kernel/drivers/net/bonding/bond_main.c

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 #include <linux/kernel.h>
  35 #include <linux/module.h>
  36 #include <linux/types.h>
  37 #include <linux/fcntl.h>
  38 #include <linux/interrupt.h>
  39 #include <linux/ptrace.h>
  40 #include <linux/ioport.h>
  41 #include <linux/in.h>
  42 #include <net/ip.h>
  43 #include <linux/ip.h>
  44 #include <linux/tcp.h>
  45 #include <linux/udp.h>
  46 #include <linux/slab.h>
  47 #include <linux/string.h>
  48 #include <linux/init.h>
  49 #include <linux/timer.h>
  50 #include <linux/socket.h>
  51 #include <linux/ctype.h>
  52 #include <linux/inet.h>
  53 #include <linux/bitops.h>
  54 #include <linux/io.h>
  55 #include <asm/dma.h>
  56 #include <linux/uaccess.h>
  57 #include <linux/errno.h>
  58 #include <linux/netdevice.h>
  59 #include <linux/inetdevice.h>
  60 #include <linux/igmp.h>
  61 #include <linux/etherdevice.h>
  62 #include <linux/skbuff.h>
  63 #include <net/sock.h>
  64 #include <linux/rtnetlink.h>
  65 #include <linux/smp.h>
  66 #include <linux/if_ether.h>
  67 #include <net/arp.h>
  68 #include <linux/mii.h>
  69 #include <linux/ethtool.h>
  70 #include <linux/if_vlan.h>
  71 #include <linux/if_bonding.h>
  72 #include <linux/jiffies.h>
  73 #include <linux/preempt.h>
  74 #include <net/route.h>
  75 #include <net/net_namespace.h>
  76 #include <net/netns/generic.h>
  77 #include <net/pkt_sched.h>
  78 #include <linux/rculist.h>
  79 #include <net/flow_dissector.h>
  80 #include <net/switchdev.h>
  81 #include <net/bonding.h>
  82 #include <net/bond_3ad.h>
  83 #include <net/bond_alb.h>
  84
  85 #include "bonding_priv.h"
  86
  87 /*---------------------------- Module parameters ----------------------------*/
  88
  89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  90
  91 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  92 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  93 static int num_peer_notif = 1;
  94 static int miimon;
  95 static int updelay;
  96 static int downdelay;
  97 static int use_carrier  = 1;
  98 static char *mode;
  99 static char *primary;
 100 static char *primary_reselect;
 101 static char *lacp_rate;
 102 static int min_links;
 103 static char *ad_select;
 104 static char *xmit_hash_policy;
 105 static int arp_interval;
 106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 107 static char *arp_validate;
 108 static char *arp_all_targets;
 109 static char *fail_over_mac;
 110 static int all_slaves_active;
 111 static struct bond_params bonding_defaults;
 112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 113 static int packets_per_slave = 1;
 114 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
 115
 116 module_param(max_bonds, int, 0);
 117 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 118 module_param(tx_queues, int, 0);
 119 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 120 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 121 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 122                                "failover event (alias of num_unsol_na)");
 123 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 124 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 125                                "failover event (alias of num_grat_arp)");
 126 module_param(miimon, int, 0);
 127 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 128 module_param(updelay, int, 0);
 129 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 130 module_param(downdelay, int, 0);
 131 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 132                             "in milliseconds");
 133 module_param(use_carrier, int, 0);
 134 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 135                               "0 for off, 1 for on (default)");
 136 module_param(mode, charp, 0);
 137 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 138                        "1 for active-backup, 2 for balance-xor, "
 139                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 140                        "6 for balance-alb");
 141 module_param(primary, charp, 0);
 142 MODULE_PARM_DESC(primary, "Primary network device to use");
 143 module_param(primary_reselect, charp, 0);
 144 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 145                                    "once it comes up; "
 146                                    "0 for always (default), "
 147                                    "1 for only if speed of primary is "
 148                                    "better, "
 149                                    "2 for only on active slave "
 150                                    "failure");
 151 module_param(lacp_rate, charp, 0);
 152 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 153                             "0 for slow, 1 for fast");
 154 module_param(ad_select, charp, 0);
 155 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 156                             "0 for stable (default), 1 for bandwidth, "
 157                             "2 for count");
 158 module_param(min_links, int, 0);
 159 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 160
 161 module_param(xmit_hash_policy, charp, 0);
 162 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 163                                    "0 for layer 2 (default), 1 for layer 3+4, "
 164                                    "2 for layer 2+3, 3 for encap layer 2+3, "
 165                                    "4 for encap layer 3+4");
 166 module_param(arp_interval, int, 0);
 167 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 168 module_param_array(arp_ip_target, charp, NULL, 0);
 169 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 170 module_param(arp_validate, charp, 0);
 171 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 172                                "0 for none (default), 1 for active, "
 173                                "2 for backup, 3 for all");
 174 module_param(arp_all_targets, charp, 0);
 175 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 176 module_param(fail_over_mac, charp, 0);
 177 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 178                                 "the same MAC; 0 for none (default), "
 179                                 "1 for active, 2 for follow");
 180 module_param(all_slaves_active, int, 0);
 181 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
 182                                      "by setting active flag for all slaves; "
 183                                      "0 for never (default), 1 for always.");
 184 module_param(resend_igmp, int, 0);
 185 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 186                               "link failure");
 187 module_param(packets_per_slave, int, 0);
 188 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
 189                                     "mode; 0 for a random slave, 1 packet per "
 190                                     "slave (default), >1 packets per slave.");
 191 module_param(lp_interval, uint, 0);
 192 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
 193                               "the bonding driver sends learning packets to "
 194                               "each slaves peer switch. The default is 1.");
 195
 196 /*----------------------------- Global variables ----------------------------*/
 197
 198 #ifdef CONFIG_NET_POLL_CONTROLLER
 199 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 200 #endif
 201
 202 int bond_net_id __read_mostly;
 203
 204 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 205 static int arp_ip_count;
 206 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 207 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 208 static int lacp_fast;
 209
 210 /*-------------------------- Forward declarations ---------------------------*/
 211
 212 static int bond_init(struct net_device *bond_dev);
 213 static void bond_uninit(struct net_device *bond_dev);
 214 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
 215                                                 struct rtnl_link_stats64 *stats);
 216 static void bond_slave_arr_handler(struct work_struct *work);
 217 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
 218                                   int mod);
 219
 220 /*---------------------------- General routines -----------------------------*/
 221
 222 const char *bond_mode_name(int mode)
 223 {
 224         static const char *names[] = {
 225                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 226                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 227                 [BOND_MODE_XOR] = "load balancing (xor)",
 228                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 229                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 230                 [BOND_MODE_TLB] = "transmit load balancing",
 231                 [BOND_MODE_ALB] = "adaptive load balancing",
 232         };
 233
 234         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 235                 return "unknown";
 236
 237         return names[mode];
 238 }
 239
 240 /*---------------------------------- VLAN -----------------------------------*/
 241
 242 /**
 243  * bond_dev_queue_xmit - Prepare skb for xmit.
 244  *
 245  * @bond: bond device that got this skb for tx.
 246  * @skb: hw accel VLAN tagged skb to transmit
 247  * @slave_dev: slave that is supposed to xmit this skbuff
 248  */
 249 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 250                         struct net_device *slave_dev)
 251 {
 252         skb->dev = slave_dev;
 253
 254         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 255                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 256         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 257
 258         if (unlikely(netpoll_tx_running(bond->dev)))
 259                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 260         else
 261                 dev_queue_xmit(skb);
 262 }
 263
 264 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 265  * We don't protect the slave list iteration with a lock because:
 266  * a. This operation is performed in IOCTL context,
 267  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 268  * c. Holding a lock with BH disabled while directly calling a base driver
 269  *    entry point is generally a BAD idea.
 270  *
 271  * The design of synchronization/protection for this operation in the 8021q
 272  * module is good for one or more VLAN devices over a single physical device
 273  * and cannot be extended for a teaming solution like bonding, so there is a
 274  * potential race condition here where a net device from the vlan group might
 275  * be referenced (either by a base driver or the 8021q code) while it is being
 276  * removed from the system. However, it turns out we're not making matters
 277  * worse, and if it works for regular VLAN usage it will work here too.
 278 */
 279
 280 /**
 281  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 282  * @bond_dev: bonding net device that got called
 283  * @vid: vlan id being added
 284  */
 285 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 286                                 __be16 proto, u16 vid)
 287 {
 288         struct bonding *bond = netdev_priv(bond_dev);
 289         struct slave *slave, *rollback_slave;
 290         struct list_head *iter;
 291         int res;
 292
 293         bond_for_each_slave(bond, slave, iter) {
 294                 res = vlan_vid_add(slave->dev, proto, vid);
 295                 if (res)
 296                         goto unwind;
 297         }
 298
 299         return 0;
 300
 301 unwind:
 302         /* unwind to the slave that failed */
 303         bond_for_each_slave(bond, rollback_slave, iter) {
 304                 if (rollback_slave == slave)
 305                         break;
 306
 307                 vlan_vid_del(rollback_slave->dev, proto, vid);
 308         }
 309
 310         return res;
 311 }
 312
 313 /**
 314  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 315  * @bond_dev: bonding net device that got called
 316  * @vid: vlan id being removed
 317  */
 318 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 319                                  __be16 proto, u16 vid)
 320 {
 321         struct bonding *bond = netdev_priv(bond_dev);
 322         struct list_head *iter;
 323         struct slave *slave;
 324
 325         bond_for_each_slave(bond, slave, iter)
 326                 vlan_vid_del(slave->dev, proto, vid);
 327
 328         if (bond_is_lb(bond))
 329                 bond_alb_clear_vlan(bond, vid);
 330
 331         return 0;
 332 }
 333
 334 /*------------------------------- Link status -------------------------------*/
 335
 336 /* Set the carrier state for the master according to the state of its
 337  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 338  * do special 802.3ad magic.
 339  *
 340  * Returns zero if carrier state does not change, nonzero if it does.
 341  */
 342 int bond_set_carrier(struct bonding *bond)
 343 {
 344         struct list_head *iter;
 345         struct slave *slave;
 346
 347         if (!bond_has_slaves(bond))
 348                 goto down;
 349
 350         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 351                 return bond_3ad_set_carrier(bond);
 352
 353         bond_for_each_slave(bond, slave, iter) {
 354                 if (slave->link == BOND_LINK_UP) {
 355                         if (!netif_carrier_ok(bond->dev)) {
 356                                 netif_carrier_on(bond->dev);
 357                                 return 1;
 358                         }
 359                         return 0;
 360                 }
 361         }
 362
 363 down:
 364         if (netif_carrier_ok(bond->dev)) {
 365                 netif_carrier_off(bond->dev);
 366                 return 1;
 367         }
 368         return 0;
 369 }
 370
 371 /* Get link speed and duplex from the slave's base driver
 372  * using ethtool. If for some reason the call fails or the
 373  * values are invalid, set speed and duplex to -1,
 374  * and return.
