          */
4121                         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",
4122                                    updelay, downdelay);
4123                 }
4124         } else {
4125                 /* don't allow arp monitoring */
4126                 if (arp_interval) {
4127                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4128                                    miimon, arp_interval);
4129                         arp_interval = 0;
4130                 }
4131
4132                 if ((updelay % miimon) != 0) {
4133                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4134                                    updelay, miimon,
4135                                    (updelay / miimon) * miimon);
4136                 }
4137
4138                 updelay /= miimon;
4139
4140                 if ((downdelay % miimon) != 0) {
4141                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4142                                    downdelay, miimon,
4143                                    (downdelay / miimon) * miimon);
4144                 }
4145
4146                 downdelay /= miimon;
4147         }
4148
4149         if (arp_interval < 0) {
4150                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to 0\n",
4151                            arp_interval, INT_MAX);
4152                 arp_interval = 0;
4153         }
4154
4155         for (arp_ip_count = 0, i = 0;
4156              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4157                 /* not complete check, but should be good enough to
4158                    catch mistakes */
4159                 __be32 ip;
4160                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4161                     IS_IP_TARGET_UNUSABLE_ADDRESS(ip)) {
4162                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4163                                    arp_ip_target[i]);
4164                         arp_interval = 0;
4165                 } else {
4166                         if (bond_get_targets_ip(arp_target, ip) == -1)
4167                                 arp_target[arp_ip_count++] = ip;
4168                         else
4169                                 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4170                                            &ip);
4171                 }
4172         }
4173
4174         if (arp_interval && !arp_ip_count) {
4175                 /* don't allow arping if no arp_ip_target given... */
4176                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4177                            arp_interval);
4178                 arp_interval = 0;
4179         }
4180
4181         if (arp_validate) {
4182                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4183                         pr_err("arp_validate only supported in active-backup mode\n");
4184                         return -EINVAL;
4185                 }
4186                 if (!arp_interval) {
4187                         pr_err("arp_validate requires arp_interval\n");
4188                         return -EINVAL;
4189                 }
4190
4191                 bond_opt_initstr(&newval, arp_validate);
4192                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4193                                         &newval);
4194                 if (!valptr) {
4195                         pr_err("Error: invalid arp_validate \"%s\"\n",
4196                                arp_validate);
4197                         return -EINVAL;
4198                 }
4199                 arp_validate_value = valptr->value;
4200         } else {
4201                 arp_validate_value = 0;
4202         }
4203
4204         arp_all_targets_value = 0;
4205         if (arp_all_targets) {
4206                 bond_opt_initstr(&newval, arp_all_targets);
4207                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4208                                         &newval);
4209                 if (!valptr) {
4210                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4211                                arp_all_targets);
4212                         arp_all_targets_value = 0;
4213                 } else {
4214                         arp_all_targets_value = valptr->value;
4215                 }
4216         }
4217
4218         if (miimon) {
4219                 pr_info("MII link monitoring set to %d ms\n", miimon);
4220         } else if (arp_interval) {
4221                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4222                                           arp_validate_value);
4223                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4224                         arp_interval, valptr->string, arp_ip_count);
4225
4226                 for (i = 0; i < arp_ip_count; i++)
4227                         pr_info(" %s", arp_ip_target[i]);
4228
4229                 pr_info("\n");
4230
4231         } else if (max_bonds) {
4232                 /* miimon and arp_interval not set, we need one so things
4233                  * work as expected, see bonding.txt for details
4234                  */
4235                 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");
4236         }
4237
4238         if (primary && !USES_PRIMARY(bond_mode)) {
4239                 /* currently, using a primary only makes sense
4240                  * in active backup, TLB or ALB modes
4241                  */
4242                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4243                            primary, bond_mode_name(bond_mode));
4244                 primary = NULL;
4245         }
4246
4247         if (primary && primary_reselect) {
4248                 bond_opt_initstr(&newval, primary_reselect);