 375  */
 376 static void bond_update_speed_duplex(struct slave *slave)
 377 {
 378         struct net_device *slave_dev = slave->dev;
 379         struct ethtool_cmd ecmd;
 380         u32 slave_speed;
 381         int res;
 382
 383         slave->speed = SPEED_UNKNOWN;
 384         slave->duplex = DUPLEX_UNKNOWN;
 385
 386         res = __ethtool_get_settings(slave_dev, &ecmd);
 387         if (res < 0)
 388                 return;
 389
 390         slave_speed = ethtool_cmd_speed(&ecmd);
 391         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 392                 return;
 393
 394         switch (ecmd.duplex) {
 395         case DUPLEX_FULL:
 396         case DUPLEX_HALF:
 397                 break;
 398         default:
 399                 return;
 400         }
 401
 402         slave->speed = slave_speed;
 403         slave->duplex = ecmd.duplex;
 404
 405         return;
 406 }
 407
 408 const char *bond_slave_link_status(s8 link)
 409 {
 410         switch (link) {
 411         case BOND_LINK_UP:
 412                 return "up";
 413         case BOND_LINK_FAIL:
 414                 return "going down";
 415         case BOND_LINK_DOWN:
 416                 return "down";
 417         case BOND_LINK_BACK:
 418                 return "going back";
 419         default:
 420                 return "unknown";
 421         }
 422 }
 423
 424 /* if <dev> supports MII link status reporting, check its link status.
 425  *
 426  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 427  * depending upon the setting of the use_carrier parameter.
 428  *
 429  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 430  * can't tell and just pretend it is), or 0, meaning that the link is
 431  * down.
 432  *
 433  * If reporting is non-zero, instead of faking link up, return -1 if
 434  * both ETHTOOL and MII ioctls fail (meaning the device does not
 435  * support them).  If use_carrier is set, return whatever it says.
 436  * It'd be nice if there was a good way to tell if a driver supports
 437  * netif_carrier, but there really isn't.
 438  */
 439 static int bond_check_dev_link(struct bonding *bond,
 440                                struct net_device *slave_dev, int reporting)
 441 {
 442         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 443         int (*ioctl)(struct net_device *, struct ifreq *, int);
 444         struct ifreq ifr;
 445         struct mii_ioctl_data *mii;
 446
 447         if (!reporting && !netif_running(slave_dev))
 448                 return 0;
 449
 450         if (bond->params.use_carrier)
 451                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 452
 453         /* Try to get link status using Ethtool first. */
 454         if (slave_dev->ethtool_ops->get_link)
 455                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 456                         BMSR_LSTATUS : 0;
 457
 458         /* Ethtool can't be used, fallback to MII ioctls. */
 459         ioctl = slave_ops->ndo_do_ioctl;
 460         if (ioctl) {
 461                 /* TODO: set pointer to correct ioctl on a per team member
 462                  *       bases to make this more efficient. that is, once
 463                  *       we determine the correct ioctl, we will always
 464                  *       call it and not the others for that team
 465                  *       member.
 466                  */
 467
 468                 /* We cannot assume that SIOCGMIIPHY will also read a
 469                  * register; not all network drivers (e.g., e100)
 470                  * support that.
 471                  */
 472
 473                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 474                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 475                 mii = if_mii(&ifr);
 476                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 477                         mii->reg_num = MII_BMSR;
 478                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 479                                 return mii->val_out & BMSR_LSTATUS;
 480                 }
 481         }
 482
 483         /* If reporting, report that either there's no dev->do_ioctl,
 484          * or both SIOCGMIIREG and get_link failed (meaning that we
 485          * cannot report link status).  If not reporting, pretend
 486          * we're ok.
 487          */
 488         return reporting ? -1 : BMSR_LSTATUS;
 489 }
 490
 491 /*----------------------------- Multicast list ------------------------------*/
 492
 493 /* Push the promiscuity flag down to appropriate slaves */
 494 static int bond_set_promiscuity(struct bonding *bond, int inc)
 495 {
 496         struct list_head *iter;
 497         int err = 0;
 498
 499         if (bond_uses_primary(bond)) {
 500                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 501
 502                 if (curr_active)
 503                         err = dev_set_promiscuity(curr_active->dev, inc);
 504         } else {
 505                 struct slave *slave;
 506
 507                 bond_for_each_slave(bond, slave, iter) {
 508                         err = dev_set_promiscuity(slave->dev, inc);
 509                         if (err)
 510                                 return err;
 511                 }
 512         }
 513         return err;
 514 }
 515
 516 /* Push the allmulti flag down to all slaves */
 517 static int bond_set_allmulti(struct bonding *bond, int inc)
 518 {
 519         struct list_head *iter;
 520         int err = 0;
 521
 522         if (bond_uses_primary(bond)) {
 523                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 524
 525                 if (curr_active)
 526                         err = dev_set_allmulti(curr_active->dev, inc);
 527         } else {
 528                 struct slave *slave;
 529
 530                 bond_for_each_slave(bond, slave, iter) {
 531                         err = dev_set_allmulti(slave->dev, inc);
 532                         if (err)
 533                                 return err;
 534                 }
 535         }
 536         return err;
 537 }
 538
 539 /* Retrieve the list of registered multicast addresses for the bonding
 540  * device and retransmit an IGMP JOIN request to the current active
 541  * slave.
 542  */
 543 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 544 {
 545         struct bonding *bond = container_of(work, struct bonding,
 546                                             mcast_work.work);
 547
 548         if (!rtnl_trylock()) {
 549                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 550                 return;
 551         }
 552         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 553
 554         if (bond->igmp_retrans > 1) {
 555                 bond->igmp_retrans--;
 556                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 557         }
 558         rtnl_unlock();
 559 }
 560
 561 /* Flush bond's hardware addresses from slave */
 562 static void bond_hw_addr_flush(struct net_device *bond_dev,
 563                                struct net_device *slave_dev)
 564 {
 565         struct bonding *bond = netdev_priv(bond_dev);
 566
 567         dev_uc_unsync(slave_dev, bond_dev);
 568         dev_mc_unsync(slave_dev, bond_dev);
 569
 570         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 571                 /* del lacpdu mc addr from mc list */
 572                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 573
 574                 dev_mc_del(slave_dev, lacpdu_multicast);
 575         }
 576 }
 577
 578 /*--------------------------- Active slave change ---------------------------*/
 579
 580 /* Update the hardware address list and promisc/allmulti for the new and
 581  * old active slaves (if any).  Modes that are not using primary keep all
 582  * slaves up date at all times; only the modes that use primary need to call
 583  * this function to swap these settings during a failover.
 584  */
 585 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 586                               struct slave *old_active)
 587 {
 588         if (old_active) {
 589                 if (bond->dev->flags & IFF_PROMISC)
 590                         dev_set_promiscuity(old_active->dev, -1);
 591
 592                 if (bond->dev->flags & IFF_ALLMULTI)
 593                         dev_set_allmulti(old_active->dev, -1);
 594
 595                 bond_hw_addr_flush(bond->dev, old_active->dev);
 596         }
 597
 598         if (new_active) {
 599                 /* FIXME: Signal errors upstream. */
 600                 if (bond->dev->flags & IFF_PROMISC)
 601                         dev_set_promiscuity(new_active->dev, 1);
 602
 603                 if (bond->dev->flags & IFF_ALLMULTI)
 604                         dev_set_allmulti(new_active->dev, 1);
 605
 606                 netif_addr_lock_bh(bond->dev);
 607                 dev_uc_sync(new_active->dev, bond->dev);
 608                 dev_mc_sync(new_active->dev, bond->dev);
 609                 netif_addr_unlock_bh(bond->dev);
 610         }
 611 }
 612
 613 /**
 614  * bond_set_dev_addr - clone slave's address to bond
 615  * @bond_dev: bond net device
 616  * @slave_dev: slave net device
 617  *
 618  * Should be called with RTNL held.
 619  */
 620 static void bond_set_dev_addr(struct net_device *bond_dev,
 621                               struct net_device *slave_dev)
 622 {
 623         netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 624                    bond_dev, slave_dev, slave_dev->addr_len);
 625         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 626         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 627         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 628 }
 629
 630 static struct slave *bond_get_old_active(struct bonding *bond,
 631                                          struct slave *new_active)
 632 {
 633         struct slave *slave;
 634         struct list_head *iter;
 635
 636         bond_for_each_slave(bond, slave, iter) {
 637                 if (slave == new_active)
 638                         continue;
 639
 640                 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
 641                         return slave;
 642         }
 643
 644         return NULL;
 645 }
 646
 647 /* bond_do_fail_over_mac
 648  *
 649  * Perform special MAC address swapping for fail_over_mac settings
 650  *
 651  * Called with RTNL
 652  */
 653 static void bond_do_fail_over_mac(struct bonding *bond,
 654                                   struct slave *new_active,
 655                                   struct slave *old_active)
 656 {
 657         u8 tmp_mac[ETH_ALEN];
 658         struct sockaddr saddr;
 659         int rv;
 660
 661         switch (bond->params.fail_over_mac) {
 662         case BOND_FOM_ACTIVE:
 663                 if (new_active)
 664                         bond_set_dev_addr(bond->dev, new_active->dev);
 665                 break;
 666         case BOND_FOM_FOLLOW:
 667                 /* if new_active && old_active, swap them
 668                  * if just old_active, do nothing (going to no active slave)
 669                  * if just new_active, set new_active to bond's MAC
 670                  */
 671                 if (!new_active)
 672                         return;
 673
 674                 if (!old_active)
 675                         old_active = bond_get_old_active(bond, new_active);
 676
 677                 if (old_active) {
 678                         ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
 679                         ether_addr_copy(saddr.sa_data,
 680                                         old_active->dev->dev_addr);
 681                         saddr.sa_family = new_active->dev->type;
 682                 } else {
 683                         ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
 684                         saddr.sa_family = bond->dev->type;
 685                 }
 686
 687                 rv = dev_set_mac_address(new_active->dev, &saddr);
 688                 if (rv) {
 689                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 690                                    -rv, new_active->dev->name);
 691                         goto out;
 692                 }
 693
 694                 if (!old_active)
 695                         goto out;
 696
 697                 ether_addr_copy(saddr.sa_data, tmp_mac);
 698                 saddr.sa_family = old_active->dev->type;
 699
 700                 rv = dev_set_mac_address(old_active->dev, &saddr);
 701                 if (rv)
 702                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 703                                    -rv, new_active->dev->name);
 704 out:
 705                 break;
 706         default:
 707                 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
 708                            bond->params.fail_over_mac);
 709                 break;
 710         }
 711
 712 }
 713
 714 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
 715 {
 716         struct slave *prim = rtnl_dereference(bond->primary_slave);
 717         struct slave *curr = rtnl_dereference(bond->curr_active_slave);
 718
 719         if (!prim || prim->link != BOND_LINK_UP) {
 720                 if (!curr || curr->link != BOND_LINK_UP)
 721                         return NULL;
 722                 return curr;
 723         }
 724
 725         if (bond->force_primary) {
 726                 bond->force_primary = false;
 727                 return prim;
 728         }
 729
 730         if (!curr || curr->link != BOND_LINK_UP)
 731                 return prim;
 732
 733         /* At this point, prim and curr are both up */
 734         switch (bond->params.primary_reselect) {
 735         case BOND_PRI_RESELECT_ALWAYS:
 736                 return prim;
 737         case BOND_PRI_RESELECT_BETTER:
 738                 if (prim->speed < curr->speed)
 739                         return curr;
 740                 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
 741                         return curr;
 742                 return prim;
 743         case BOND_PRI_RESELECT_FAILURE:
 744                 return curr;
 745         default:
 746                 netdev_err(bond->dev, "impossible primary_reselect %d\n",
 747                            bond->params.primary_reselect);
 748                 return curr;
 749         }
 750 }
 751
 752 /**
 753  * bond_find_best_slave - select the best available slave to be the active one
 754  * @bond: our bonding struct
 755  */
 756 static struct slave *bond_find_best_slave(struct bonding *bond)
 757 {
 758         struct slave *slave, *bestslave = NULL;
 759         struct list_head *iter;
 760         int mintime = bond->params.updelay;
 761
 762         slave = bond_choose_primary_or_current(bond);
 763         if (slave)
 764                 return slave;
 765
 766         bond_for_each_slave(bond, slave, iter) {
 767                 if (slave->link == BOND_LINK_UP)
 768                         return slave;
 769                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
 770                     slave->delay < mintime) {
 771                         mintime = slave->delay;
 772                         bestslave = slave;
 773                 }
 774         }
 775
 776         return bestslave;
 777 }
 778
 779 static bool bond_should_notify_peers(struct bonding *bond)
 780 {
 781         struct slave *slave;
 782
 783         rcu_read_lock();
 784         slave = rcu_dereference(bond->curr_active_slave);
 785         rcu_read_unlock();
 786
 787         netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
 788                    slave ? slave->dev->name : "NULL");
 789
 790         if (!slave || !bond->send_peer_notif ||
 791             !netif_carrier_ok(bond->dev) ||
 792             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 793                 return false;
 794
 795         return true;
 796 }
 797
 798 /**
 799  * change_active_interface - change the active slave into the specified one
 800  * @bond: our bonding struct
 801  * @new: the new slave to make the active one
 802  *
 803  * Set the new slave to the bond's settings and unset them on the old
 804  * curr_active_slave.
 805  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 806  *
 807  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 808  * because it is apparently the best available slave we have, even though its
 809  * updelay hasn't timed out yet.
 810  *
 811  * Caller must hold RTNL.
 812  */
 813 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 814 {
 815         struct slave *old_active;
 816
 817         ASSERT_RTNL();
 818
 819         old_active = rtnl_dereference(bond->curr_active_slave);
 820
 821         if (old_active == new_active)
 822                 return;
 823
 824         if (new_active) {
 825                 new_active->last_link_up = jiffies;
 826
 827                 if (new_active->link == BOND_LINK_BACK) {
 828                         if (bond_uses_primary(bond)) {
 829                                 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
 830                                             new_active->dev->name,
 831                                             (bond->params.updelay - new_active->delay) * bond->params.miimon);
 832                         }
 833
 834                         new_active->delay = 0;
 835                         bond_set_slave_link_state(new_active, BOND_LINK_UP);
 836
 837                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 838                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 839
 840                         if (bond_is_lb(bond))
 841                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 842                 } else {
 843                         if (bond_uses_primary(bond)) {
 844                                 netdev_info(bond->dev, "making interface %s the new active one\n",
 845                                             new_active->dev->name);
 846                         }
 847                 }
 848         }
 849
 850         if (bond_uses_primary(bond))
 851                 bond_hw_addr_swap(bond, new_active, old_active);
 852
 853         if (bond_is_lb(bond)) {
 854                 bond_alb_handle_active_change(bond, new_active);
 855                 if (old_active)
 856                         bond_set_slave_inactive_flags(old_active,
 857                                                       BOND_SLAVE_NOTIFY_NOW);
 858                 if (new_active)
 859                         bond_set_slave_active_flags(new_active,
 860                                                     BOND_SLAVE_NOTIFY_NOW);
 861         } else {
 862                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 863         }
 864
 865         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
 866                 if (old_active)
 867                         bond_set_slave_inactive_flags(old_active,
 868                                                       BOND_SLAVE_NOTIFY_NOW);
 869
 870                 if (new_active) {
 871                         bool should_notify_peers = false;
 872
 873                         bond_set_slave_active_flags(new_active,
 874                                                     BOND_SLAVE_NOTIFY_NOW);
 875
 876                         if (bond->params.fail_over_mac)
 877                                 bond_do_fail_over_mac(bond, new_active,
 878                                                       old_active);
 879
 880                         if (netif_running(bond->dev)) {
 881                                 bond->send_peer_notif =
 882                                         bond->params.num_peer_notif;
 883                                 should_notify_peers =
 884                                         bond_should_notify_peers(bond);
 885                         }
 886
 887                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 888                         if (should_notify_peers)
 889                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 890                                                          bond->dev);
 891                 }
 892         }
 893
 894         /* resend IGMP joins since active slave has changed or
 895          * all were sent on curr_active_slave.
 896          * resend only if bond is brought up with the affected
 897          * bonding modes and the retransmission is enabled
 898          */
 899         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 900             ((bond_uses_primary(bond) && new_active) ||
 901              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
 902                 bond->igmp_retrans = bond->params.resend_igmp;
 903                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 904         }
 905 }
 906
 907 /**
 908  * bond_select_active_slave - select a new active slave, if needed
 909  * @bond: our bonding struct
 910  *
 911  * This functions should be called when one of the following occurs:
 912  * - The old curr_active_slave has been released or lost its link.
 913  * - The primary_slave has got its link back.
 914  * - A slave has got its link back and there's no old curr_active_slave.
 915  *
 916  * Caller must hold RTNL.
 917  */
 918 void bond_select_active_slave(struct bonding *bond)
 919 {
 920         struct slave *best_slave;
 921         int rv;
 922
 923         ASSERT_RTNL();
 924
 925         best_slave = bond_find_best_slave(bond);
 926         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
 927                 bond_change_active_slave(bond, best_slave);
 928                 rv = bond_set_carrier(bond);
 929                 if (!rv)
 930                         return;
 931
 932                 if (netif_carrier_ok(bond->dev)) {
 933                         netdev_info(bond->dev, "first active interface up!\n");
 934                 } else {
 935                         netdev_info(bond->dev, "now running without any active interface!\n");
 936                 }
 937         }
 938 }
 939
 940 #ifdef CONFIG_NET_POLL_CONTROLLER
 941 static inline int slave_enable_netpoll(struct slave *slave)
 942 {
 943         struct netpoll *np;
 944         int err = 0;
 945
 946         np = kzalloc(sizeof(*np), GFP_KERNEL);
 947         err = -ENOMEM;
 948         if (!np)
 949                 goto out;
 950
 951         err = __netpoll_setup(np, slave->dev);
 952         if (err) {
 953                 kfree(np);
 954                 goto out;
 955         }
 956         slave->np = np;
 957 out:
 958         return err;
 959 }
 960 static inline void slave_disable_netpoll(struct slave *slave)
 961 {
 962         struct netpoll *np = slave->np;
 963
 964         if (!np)
 965                 return;
 966
 967         slave->np = NULL;
 968         __netpoll_free_async(np);
 969 }
 970
 971 static void bond_poll_controller(struct net_device *bond_dev)
 972 {
 973         struct bonding *bond = netdev_priv(bond_dev);
 974         struct slave *slave = NULL;
 975         struct list_head *iter;
 976         struct ad_info ad_info;
 977         struct netpoll_info *ni;
 978         const struct net_device_ops *ops;
 979
 980         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 981                 if (bond_3ad_get_active_agg_info(bond, &ad_info))
 982                         return;
 983
 984         bond_for_each_slave_rcu(bond, slave, iter) {
 985                 ops = slave->dev->netdev_ops;
 986                 if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller)
 987                         continue;
 988
 989                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 990                         struct aggregator *agg =
 991                             SLAVE_AD_INFO(slave)->port.aggregator;
 992
 993                         if (agg &&
 994                             agg->aggregator_identifier != ad_info.aggregator_id)
 995                                 continue;
 996                 }
 997
 998                 ni = rcu_dereference_bh(slave->dev->npinfo);
 999                 if (down_trylock(&ni->dev_lock))
1000                         continue;
1001                 ops->ndo_poll_controller(slave->dev);
1002                 up(&ni->dev_lock);
1003         }
1004 }
1005
1006 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1007 {
1008         struct bonding *bond = netdev_priv(bond_dev);
1009         struct list_head *iter;
1010         struct slave *slave;
1011
1012         bond_for_each_slave(bond, slave, iter)
1013                 if (bond_slave_is_up(slave))
1014                         slave_disable_netpoll(slave);
1015 }
1016
1017 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1018 {
1019         struct bonding *bond = netdev_priv(dev);
1020         struct list_head *iter;
1021         struct slave *slave;
1022         int err = 0;
1023
1024         bond_for_each_slave(bond, slave, iter) {
1025                 err = slave_enable_netpoll(slave);
1026                 if (err) {
1027                         bond_netpoll_cleanup(dev);
1028                         break;
1029                 }
1030         }
1031         return err;
1032 }
1033 #else
1034 static inline int slave_enable_netpoll(struct slave *slave)
1035 {
1036         return 0;
1037 }
1038 static inline void slave_disable_netpoll(struct slave *slave)
1039 {
1040 }
1041 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1042 {
1043 }
1044 #endif
1045
1046 /*---------------------------------- IOCTL ----------------------------------*/
1047
1048 static netdev_features_t bond_fix_features(struct net_device *dev,
1049                                            netdev_features_t features)
1050 {
1051         struct bonding *bond = netdev_priv(dev);
1052         struct list_head *iter;
1053         netdev_features_t mask;
1054         struct slave *slave;
1055
1056         mask = features;
1057
1058         features &= ~NETIF_F_ONE_FOR_ALL;
1059         features |= NETIF_F_ALL_FOR_ALL;
1060
1061         bond_for_each_slave(bond, slave, iter) {
1062                 features = netdev_increment_features(features,
1063                                                      slave->dev->features,
1064                                                      mask);
1065         }
1066         features = netdev_add_tso_features(features, mask);
1067
1068         return features;
1069 }
1070
1071 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1072                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1073                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1074
1075 #define BOND_ENC_FEATURES       (NETIF_F_ALL_CSUM | NETIF_F_SG | NETIF_F_RXCSUM |\
1076                                  NETIF_F_ALL_TSO)
1077
1078 static void bond_compute_features(struct bonding *bond)
1079 {
1080         unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1081                                         IFF_XMIT_DST_RELEASE_PERM;
1082         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1083         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1084         struct net_device *bond_dev = bond->dev;
1085         struct list_head *iter;
1086         struct slave *slave;
1087         unsigned short max_hard_header_len = ETH_HLEN;
1088         unsigned int gso_max_size = GSO_MAX_SIZE;
1089         u16 gso_max_segs = GSO_MAX_SEGS;
1090
1091         if (!bond_has_slaves(bond))
1092                 goto done;
1093         vlan_features &= NETIF_F_ALL_FOR_ALL;
1094
1095         bond_for_each_slave(bond, slave, iter) {
1096                 vlan_features = netdev_increment_features(vlan_features,
1097                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1098
1099                 enc_features = netdev_increment_features(enc_features,
1100                                                          slave->dev->hw_enc_features,
1101                                                          BOND_ENC_FEATURES);
1102                 dst_release_flag &= slave->dev->priv_flags;
1103                 if (slave->dev->hard_header_len > max_hard_header_len)
1104                         max_hard_header_len = slave->dev->hard_header_len;
1105
1106                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1107                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1108         }
1109
1110 done:
1111         bond_dev->vlan_features = vlan_features;
1112         bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
1113         bond_dev->hard_header_len = max_hard_header_len;
1114         bond_dev->gso_max_segs = gso_max_segs;
1115         netif_set_gso_max_size(bond_dev, gso_max_size);
1116
1117         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1118         if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1119             dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1120                 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1121
1122         netdev_change_features(bond_dev);
1123 }
1124
1125 static void bond_setup_by_slave(struct net_device *bond_dev,
1126                                 struct net_device *slave_dev)
1127 {
1128         bond_dev->header_ops        = slave_dev->header_ops;
1129
1130         bond_dev->type              = slave_dev->type;
1131         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1132         bond_dev->addr_len          = slave_dev->addr_len;
1133
1134         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1135                 slave_dev->addr_len);
1136 }
1137
1138 /* On bonding slaves other than the currently active slave, suppress
1139  * duplicates except for alb non-mcast/bcast.