4249                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4250                                         &newval);
4251                 if (!valptr) {
4252                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4253                                primary_reselect);
4254                         return -EINVAL;
4255                 }
4256                 primary_reselect_value = valptr->value;
4257         } else {
4258                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4259         }
4260
4261         if (fail_over_mac) {
4262                 bond_opt_initstr(&newval, fail_over_mac);
4263                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4264                                         &newval);
4265                 if (!valptr) {
4266                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4267                                fail_over_mac);
4268                         return -EINVAL;
4269                 }
4270                 fail_over_mac_value = valptr->value;
4271                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4272                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4273         } else {
4274                 fail_over_mac_value = BOND_FOM_NONE;
4275         }
4276
4277         if (lp_interval == 0) {
4278                 pr_warning("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4279                            INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4280                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4281         }
4282
4283         /* fill params struct with the proper values */
4284         params->mode = bond_mode;
4285         params->xmit_policy = xmit_hashtype;
4286         params->miimon = miimon;
4287         params->num_peer_notif = num_peer_notif;
4288         params->arp_interval = arp_interval;
4289         params->arp_validate = arp_validate_value;
4290         params->arp_all_targets = arp_all_targets_value;
4291         params->updelay = updelay;
4292         params->downdelay = downdelay;
4293         params->use_carrier = use_carrier;
4294         params->lacp_fast = lacp_fast;
4295         params->primary[0] = 0;
4296         params->primary_reselect = primary_reselect_value;
4297         params->fail_over_mac = fail_over_mac_value;
4298         params->tx_queues = tx_queues;
4299         params->all_slaves_active = all_slaves_active;
4300         params->resend_igmp = resend_igmp;
4301         params->min_links = min_links;
4302         params->lp_interval = lp_interval;
4303         params->packets_per_slave = packets_per_slave;
4304         if (packets_per_slave > 0) {
4305                 params->reciprocal_packets_per_slave =
4306                         reciprocal_value(packets_per_slave);
4307         } else {
4308                 /* reciprocal_packets_per_slave is unused if
4309                  * packets_per_slave is 0 or 1, just initialize it
4310                  */
4311                 params->reciprocal_packets_per_slave =
4312                         (struct reciprocal_value) { 0 };
4313         }
4314
4315         if (primary) {
4316                 strncpy(params->primary, primary, IFNAMSIZ);
4317                 params->primary[IFNAMSIZ - 1] = 0;
4318         }
4319
4320         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4321
4322         return 0;
4323 }
4324
4325 static struct lock_class_key bonding_netdev_xmit_lock_key;
4326 static struct lock_class_key bonding_netdev_addr_lock_key;
4327 static struct lock_class_key bonding_tx_busylock_key;
4328
4329 static void bond_set_lockdep_class_one(struct net_device *dev,
4330                                        struct netdev_queue *txq,
4331                                        void *_unused)
4332 {
4333         lockdep_set_class(&txq->_xmit_lock,
4334                           &bonding_netdev_xmit_lock_key);
4335 }
4336
4337 static void bond_set_lockdep_class(struct net_device *dev)
4338 {
4339         lockdep_set_class(&dev->addr_list_lock,
4340                           &bonding_netdev_addr_lock_key);
4341         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4342         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4343 }
4344
4345 /*
4346  * Called from registration process
4347  */
4348 static int bond_init(struct net_device *bond_dev)
4349 {
4350         struct bonding *bond = netdev_priv(bond_dev);
4351         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4352         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4353
4354         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4355
4356         /*
4357          * Initialize locks that may be required during
4358          * en/deslave operations.  All of the bond_open work
4359          * (of which this is part) should really be moved to
4360          * a phase prior to dev_open
4361          */
4362         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4363         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4364
4365         bond->wq = create_singlethread_workqueue(bond_dev->name);
4366         if (!bond->wq)
4367                 return -ENOMEM;
4368
4369         bond_set_lockdep_class(bond_dev);
4370
4371         list_add_tail(&bond->bond_list, &bn->dev_list);
4372
4373         bond_prepare_sysfs_group(bond);
4374
4375         bond_debug_register(bond);
4376
4377         /* Ensure valid dev_addr */
4378         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4379             bond_dev->addr_assign_type == NET_ADDR_PERM)
4380                 eth_hw_addr_random(bond_dev);
4381
4382         return 0;
4383 }
4384
4385 unsigned int bond_get_num_tx_queues(void)
4386 {
4387         return tx_queues;
4388 }
4389
4390 /* Create a new bond based on the specified name and bonding parameters.