1140  */
1141 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1142                                             struct slave *slave,
1143                                             struct bonding *bond)
1144 {
1145         if (bond_is_slave_inactive(slave)) {
1146                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1147                     skb->pkt_type != PACKET_BROADCAST &&
1148                     skb->pkt_type != PACKET_MULTICAST)
1149                         return false;
1150                 return true;
1151         }
1152         return false;
1153 }
1154
1155 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1156 {
1157         struct sk_buff *skb = *pskb;
1158         struct slave *slave;
1159         struct bonding *bond;
1160         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1161                           struct slave *);
1162         int ret = RX_HANDLER_ANOTHER;
1163
1164         skb = skb_share_check(skb, GFP_ATOMIC);
1165         if (unlikely(!skb))
1166                 return RX_HANDLER_CONSUMED;
1167
1168         *pskb = skb;
1169
1170         slave = bond_slave_get_rcu(skb->dev);
1171         bond = slave->bond;
1172
1173         recv_probe = ACCESS_ONCE(bond->recv_probe);
1174         if (recv_probe) {
1175                 ret = recv_probe(skb, bond, slave);
1176                 if (ret == RX_HANDLER_CONSUMED) {
1177                         consume_skb(skb);
1178                         return ret;
1179                 }
1180         }
1181
1182         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1183                 return RX_HANDLER_EXACT;
1184         }
1185
1186         skb->dev = bond->dev;
1187
1188         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1189             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1190             skb->pkt_type == PACKET_HOST) {
1191
1192                 if (unlikely(skb_cow_head(skb,
1193                                           skb->data - skb_mac_header(skb)))) {
1194                         kfree_skb(skb);
1195                         return RX_HANDLER_CONSUMED;
1196                 }
1197                 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1198         }
1199
1200         return ret;
1201 }
1202
1203 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1204                                       struct net_device *slave_dev,
1205                                       struct slave *slave)
1206 {
1207         int err;
1208
1209         err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1210         if (err)
1211                 return err;
1212         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1213         return 0;
1214 }
1215
1216 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1217                                   struct net_device *slave_dev)
1218 {
1219         netdev_upper_dev_unlink(slave_dev, bond_dev);
1220         slave_dev->flags &= ~IFF_SLAVE;
1221         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1222 }
1223
1224 static struct slave *bond_alloc_slave(struct bonding *bond)
1225 {
1226         struct slave *slave = NULL;
1227
1228         slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1229         if (!slave)
1230                 return NULL;
1231
1232         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1233                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1234                                                GFP_KERNEL);
1235                 if (!SLAVE_AD_INFO(slave)) {
1236                         kfree(slave);
1237                         return NULL;
1238                 }
1239         }
1240         return slave;
1241 }
1242
1243 static void bond_free_slave(struct slave *slave)
1244 {
1245         struct bonding *bond = bond_get_bond_by_slave(slave);
1246
1247         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1248                 kfree(SLAVE_AD_INFO(slave));
1249
1250         kfree(slave);
1251 }
1252
1253 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1254 {
1255         info->bond_mode = BOND_MODE(bond);
1256         info->miimon = bond->params.miimon;
1257         info->num_slaves = bond->slave_cnt;
1258 }
1259
1260 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1261 {
1262         strcpy(info->slave_name, slave->dev->name);
1263         info->link = slave->link;
1264         info->state = bond_slave_state(slave);
1265         info->link_failure_count = slave->link_failure_count;
1266 }
1267
1268 static void bond_netdev_notify(struct net_device *dev,
1269                                struct netdev_bonding_info *info)
1270 {
1271         rtnl_lock();
1272         netdev_bonding_info_change(dev, info);
1273         rtnl_unlock();
1274 }
1275
1276 static void bond_netdev_notify_work(struct work_struct *_work)
1277 {
1278         struct netdev_notify_work *w =
1279                 container_of(_work, struct netdev_notify_work, work.work);
1280
1281         bond_netdev_notify(w->dev, &w->bonding_info);
1282         dev_put(w->dev);
1283         kfree(w);
1284 }
1285
1286 void bond_queue_slave_event(struct slave *slave)
1287 {
1288         struct bonding *bond = slave->bond;
1289         struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1290
1291         if (!nnw)
1292                 return;
1293
1294         dev_hold(slave->dev);
1295         nnw->dev = slave->dev;
1296         bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1297         bond_fill_ifbond(bond, &nnw->bonding_info.master);
1298         INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1299
1300         queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1301 }
1302
1303 /* enslave device <slave> to bond device <master> */
1304 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1305 {
1306         struct bonding *bond = netdev_priv(bond_dev);
1307         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1308         struct slave *new_slave = NULL, *prev_slave;
1309         struct sockaddr addr;
1310         int link_reporting;
1311         int res = 0, i;
1312
1313         if (!bond->params.use_carrier &&
1314             slave_dev->ethtool_ops->get_link == NULL &&
1315             slave_ops->ndo_do_ioctl == NULL) {
1316                 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1317                             slave_dev->name);
1318         }
1319
1320         /* already in-use? */
1321         if (netdev_is_rx_handler_busy(slave_dev)) {
1322                 netdev_err(bond_dev,
1323                            "Error: Device is in use and cannot be enslaved\n");
1324                 return -EBUSY;
1325         }
1326
1327         if (bond_dev == slave_dev) {
1328                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1329                 return -EPERM;
1330         }
1331
1332         /* vlan challenged mutual exclusion */
1333         /* no need to lock since we're protected by rtnl_lock */
1334         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1335                 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1336                            slave_dev->name);
1337                 if (vlan_uses_dev(bond_dev)) {
1338                         netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1339                                    slave_dev->name, bond_dev->name);
1340                         return -EPERM;
1341                 } else {
1342                         netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1343                                     slave_dev->name, slave_dev->name,
1344                                     bond_dev->name);
1345                 }
1346         } else {
1347                 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1348                            slave_dev->name);
1349         }
1350
1351         /* Old ifenslave binaries are no longer supported.  These can
1352          * be identified with moderate accuracy by the state of the slave:
1353          * the current ifenslave will set the interface down prior to
1354          * enslaving it; the old ifenslave will not.
1355          */
1356         if ((slave_dev->flags & IFF_UP)) {
1357                 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1358                            slave_dev->name);
1359                 res = -EPERM;
1360                 goto err_undo_flags;
1361         }
1362
1363         /* set bonding device ether type by slave - bonding netdevices are
1364          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1365          * there is a need to override some of the type dependent attribs/funcs.
1366          *
1367          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1368          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1369          */
1370         if (!bond_has_slaves(bond)) {
1371                 if (bond_dev->type != slave_dev->type) {
1372                         netdev_dbg(bond_dev, "change device type from %d to %d\n",
1373                                    bond_dev->type, slave_dev->type);
1374
1375                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1376                                                        bond_dev);
1377                         res = notifier_to_errno(res);
1378                         if (res) {
1379                                 netdev_err(bond_dev, "refused to change device type\n");
1380                                 res = -EBUSY;
1381                                 goto err_undo_flags;
1382                         }
1383
1384                         /* Flush unicast and multicast addresses */
1385                         dev_uc_flush(bond_dev);
1386                         dev_mc_flush(bond_dev);
1387
1388                         if (slave_dev->type != ARPHRD_ETHER)
1389                                 bond_setup_by_slave(bond_dev, slave_dev);
1390                         else {
1391                                 ether_setup(bond_dev);
1392                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1393                         }
1394
1395                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1396                                                  bond_dev);
1397                 }
1398         } else if (bond_dev->type != slave_dev->type) {
1399                 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1400                            slave_dev->name, slave_dev->type, bond_dev->type);
1401                 res = -EINVAL;
1402                 goto err_undo_flags;
1403         }
1404
1405         if (slave_ops->ndo_set_mac_address == NULL) {
1406                 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1407                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1408                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1409                         if (!bond_has_slaves(bond)) {
1410                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1411                                 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1412                         } else {
1413                                 netdev_err(bond_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
1414                                 res = -EOPNOTSUPP;
1415                                 goto err_undo_flags;
1416                         }
1417                 }
1418         }
1419
1420         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1421
1422         /* If this is the first slave, then we need to set the master's hardware
1423          * address to be the same as the slave's.
1424          */
1425         if (!bond_has_slaves(bond) &&
1426             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1427                 bond_set_dev_addr(bond->dev, slave_dev);
1428
1429         new_slave = bond_alloc_slave(bond);
1430         if (!new_slave) {
1431                 res = -ENOMEM;
1432                 goto err_undo_flags;
1433         }
1434
1435         new_slave->bond = bond;
1436         new_slave->dev = slave_dev;
1437         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1438          * is set via sysfs or module option if desired.
1439          */
1440         new_slave->queue_id = 0;
1441
1442         /* Save slave's original mtu and then set it to match the bond */
1443         new_slave->original_mtu = slave_dev->mtu;
1444         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1445         if (res) {
1446                 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1447                 goto err_free;
1448         }
1449
1450         /* Save slave's original ("permanent") mac address for modes
1451          * that need it, and for restoring it upon release, and then
1452          * set it to the master's address
1453          */
1454         ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1455
1456         if (!bond->params.fail_over_mac ||
1457             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1458                 /* Set slave to master's mac address.  The application already
1459                  * set the master's mac address to that of the first slave
1460                  */
1461                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1462                 addr.sa_family = slave_dev->type;
1463                 res = dev_set_mac_address(slave_dev, &addr);
1464                 if (res) {
1465                         netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1466                         goto err_restore_mtu;
1467                 }
1468         }
1469
1470         /* set slave flag before open to prevent IPv6 addrconf */
1471         slave_dev->flags |= IFF_SLAVE;
1472
1473         /* open the slave since the application closed it */
1474         res = dev_open(slave_dev);
1475         if (res) {
1476                 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1477                 goto err_restore_mac;
1478         }
1479
1480         slave_dev->priv_flags |= IFF_BONDING;
1481         /* initialize slave stats */
1482         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1483
1484         if (bond_is_lb(bond)) {
1485                 /* bond_alb_init_slave() must be called before all other stages since
1486                  * it might fail and we do not want to have to undo everything
1487                  */
1488                 res = bond_alb_init_slave(bond, new_slave);
1489                 if (res)
1490                         goto err_close;
1491         }
1492
1493         /* If the mode uses primary, then the following is handled by
1494          * bond_change_active_slave().
1495          */
1496         if (!bond_uses_primary(bond)) {
1497                 /* set promiscuity level to new slave */
1498                 if (bond_dev->flags & IFF_PROMISC) {
1499                         res = dev_set_promiscuity(slave_dev, 1);
1500                         if (res)
1501                                 goto err_close;
1502                 }
1503
1504                 /* set allmulti level to new slave */
1505                 if (bond_dev->flags & IFF_ALLMULTI) {
1506                         res = dev_set_allmulti(slave_dev, 1);
1507                         if (res)
1508                                 goto err_close;
1509                 }
1510
1511                 netif_addr_lock_bh(bond_dev);
1512
1513                 dev_mc_sync_multiple(slave_dev, bond_dev);
1514                 dev_uc_sync_multiple(slave_dev, bond_dev);
1515
1516                 netif_addr_unlock_bh(bond_dev);
1517         }
1518
1519         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1520                 /* add lacpdu mc addr to mc list */
1521                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1522
1523                 dev_mc_add(slave_dev, lacpdu_multicast);
1524         }
1525
1526         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1527         if (res) {
1528                 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1529                            slave_dev->name);
1530                 goto err_close;
1531         }
1532
1533         prev_slave = bond_last_slave(bond);
1534
1535         new_slave->delay = 0;
1536         new_slave->link_failure_count = 0;
1537
1538         bond_update_speed_duplex(new_slave);
1539
1540         new_slave->last_rx = jiffies -
1541                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1542         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1543                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1544
1545         if (bond->params.miimon && !bond->params.use_carrier) {
1546                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1547
1548                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1549                         /* miimon is set but a bonded network driver
1550                          * does not support ETHTOOL/MII and
1551                          * arp_interval is not set.  Note: if
1552                          * use_carrier is enabled, we will never go
1553                          * here (because netif_carrier is always
1554                          * supported); thus, we don't need to change
1555                          * the messages for netif_carrier.
1556                          */
1557                         netdev_warn(bond_dev, "MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n",
1558                                     slave_dev->name);
1559                 } else if (link_reporting == -1) {
1560                         /* unable get link status using mii/ethtool */
1561                         netdev_warn(bond_dev, "can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface\n",
1562                                     slave_dev->name);
1563                 }
1564         }
1565
1566         /* check for initial state */
1567         if (bond->params.miimon) {
1568                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1569                         if (bond->params.updelay) {
1570                                 bond_set_slave_link_state(new_slave,
1571                                                           BOND_LINK_BACK);
1572                                 new_slave->delay = bond->params.updelay;
1573                         } else {
1574                                 bond_set_slave_link_state(new_slave,
1575                                                           BOND_LINK_UP);
1576                         }
1577                 } else {
1578                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN);
1579                 }
1580         } else if (bond->params.arp_interval) {
1581                 bond_set_slave_link_state(new_slave,
1582                                           (netif_carrier_ok(slave_dev) ?
1583                                           BOND_LINK_UP : BOND_LINK_DOWN));
1584         } else {
1585                 bond_set_slave_link_state(new_slave, BOND_LINK_UP);
1586         }
1587
1588         if (new_slave->link != BOND_LINK_DOWN)
1589                 new_slave->last_link_up = jiffies;
1590         netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1591                    new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1592                    (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1593
1594         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1595                 /* if there is a primary slave, remember it */
1596                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1597                         rcu_assign_pointer(bond->primary_slave, new_slave);
1598                         bond->force_primary = true;
1599                 }
1600         }
1601
1602         switch (BOND_MODE(bond)) {
1603         case BOND_MODE_ACTIVEBACKUP:
1604                 bond_set_slave_inactive_flags(new_slave,
1605                                               BOND_SLAVE_NOTIFY_NOW);
1606                 break;
1607         case BOND_MODE_8023AD:
1608                 /* in 802.3ad mode, the internal mechanism
1609                  * will activate the slaves in the selected
1610                  * aggregator
1611                  */
1612                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1613                 /* if this is the first slave */
1614                 if (!prev_slave) {
1615                         SLAVE_AD_INFO(new_slave)->id = 1;
1616                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1617                          * can be called only after the mac address of the bond is set
1618                          */
1619                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1620                 } else {
1621                         SLAVE_AD_INFO(new_slave)->id =
1622                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1623                 }
1624
1625                 bond_3ad_bind_slave(new_slave);
1626                 break;
1627         case BOND_MODE_TLB:
1628         case BOND_MODE_ALB:
1629                 bond_set_active_slave(new_slave);
1630                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1631                 break;
1632         default:
1633                 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1634
1635                 /* always active in trunk mode */
1636                 bond_set_active_slave(new_slave);
1637
1638                 /* In trunking mode there is little meaning to curr_active_slave
1639                  * anyway (it holds no special properties of the bond device),
1640                  * so we can change it without calling change_active_interface()
1641                  */
1642                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1643                     new_slave->link == BOND_LINK_UP)
1644                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1645
1646                 break;
1647         } /* switch(bond_mode) */
1648
1649 #ifdef CONFIG_NET_POLL_CONTROLLER
1650         slave_dev->npinfo = bond->dev->npinfo;
1651         if (slave_dev->npinfo) {
1652                 if (slave_enable_netpoll(new_slave)) {
1653                         netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1654                         res = -EBUSY;
1655                         goto err_detach;
1656                 }
1657         }
1658 #endif
1659
1660         if (!(bond_dev->features & NETIF_F_LRO))
1661                 dev_disable_lro(slave_dev);
1662
1663         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1664                                          new_slave);
1665         if (res) {
1666                 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1667                 goto err_detach;
1668         }
1669
1670         res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1671         if (res) {
1672                 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1673                 goto err_unregister;
1674         }
1675
1676         res = bond_sysfs_slave_add(new_slave);
1677         if (res) {
1678                 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1679                 goto err_upper_unlink;
1680         }
1681
1682         bond->slave_cnt++;
1683         bond_compute_features(bond);
1684         bond_set_carrier(bond);
1685
1686         if (bond_uses_primary(bond)) {
1687                 block_netpoll_tx();
1688                 bond_select_active_slave(bond);
1689                 unblock_netpoll_tx();
1690         }
1691
1692         if (bond_mode_uses_xmit_hash(bond))
1693                 bond_update_slave_arr(bond, NULL);
1694
1695         netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1696                     slave_dev->name,
1697                     bond_is_active_slave(new_slave) ? "an active" : "a backup",
1698                     new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1699
1700         /* enslave is successful */
1701         bond_queue_slave_event(new_slave);
1702         return 0;
1703
1704 /* Undo stages on error */
1705 err_upper_unlink:
1706         bond_upper_dev_unlink(bond_dev, slave_dev);
1707
1708 err_unregister:
1709         netdev_rx_handler_unregister(slave_dev);
1710
1711 err_detach:
1712         if (!bond_uses_primary(bond))
1713                 bond_hw_addr_flush(bond_dev, slave_dev);
1714
1715         vlan_vids_del_by_dev(slave_dev, bond_dev);
1716         if (rcu_access_pointer(bond->primary_slave) == new_slave)
1717                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1718         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1719                 block_netpoll_tx();
1720                 bond_change_active_slave(bond, NULL);
1721                 bond_select_active_slave(bond);
1722                 unblock_netpoll_tx();
1723         }
1724         /* either primary_slave or curr_active_slave might've changed */
1725         synchronize_rcu();
1726         slave_disable_netpoll(new_slave);
1727
1728 err_close:
1729         slave_dev->priv_flags &= ~IFF_BONDING;
1730         dev_close(slave_dev);
1731
1732 err_restore_mac:
1733         slave_dev->flags &= ~IFF_SLAVE;
1734         if (!bond->params.fail_over_mac ||
1735             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1736                 /* XXX TODO - fom follow mode needs to change master's
1737                  * MAC if this slave's MAC is in use by the bond, or at
1738                  * least print a warning.
1739                  */
1740                 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1741                 addr.sa_family = slave_dev->type;
1742                 dev_set_mac_address(slave_dev, &addr);
1743         }
1744
1745 err_restore_mtu:
1746         dev_set_mtu(slave_dev, new_slave->original_mtu);
1747
1748 err_free:
1749         bond_free_slave(new_slave);
1750
1751 err_undo_flags:
1752         /* Enslave of first slave has failed and we need to fix master's mac */
1753         if (!bond_has_slaves(bond)) {
1754                 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1755                                             slave_dev->dev_addr))
1756                         eth_hw_addr_random(bond_dev);
1757                 if (bond_dev->type != ARPHRD_ETHER) {
1758                         dev_close(bond_dev);
1759                         ether_setup(bond_dev);
1760                         bond_dev->flags |= IFF_MASTER;
1761                         bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1762                 }
1763         }
1764
1765         return res;
1766 }
1767
1768 /* Try to release the slave device <slave> from the bond device <master>
1769  * It is legal to access curr_active_slave without a lock because all the function
1770  * is RTNL-locked. If "all" is true it means that the function is being called
1771  * while destroying a bond interface and all slaves are being released.
1772  *
1773  * The rules for slave state should be:
1774  *   for Active/Backup:
1775  *     Active stays on all backups go down
1776  *   for Bonded connections:
1777  *     The first up interface should be left on and all others downed.
1778  */
1779 static int __bond_release_one(struct net_device *bond_dev,
1780                               struct net_device *slave_dev,
1781                               bool all)
1782 {
1783         struct bonding *bond = netdev_priv(bond_dev);
1784         struct slave *slave, *oldcurrent;
1785         struct sockaddr addr;
1786         int old_flags = bond_dev->flags;
1787         netdev_features_t old_features = bond_dev->features;
1788
1789         /* slave is not a slave or master is not master of this slave */
1790         if (!(slave_dev->flags & IFF_SLAVE) ||
1791             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1792                 netdev_dbg(bond_dev, "cannot release %s\n",
1793                            slave_dev->name);
1794                 return -EINVAL;
1795         }
1796
1797         block_netpoll_tx();
1798
1799         slave = bond_get_slave_by_dev(bond, slave_dev);
1800         if (!slave) {
1801                 /* not a slave of this bond */
1802                 netdev_info(bond_dev, "%s not enslaved\n",
1803                             slave_dev->name);
1804                 unblock_netpoll_tx();
1805                 return -EINVAL;
1806         }
1807
1808         bond_sysfs_slave_del(slave);
1809
1810         /* recompute stats just before removing the slave */
1811         bond_get_stats(bond->dev, &bond->bond_stats);
1812
1813         bond_upper_dev_unlink(bond_dev, slave_dev);
1814         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1815          * for this slave anymore.
1816          */
1817         netdev_rx_handler_unregister(slave_dev);
1818
1819         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1820                 bond_3ad_unbind_slave(slave);
1821
1822         if (bond_mode_uses_xmit_hash(bond))
1823                 bond_update_slave_arr(bond, slave);
1824
1825         netdev_info(bond_dev, "Releasing %s interface %s\n",
1826                     bond_is_active_slave(slave) ? "active" : "backup",
1827                     slave_dev->name);
1828
1829         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1830
1831         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1832
1833         if (!all && (!bond->params.fail_over_mac ||
1834                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1835                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1836                     bond_has_slaves(bond))
1837                         netdev_warn(bond_dev, "the permanent HWaddr of %s - %pM - is still in use by %s - set the HWaddr of %s to a different address to avoid conflicts\n",
1838                                     slave_dev->name, slave->perm_hwaddr,
1839                                     bond_dev->name, slave_dev->name);
1840         }
1841
1842         if (rtnl_dereference(bond->primary_slave) == slave)
1843                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1844
1845         if (oldcurrent == slave)
1846                 bond_change_active_slave(bond, NULL);
1847
1848         if (bond_is_lb(bond)) {
1849                 /* Must be called only after the slave has been
1850                  * detached from the list and the curr_active_slave
1851                  * has been cleared (if our_slave == old_current),
1852                  * but before a new active slave is selected.
1853                  */
1854                 bond_alb_deinit_slave(bond, slave);
1855         }
1856
1857         if (all) {
1858                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1859         } else if (oldcurrent == slave) {
1860                 /* Note that we hold RTNL over this sequence, so there
1861                  * is no concern that another slave add/remove event
1862                  * will interfere.