4391  * If name is NULL, obtain a suitable "bond%d" name for us.
4392  * Caller must NOT hold rtnl_lock; we need to release it here before we
4393  * set up our sysfs entries.
4394  */
4395 int bond_create(struct net *net, const char *name)
4396 {
4397         struct net_device *bond_dev;
4398         int res;
4399
4400         rtnl_lock();
4401
4402         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4403                                    name ? name : "bond%d",
4404                                    bond_setup, tx_queues);
4405         if (!bond_dev) {
4406                 pr_err("%s: eek! can't alloc netdev!\n", name);
4407                 rtnl_unlock();
4408                 return -ENOMEM;
4409         }
4410
4411         dev_net_set(bond_dev, net);
4412         bond_dev->rtnl_link_ops = &bond_link_ops;
4413
4414         res = register_netdevice(bond_dev);
4415
4416         netif_carrier_off(bond_dev);
4417
4418         rtnl_unlock();
4419         if (res < 0)
4420                 bond_destructor(bond_dev);
4421         return res;
4422 }
4423
4424 static int __net_init bond_net_init(struct net *net)
4425 {
4426         struct bond_net *bn = net_generic(net, bond_net_id);
4427
4428         bn->net = net;
4429         INIT_LIST_HEAD(&bn->dev_list);
4430
4431         bond_create_proc_dir(bn);
4432         bond_create_sysfs(bn);
4433
4434         return 0;
4435 }
4436
4437 static void __net_exit bond_net_exit(struct net *net)
4438 {
4439         struct bond_net *bn = net_generic(net, bond_net_id);
4440         struct bonding *bond, *tmp_bond;
4441         LIST_HEAD(list);
4442
4443         bond_destroy_sysfs(bn);
4444         bond_destroy_proc_dir(bn);
4445
4446         /* Kill off any bonds created after unregistering bond rtnl ops */
4447         rtnl_lock();
4448         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4449                 unregister_netdevice_queue(bond->dev, &list);
4450         unregister_netdevice_many(&list);
4451         rtnl_unlock();
4452 }
4453
4454 static struct pernet_operations bond_net_ops = {
4455         .init = bond_net_init,
4456         .exit = bond_net_exit,
4457         .id   = &bond_net_id,
4458         .size = sizeof(struct bond_net),
4459 };
4460
4461 static int __init bonding_init(void)
4462 {
4463         int i;
4464         int res;
4465
4466         pr_info("%s", bond_version);
4467
4468         res = bond_check_params(&bonding_defaults);
4469         if (res)
4470                 goto out;
4471
4472         res = register_pernet_subsys(&bond_net_ops);
4473         if (res)
4474                 goto out;
4475
4476         res = bond_netlink_init();
4477         if (res)
4478                 goto err_link;
4479
4480         bond_create_debugfs();
4481
4482         for (i = 0; i < max_bonds; i++) {
4483                 res = bond_create(&init_net, NULL);
4484                 if (res)
4485                         goto err;
4486         }
4487
4488         register_netdevice_notifier(&bond_netdev_notifier);
4489 out:
4490         return res;
4491 err:
4492         bond_netlink_fini();
4493 err_link:
4494         unregister_pernet_subsys(&bond_net_ops);
4495         goto out;
4496
4497 }
4498
4499 static void __exit bonding_exit(void)
4500 {
4501         unregister_netdevice_notifier(&bond_netdev_notifier);
4502
4503         bond_destroy_debugfs();
4504
4505         bond_netlink_fini();
4506         unregister_pernet_subsys(&bond_net_ops);
4507
4508 #ifdef CONFIG_NET_POLL_CONTROLLER
4509         /*
4510          * Make sure we don't have an imbalance on our netpoll blocking
4511          */
4512         WARN_ON(atomic_read(&netpoll_block_tx));
4513 #endif
4514 }
4515
4516 module_init(bonding_init);
4517 module_exit(bonding_exit);
4518 MODULE_LICENSE("GPL");
4519 MODULE_VERSION(DRV_VERSION);
4520 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4521 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");