1863                  */
1864                 bond_select_active_slave(bond);
1865         }
1866
1867         if (!bond_has_slaves(bond)) {
1868                 bond_set_carrier(bond);
1869                 eth_hw_addr_random(bond_dev);
1870         }
1871
1872         unblock_netpoll_tx();
1873         synchronize_rcu();
1874         bond->slave_cnt--;
1875
1876         if (!bond_has_slaves(bond)) {
1877                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1878                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1879         }
1880
1881         bond_compute_features(bond);
1882         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1883             (old_features & NETIF_F_VLAN_CHALLENGED))
1884                 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1885                             slave_dev->name, bond_dev->name);
1886
1887         vlan_vids_del_by_dev(slave_dev, bond_dev);
1888
1889         /* If the mode uses primary, then this case was handled above by
1890          * bond_change_active_slave(..., NULL)
1891          */
1892         if (!bond_uses_primary(bond)) {
1893                 /* unset promiscuity level from slave
1894                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1895                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1896                  * value of that flag before that change, as that was the value
1897                  * when this slave was attached, so we cache at the start of the
1898                  * function and use it here. Same goes for ALLMULTI below
1899                  */
1900                 if (old_flags & IFF_PROMISC)
1901                         dev_set_promiscuity(slave_dev, -1);
1902
1903                 /* unset allmulti level from slave */
1904                 if (old_flags & IFF_ALLMULTI)
1905                         dev_set_allmulti(slave_dev, -1);
1906
1907                 bond_hw_addr_flush(bond_dev, slave_dev);
1908         }
1909
1910         slave_disable_netpoll(slave);
1911
1912         /* close slave before restoring its mac address */
1913         dev_close(slave_dev);
1914
1915         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1916             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1917                 /* restore original ("permanent") mac address */
1918                 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1919                 addr.sa_family = slave_dev->type;
1920                 dev_set_mac_address(slave_dev, &addr);
1921         }
1922
1923         dev_set_mtu(slave_dev, slave->original_mtu);
1924
1925         slave_dev->priv_flags &= ~IFF_BONDING;
1926
1927         bond_free_slave(slave);
1928
1929         return 0;
1930 }
1931
1932 /* A wrapper used because of ndo_del_link */
1933 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1934 {
1935         return __bond_release_one(bond_dev, slave_dev, false);
1936 }
1937
1938 /* First release a slave and then destroy the bond if no more slaves are left.
1939  * Must be under rtnl_lock when this function is called.
1940  */
1941 static int  bond_release_and_destroy(struct net_device *bond_dev,
1942                                      struct net_device *slave_dev)
1943 {
1944         struct bonding *bond = netdev_priv(bond_dev);
1945         int ret;
1946
1947         ret = bond_release(bond_dev, slave_dev);
1948         if (ret == 0 && !bond_has_slaves(bond)) {
1949                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1950                 netdev_info(bond_dev, "Destroying bond %s\n",
1951                             bond_dev->name);
1952                 bond_remove_proc_entry(bond);
1953                 unregister_netdevice(bond_dev);
1954         }
1955         return ret;
1956 }
1957
1958 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1959 {
1960         struct bonding *bond = netdev_priv(bond_dev);
1961         bond_fill_ifbond(bond, info);
1962         return 0;
1963 }
1964
1965 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1966 {
1967         struct bonding *bond = netdev_priv(bond_dev);
1968         struct list_head *iter;
1969         int i = 0, res = -ENODEV;
1970         struct slave *slave;
1971
1972         bond_for_each_slave(bond, slave, iter) {
1973                 if (i++ == (int)info->slave_id) {
1974                         res = 0;
1975                         bond_fill_ifslave(slave, info);
1976                         break;
1977                 }
1978         }
1979
1980         return res;
1981 }
1982
1983 /*-------------------------------- Monitoring -------------------------------*/
1984
1985 /* called with rcu_read_lock() */
1986 static int bond_miimon_inspect(struct bonding *bond)
1987 {
1988         int link_state, commit = 0;
1989         struct list_head *iter;
1990         struct slave *slave;
1991         bool ignore_updelay;
1992
1993         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
1994
1995         bond_for_each_slave_rcu(bond, slave, iter) {
1996                 slave->new_link = BOND_LINK_NOCHANGE;
1997
1998                 link_state = bond_check_dev_link(bond, slave->dev, 0);
1999
2000                 switch (slave->link) {
2001                 case BOND_LINK_UP:
2002                         if (link_state)
2003                                 continue;
2004
2005                         bond_set_slave_link_state(slave, BOND_LINK_FAIL);
2006                         slave->delay = bond->params.downdelay;
2007                         if (slave->delay) {
2008                                 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2009                                             (BOND_MODE(bond) ==
2010                                              BOND_MODE_ACTIVEBACKUP) ?
2011                                              (bond_is_active_slave(slave) ?
2012                                               "active " : "backup ") : "",
2013                                             slave->dev->name,
2014                                             bond->params.downdelay * bond->params.miimon);
2015                         }
2016                         /*FALLTHRU*/
2017                 case BOND_LINK_FAIL:
2018                         if (link_state) {
2019                                 /* recovered before downdelay expired */
2020                                 bond_set_slave_link_state(slave, BOND_LINK_UP);
2021                                 slave->last_link_up = jiffies;
2022                                 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2023                                             (bond->params.downdelay - slave->delay) *
2024                                             bond->params.miimon,
2025                                             slave->dev->name);
2026                                 continue;
2027                         }
2028
2029                         if (slave->delay <= 0) {
2030                                 slave->new_link = BOND_LINK_DOWN;
2031                                 commit++;
2032                                 continue;
2033                         }
2034
2035                         slave->delay--;
2036                         break;
2037
2038                 case BOND_LINK_DOWN:
2039                         if (!link_state)
2040                                 continue;
2041
2042                         bond_set_slave_link_state(slave, BOND_LINK_BACK);
2043                         slave->delay = bond->params.updelay;
2044
2045                         if (slave->delay) {
2046                                 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2047                                             slave->dev->name,
2048                                             ignore_updelay ? 0 :
2049                                             bond->params.updelay *
2050                                             bond->params.miimon);
2051                         }
2052                         /*FALLTHRU*/
2053                 case BOND_LINK_BACK:
2054                         if (!link_state) {
2055                                 bond_set_slave_link_state(slave,
2056                                                           BOND_LINK_DOWN);
2057                                 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2058                                             (bond->params.updelay - slave->delay) *
2059                                             bond->params.miimon,
2060                                             slave->dev->name);
2061
2062                                 continue;
2063                         }
2064
2065                         if (ignore_updelay)
2066                                 slave->delay = 0;
2067
2068                         if (slave->delay <= 0) {
2069                                 slave->new_link = BOND_LINK_UP;
2070                                 commit++;
2071                                 ignore_updelay = false;
2072                                 continue;
2073                         }
2074
2075                         slave->delay--;
2076                         break;
2077                 }
2078         }
2079
2080         return commit;
2081 }
2082
2083 static void bond_miimon_commit(struct bonding *bond)
2084 {
2085         struct list_head *iter;
2086         struct slave *slave, *primary;
2087
2088         bond_for_each_slave(bond, slave, iter) {
2089                 switch (slave->new_link) {
2090                 case BOND_LINK_NOCHANGE:
2091                         continue;
2092
2093                 case BOND_LINK_UP:
2094                         bond_set_slave_link_state(slave, BOND_LINK_UP);
2095                         slave->last_link_up = jiffies;
2096
2097                         primary = rtnl_dereference(bond->primary_slave);
2098                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2099                                 /* prevent it from being the active one */
2100                                 bond_set_backup_slave(slave);
2101                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2102                                 /* make it immediately active */
2103                                 bond_set_active_slave(slave);
2104                         } else if (slave != primary) {
2105                                 /* prevent it from being the active one */
2106                                 bond_set_backup_slave(slave);
2107                         }
2108
2109                         netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2110                                     slave->dev->name,
2111                                     slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2112                                     slave->duplex ? "full" : "half");
2113
2114                         /* notify ad that the link status has changed */
2115                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2116                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2117
2118                         if (bond_is_lb(bond))
2119                                 bond_alb_handle_link_change(bond, slave,
2120                                                             BOND_LINK_UP);
2121
2122                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2123                                 bond_update_slave_arr(bond, NULL);
2124
2125                         if (!bond->curr_active_slave || slave == primary)
2126                                 goto do_failover;
2127
2128                         continue;
2129
2130                 case BOND_LINK_DOWN:
2131                         if (slave->link_failure_count < UINT_MAX)
2132                                 slave->link_failure_count++;
2133
2134                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2135
2136                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2137                             BOND_MODE(bond) == BOND_MODE_8023AD)
2138                                 bond_set_slave_inactive_flags(slave,
2139                                                               BOND_SLAVE_NOTIFY_NOW);
2140
2141                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2142                                     slave->dev->name);
2143
2144                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2145                                 bond_3ad_handle_link_change(slave,
2146                                                             BOND_LINK_DOWN);
2147
2148                         if (bond_is_lb(bond))
2149                                 bond_alb_handle_link_change(bond, slave,
2150                                                             BOND_LINK_DOWN);
2151
2152                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2153                                 bond_update_slave_arr(bond, NULL);
2154
2155                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2156                                 goto do_failover;
2157
2158                         continue;
2159
2160                 default:
2161                         netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2162                                    slave->new_link, slave->dev->name);
2163                         slave->new_link = BOND_LINK_NOCHANGE;
2164
2165                         continue;
2166                 }
2167
2168 do_failover:
2169                 block_netpoll_tx();
2170                 bond_select_active_slave(bond);
2171                 unblock_netpoll_tx();
2172         }
2173
2174         bond_set_carrier(bond);
2175 }
2176
2177 /* bond_mii_monitor
2178  *
2179  * Really a wrapper that splits the mii monitor into two phases: an
2180  * inspection, then (if inspection indicates something needs to be done)
2181  * an acquisition of appropriate locks followed by a commit phase to
2182  * implement whatever link state changes are indicated.
2183  */
2184 static void bond_mii_monitor(struct work_struct *work)
2185 {
2186         struct bonding *bond = container_of(work, struct bonding,
2187                                             mii_work.work);
2188         bool should_notify_peers = false;
2189         unsigned long delay;
2190
2191         delay = msecs_to_jiffies(bond->params.miimon);
2192
2193         if (!bond_has_slaves(bond))
2194                 goto re_arm;
2195
2196         rcu_read_lock();
2197
2198         should_notify_peers = bond_should_notify_peers(bond);
2199
2200         if (bond_miimon_inspect(bond)) {
2201                 rcu_read_unlock();
2202
2203                 /* Race avoidance with bond_close cancel of workqueue */
2204                 if (!rtnl_trylock()) {
2205                         delay = 1;
2206                         should_notify_peers = false;
2207                         goto re_arm;
2208                 }
2209
2210                 bond_miimon_commit(bond);
2211
2212                 rtnl_unlock();  /* might sleep, hold no other locks */
2213         } else
2214                 rcu_read_unlock();
2215
2216 re_arm:
2217         if (bond->params.miimon)
2218                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2219
2220         if (should_notify_peers) {
2221                 if (!rtnl_trylock())
2222                         return;
2223                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2224                 rtnl_unlock();
2225         }
2226 }
2227
2228 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2229 {
2230         struct net_device *upper;
2231         struct list_head *iter;
2232         bool ret = false;
2233
2234         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2235                 return true;
2236
2237         rcu_read_lock();
2238         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2239                 if (ip == bond_confirm_addr(upper, 0, ip)) {
2240                         ret = true;
2241                         break;
2242                 }
2243         }
2244         rcu_read_unlock();
2245
2246         return ret;
2247 }
2248
2249 /* We go to the (large) trouble of VLAN tagging ARP frames because
2250  * switches in VLAN mode (especially if ports are configured as
2251  * "native" to a VLAN) might not pass non-tagged frames.
2252  */
2253 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2254                           __be32 dest_ip, __be32 src_ip,
2255                           struct bond_vlan_tag *tags)
2256 {
2257         struct sk_buff *skb;
2258         struct bond_vlan_tag *outer_tag = tags;
2259
2260         netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2261                    arp_op, slave_dev->name, &dest_ip, &src_ip);
2262
2263         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2264                          NULL, slave_dev->dev_addr, NULL);
2265
2266         if (!skb) {
2267                 net_err_ratelimited("ARP packet allocation failed\n");
2268                 return;
2269         }
2270
2271         if (!tags || tags->vlan_proto == VLAN_N_VID)
2272                 goto xmit;
2273
2274         tags++;
2275
2276         /* Go through all the tags backwards and add them to the packet */
2277         while (tags->vlan_proto != VLAN_N_VID) {
2278                 if (!tags->vlan_id) {
2279                         tags++;
2280                         continue;
2281                 }
2282
2283                 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2284                            ntohs(outer_tag->vlan_proto), tags->vlan_id);
2285                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2286                                                 tags->vlan_id);
2287                 if (!skb) {
2288                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2289                         return;
2290                 }
2291
2292                 tags++;
2293         }
2294         /* Set the outer tag */
2295         if (outer_tag->vlan_id) {
2296                 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2297                            ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2298                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2299                                        outer_tag->vlan_id);
2300         }
2301
2302 xmit:
2303         arp_xmit(skb);
2304 }
2305
2306 /* Validate the device path between the @start_dev and the @end_dev.
2307  * The path is valid if the @end_dev is reachable through device
2308  * stacking.
2309  * When the path is validated, collect any vlan information in the
2310  * path.
2311  */
2312 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2313                                               struct net_device *end_dev,
2314                                               int level)
2315 {
2316         struct bond_vlan_tag *tags;
2317         struct net_device *upper;
2318         struct list_head  *iter;
2319
2320         if (start_dev == end_dev) {
2321                 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2322                 if (!tags)
2323                         return ERR_PTR(-ENOMEM);
2324                 tags[level].vlan_proto = VLAN_N_VID;
2325                 return tags;
2326         }
2327
2328         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2329                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2330                 if (IS_ERR_OR_NULL(tags)) {
2331                         if (IS_ERR(tags))
2332                                 return tags;
2333                         continue;
2334                 }
2335                 if (is_vlan_dev(upper)) {
2336                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2337                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2338                 }
2339
2340                 return tags;
2341         }
2342
2343         return NULL;
2344 }
2345
2346 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2347 {
2348         struct rtable *rt;
2349         struct bond_vlan_tag *tags;
2350         __be32 *targets = bond->params.arp_targets, addr;
2351         int i;
2352
2353         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2354                 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2355                 tags = NULL;
2356
2357                 /* Find out through which dev should the packet go */
2358                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2359                                      RTO_ONLINK, 0);
2360                 if (IS_ERR(rt)) {
2361                         /* there's no route to target - try to send arp
2362                          * probe to generate any traffic (arp_validate=0)
2363                          */
2364                         if (bond->params.arp_validate)
2365                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2366                                                      bond->dev->name,
2367                                                      &targets[i]);
2368                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2369                                       0, tags);
2370                         continue;
2371                 }
2372
2373                 /* bond device itself */
2374                 if (rt->dst.dev == bond->dev)
2375                         goto found;
2376
2377                 rcu_read_lock();
2378                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2379                 rcu_read_unlock();
2380
2381                 if (!IS_ERR_OR_NULL(tags))
2382                         goto found;
2383
2384                 /* Not our device - skip */
2385                 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2386                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2387
2388                 ip_rt_put(rt);
2389                 continue;
2390
2391 found:
2392                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2393                 ip_rt_put(rt);
2394                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2395                               addr, tags);
2396                 kfree(tags);
2397         }
2398 }
2399
2400 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2401 {
2402         int i;
2403
2404         if (!sip || !bond_has_this_ip(bond, tip)) {
2405                 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2406                            &sip, &tip);
2407                 return;
2408         }
2409
2410         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2411         if (i == -1) {
2412                 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2413                            &sip);
2414                 return;
2415         }
2416         slave->last_rx = jiffies;
2417         slave->target_last_arp_rx[i] = jiffies;
2418 }
2419
2420 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2421                  struct slave *slave)
2422 {
2423         struct arphdr *arp = (struct arphdr *)skb->data;
2424         struct slave *curr_active_slave, *curr_arp_slave;
2425         unsigned char *arp_ptr;
2426         __be32 sip, tip;
2427         int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2428
2429         if (!slave_do_arp_validate(bond, slave)) {
2430                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2431                     !slave_do_arp_validate_only(bond))
2432                         slave->last_rx = jiffies;
2433                 return RX_HANDLER_ANOTHER;
2434         } else if (!is_arp) {
2435                 return RX_HANDLER_ANOTHER;
2436         }
2437
2438         alen = arp_hdr_len(bond->dev);
2439
2440         netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2441                    skb->dev->name);
2442
2443         if (alen > skb_headlen(skb)) {
2444                 arp = kmalloc(alen, GFP_ATOMIC);
2445                 if (!arp)
2446                         goto out_unlock;
2447                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2448                         goto out_unlock;
2449         }
2450
2451         if (arp->ar_hln != bond->dev->addr_len ||
2452             skb->pkt_type == PACKET_OTHERHOST ||
2453             skb->pkt_type == PACKET_LOOPBACK ||
2454             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2455             arp->ar_pro != htons(ETH_P_IP) ||
2456             arp->ar_pln != 4)
2457                 goto out_unlock;
2458
2459         arp_ptr = (unsigned char *)(arp + 1);
2460         arp_ptr += bond->dev->addr_len;
2461         memcpy(&sip, arp_ptr, 4);
2462         arp_ptr += 4 + bond->dev->addr_len;
2463         memcpy(&tip, arp_ptr, 4);
2464
2465         netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2466                    slave->dev->name, bond_slave_state(slave),
2467                      bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2468                      &sip, &tip);
2469
2470         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2471         curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2472
2473         /* We 'trust' the received ARP enough to validate it if:
2474          *
2475          * (a) the slave receiving the ARP is active (which includes the
2476          * current ARP slave, if any), or
2477          *
2478          * (b) the receiving slave isn't active, but there is a currently
2479          * active slave and it received valid arp reply(s) after it became
2480          * the currently active slave, or
2481          *
2482          * (c) there is an ARP slave that sent an ARP during the prior ARP
2483          * interval, and we receive an ARP reply on any slave.  We accept
2484          * these because switch FDB update delays may deliver the ARP
2485          * reply to a slave other than the sender of the ARP request.
2486          *
2487          * Note: for (b), backup slaves are receiving the broadcast ARP
2488          * request, not a reply.  This request passes from the sending
2489          * slave through the L2 switch(es) to the receiving slave.  Since
2490          * this is checking the request, sip/tip are swapped for
2491          * validation.
2492          *
2493          * This is done to avoid endless looping when we can't reach the
2494          * arp_ip_target and fool ourselves with our own arp requests.
2495          */
2496         if (bond_is_active_slave(slave))
2497                 bond_validate_arp(bond, slave, sip, tip);
2498         else if (curr_active_slave &&
2499                  time_after(slave_last_rx(bond, curr_active_slave),
2500                             curr_active_slave->last_link_up))
2501                 bond_validate_arp(bond, slave, tip, sip);
2502         else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2503                  bond_time_in_interval(bond,
2504                                        dev_trans_start(curr_arp_slave->dev), 1))
2505                 bond_validate_arp(bond, slave, sip, tip);
2506
2507 out_unlock:
2508         if (arp != (struct arphdr *)skb->data)
2509                 kfree(arp);
2510         return RX_HANDLER_ANOTHER;
2511 }
2512
2513 /* function to verify if we're in the arp_interval timeslice, returns true if
2514  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2515  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2516  */
2517 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2518                                   int mod)
2519 {
2520         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2521
2522         return time_in_range(jiffies,
2523                              last_act - delta_in_ticks,
2524                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2525 }
2526
2527 /* This function is called regularly to monitor each slave's link
2528  * ensuring that traffic is being sent and received when arp monitoring
2529  * is used in load-balancing mode. if the adapter has been dormant, then an
2530  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2531  * arp monitoring in active backup mode.
2532  */
2533 static void bond_loadbalance_arp_mon(struct work_struct *work)
2534 {
2535         struct bonding *bond = container_of(work, struct bonding,
2536                                             arp_work.work);
2537         struct slave *slave, *oldcurrent;
2538         struct list_head *iter;
2539         int do_failover = 0, slave_state_changed = 0;
2540
2541         if (!bond_has_slaves(bond))
2542                 goto re_arm;
2543
2544         rcu_read_lock();
2545
2546         oldcurrent = rcu_dereference(bond->curr_active_slave);
2547         /* see if any of the previous devices are up now (i.e. they have
2548          * xmt and rcv traffic). the curr_active_slave does not come into
2549          * the picture unless it is null. also, slave->last_link_up is not
2550          * needed here because we send an arp on each slave and give a slave
2551          * as long as it needs to get the tx/rx within the delta.
2552          * TODO: what about up/down delay in arp mode? it wasn't here before
2553          *       so it can wait
2554          */
2555         bond_for_each_slave_rcu(bond, slave, iter) {
2556                 unsigned long trans_start = dev_trans_start(slave->dev);
2557
2558                 if (slave->link != BOND_LINK_UP) {
2559                         if (bond_time_in_interval(bond, trans_start, 1) &&
2560                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2561
2562                                 slave->link  = BOND_LINK_UP;
2563                                 slave_state_changed = 1;
2564
2565                                 /* primary_slave has no meaning in round-robin
2566                                  * mode. the window of a slave being up and
2567                                  * curr_active_slave being null after enslaving
2568                                  * is closed.
2569                                  */
2570                                 if (!oldcurrent) {
2571                                         netdev_info(bond->dev, "link status definitely up for interface %s\n",
2572                                                     slave->dev->name);
2573                                         do_failover = 1;
2574                                 } else {
2575                                         netdev_info(bond->dev, "interface %s is now up\n",
2576                                                     slave->dev->name);
2577                                 }
2578                         }
2579                 } else {
2580                         /* slave->link == BOND_LINK_UP */
2581
2582                         /* not all switches will respond to an arp request
2583                          * when the source ip is 0, so don't take the link down
2584                          * if we don't know our ip yet
2585                          */
2586                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2587                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2588
2589                                 slave->link  = BOND_LINK_DOWN;
2590                                 slave_state_changed = 1;
2591
2592                                 if (slave->link_failure_count < UINT_MAX)
2593                                         slave->link_failure_count++;
2594
2595                                 netdev_info(bond->dev, "interface %s is now down\n",
2596                                             slave->dev->name);
2597
2598                                 if (slave == oldcurrent)
2599                                         do_failover = 1;
2600                         }
2601                 }
2602
2603                 /* note: if switch is in round-robin mode, all links
2604                  * must tx arp to ensure all links rx an arp - otherwise
2605                  * links may oscillate or not come up at all; if switch is
2606                  * in something like xor mode, there is nothing we can
2607                  * do - all replies will be rx'ed on same link causing slaves
2608                  * to be unstable during low/no traffic periods
2609                  */
2610                 if (bond_slave_is_up(slave))
2611                         bond_arp_send_all(bond, slave);
2612         }
2613
2614         rcu_read_unlock();
2615
2616         if (do_failover || slave_state_changed) {
2617                 if (!rtnl_trylock())
2618                         goto re_arm;
2619
2620                 if (slave_state_changed) {
2621                         bond_slave_state_change(bond);
2622                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2623                                 bond_update_slave_arr(bond, NULL);
2624                 }
2625                 if (do_failover) {
2626                         block_netpoll_tx();
2627                         bond_select_active_slave(bond);
2628                         unblock_netpoll_tx();
2629                 }
2630                 rtnl_unlock();
2631         }
2632
2633 re_arm:
2634         if (bond->params.arp_interval)
2635                 queue_delayed_work(bond->wq, &bond->arp_work,
2636                                    msecs_to_jiffies(bond->params.arp_interval));
2637 }
2638
2639 /* Called to inspect slaves for active-backup mode ARP monitor link state
2640  * changes.  Sets new_link in slaves to specify what action should take
2641  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2642  * to link states must be committed.
2643  *
2644  * Called with rcu_read_lock held.
2645  */
2646 static int bond_ab_arp_inspect(struct bonding *bond)
2647 {
2648         unsigned long trans_start, last_rx;
2649         struct list_head *iter;
2650         struct slave *slave;
2651         int commit = 0;
2652
2653         bond_for_each_slave_rcu(bond, slave, iter) {
2654                 slave->new_link = BOND_LINK_NOCHANGE;
2655                 last_rx = slave_last_rx(bond, slave);
2656
2657                 if (slave->link != BOND_LINK_UP) {
2658                         if (bond_time_in_interval(bond, last_rx, 1)) {
2659                                 slave->new_link = BOND_LINK_UP;
2660                                 commit++;
2661                         }
2662                         continue;
2663                 }
2664
2665                 /* Give slaves 2*delta after being enslaved or made
2666                  * active.  This avoids bouncing, as the last receive
2667                  * times need a full ARP monitor cycle to be updated.
2668                  */
2669                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2670                         continue;
2671
2672                 /* Backup slave is down if:
2673                  * - No current_arp_slave AND
2674                  * - more than 3*delta since last receive AND
2675                  * - the bond has an IP address
2676                  *
2677                  * Note: a non-null current_arp_slave indicates
2678                  * the curr_active_slave went down and we are
2679                  * searching for a new one; under this condition
2680                  * we only take the curr_active_slave down - this
2681                  * gives each slave a chance to tx/rx traffic
2682                  * before being taken out
2683                  */
2684                 if (!bond_is_active_slave(slave) &&
2685                     !rcu_access_pointer(bond->current_arp_slave) &&
2686                     !bond_time_in_interval(bond, last_rx, 3)) {
2687                         slave->new_link = BOND_LINK_DOWN;
2688                         commit++;
2689                 }
2690
2691                 /* Active slave is down if:
2692                  * - more than 2*delta since transmitting OR
2693                  * - (more than 2*delta since receive AND
2694                  *    the bond has an IP address)
2695                  */
2696                 trans_start = dev_trans_start(slave->dev);
2697                 if (bond_is_active_slave(slave) &&
2698                     (!bond_time_in_interval(bond, trans_start, 2) ||
2699                      !bond_time_in_interval(bond, last_rx, 2))) {
2700                         slave->new_link = BOND_LINK_DOWN;
2701                         commit++;
2702                 }
2703         }
2704
2705         return commit;
2706 }
2707
2708 /* Called to commit link state changes noted by inspection step of
2709  * active-backup mode ARP monitor.
2710  *
2711  * Called with RTNL hold.
2712  */
2713 static void bond_ab_arp_commit(struct bonding *bond)
2714 {
2715         unsigned long trans_start;
2716         struct list_head *iter;
2717         struct slave *slave;
2718
2719         bond_for_each_slave(bond, slave, iter) {
2720                 switch (slave->new_link) {
2721                 case BOND_LINK_NOCHANGE:
2722                         continue;
2723
2724                 case BOND_LINK_UP:
2725                         trans_start = dev_trans_start(slave->dev);
2726                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
2727                             (!rtnl_dereference(bond->curr_active_slave) &&
2728                              bond_time_in_interval(bond, trans_start, 1))) {
2729                                 struct slave *current_arp_slave;
2730
2731                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2732                                 bond_set_slave_link_state(slave, BOND_LINK_UP);
2733                                 if (current_arp_slave) {
2734                                         bond_set_slave_inactive_flags(
2735                                                 current_arp_slave,
2736                                                 BOND_SLAVE_NOTIFY_NOW);
2737                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2738                                 }
2739
2740                                 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2741                                             slave->dev->name);
2742
2743                                 if (!rtnl_dereference(bond->curr_active_slave) ||
2744                                     slave == rtnl_dereference(bond->primary_slave))
2745                                         goto do_failover;
2746
2747                         }
2748
2749                         continue;
2750
2751                 case BOND_LINK_DOWN:
2752                         if (slave->link_failure_count < UINT_MAX)
2753                                 slave->link_failure_count++;
2754
2755                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2756                         bond_set_slave_inactive_flags(slave,
2757                                                       BOND_SLAVE_NOTIFY_NOW);
2758
2759                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2760                                     slave->dev->name);
2761
2762                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
2763                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2764                                 goto do_failover;
2765                         }
2766
2767                         continue;
2768
2769                 default:
2770                         netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2771                                    slave->new_link, slave->dev->name);
2772                         continue;
2773                 }
2774
2775 do_failover:
2776                 block_netpoll_tx();
2777                 bond_select_active_slave(bond);
2778                 unblock_netpoll_tx();
2779         }
2780
2781         bond_set_carrier(bond);
2782 }
2783
2784 /* Send ARP probes for active-backup mode ARP monitor.
2785  *
2786  * Called with rcu_read_lock held.
2787  */
2788 static bool bond_ab_arp_probe(struct bonding *bond)
2789 {
2790         struct slave *slave, *before = NULL, *new_slave = NULL,
2791                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2792                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2793         struct list_head *iter;
2794         bool found = false;
2795         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2796
2797         if (curr_arp_slave && curr_active_slave)
2798                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2799                             curr_arp_slave->dev->name,
2800                             curr_active_slave->dev->name);
2801
2802         if (curr_active_slave) {
2803                 bond_arp_send_all(bond, curr_active_slave);
2804                 return should_notify_rtnl;
2805         }
2806
2807         /* if we don't have a curr_active_slave, search for the next available
2808          * backup slave from the current_arp_slave and make it the candidate
2809          * for becoming the curr_active_slave
2810          */
2811
2812         if (!curr_arp_slave) {
2813                 curr_arp_slave = bond_first_slave_rcu(bond);
2814                 if (!curr_arp_slave)
2815                         return should_notify_rtnl;
2816         }
2817
2818         bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2819
2820         bond_for_each_slave_rcu(bond, slave, iter) {
2821                 if (!found && !before && bond_slave_is_up(slave))
2822                         before = slave;
2823
2824                 if (found && !new_slave && bond_slave_is_up(slave))
2825                         new_slave = slave;
2826                 /* if the link state is up at this point, we
2827                  * mark it down - this can happen if we have
2828                  * simultaneous link failures and
2829                  * reselect_active_interface doesn't make this
2830                  * one the current slave so it is still marked
2831                  * up when it is actually down
2832                  */
2833                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2834                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2835                         if (slave->link_failure_count < UINT_MAX)
2836                                 slave->link_failure_count++;
2837
2838                         bond_set_slave_inactive_flags(slave,
2839                                                       BOND_SLAVE_NOTIFY_LATER);
2840
2841                         netdev_info(bond->dev, "backup interface %s is now down\n",
2842                                     slave->dev->name);
2843                 }
2844                 if (slave == curr_arp_slave)
2845                         found = true;
2846         }
2847
2848         if (!new_slave && before)
2849                 new_slave = before;
2850
2851         if (!new_slave)
2852                 goto check_state;
2853
2854         bond_set_slave_link_state(new_slave, BOND_LINK_BACK);
2855         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2856         bond_arp_send_all(bond, new_slave);
2857         new_slave->last_link_up = jiffies;
2858         rcu_assign_pointer(bond->current_arp_slave, new_slave);
2859
2860 check_state:
2861         bond_for_each_slave_rcu(bond, slave, iter) {
2862                 if (slave->should_notify) {
2863                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2864                         break;
2865                 }
2866         }
2867         return should_notify_rtnl;
2868 }
2869
2870 static void bond_activebackup_arp_mon(struct work_struct *work)
2871 {
2872         struct bonding *bond = container_of(work, struct bonding,
2873                                             arp_work.work);
2874         bool should_notify_peers = false;
2875         bool should_notify_rtnl = false;
2876         int delta_in_ticks;
2877
2878         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2879
2880         if (!bond_has_slaves(bond))
2881                 goto re_arm;
2882
2883         rcu_read_lock();
2884
2885         should_notify_peers = bond_should_notify_peers(bond);
2886
2887         if (bond_ab_arp_inspect(bond)) {
2888                 rcu_read_unlock();
2889
2890                 /* Race avoidance with bond_close flush of workqueue */
2891                 if (!rtnl_trylock()) {
2892                         delta_in_ticks = 1;
2893                         should_notify_peers = false;
2894                         goto re_arm;
2895                 }
2896
2897                 bond_ab_arp_commit(bond);
2898
2899                 rtnl_unlock();
2900                 rcu_read_lock();
2901         }
2902
2903         should_notify_rtnl = bond_ab_arp_probe(bond);
2904         rcu_read_unlock();
2905
2906 re_arm:
2907         if (bond->params.arp_interval)
2908                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2909
2910         if (should_notify_peers || should_notify_rtnl) {
2911                 if (!rtnl_trylock())
2912                         return;
2913
2914                 if (should_notify_peers)
2915                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2916                                                  bond->dev);
2917                 if (should_notify_rtnl)
2918                         bond_slave_state_notify(bond);
2919
2920                 rtnl_unlock();
2921         }
2922 }
2923
2924 /*-------------------------- netdev event handling --------------------------*/
2925
2926 /* Change device name */
2927 static int bond_event_changename(struct bonding *bond)
2928 {
2929         bond_remove_proc_entry(bond);
2930         bond_create_proc_entry(bond);
2931
2932         bond_debug_reregister(bond);
2933
2934         return NOTIFY_DONE;
2935 }
2936
2937 static int bond_master_netdev_event(unsigned long event,
2938                                     struct net_device *bond_dev)
2939 {
2940         struct bonding *event_bond = netdev_priv(bond_dev);
2941
2942         switch (event) {
2943         case NETDEV_CHANGENAME:
2944                 return bond_event_changename(event_bond);
2945         case NETDEV_UNREGISTER:
2946                 bond_remove_proc_entry(event_bond);
2947                 break;
2948         case NETDEV_REGISTER:
2949                 bond_create_proc_entry(event_bond);
2950                 break;
2951         case NETDEV_NOTIFY_PEERS:
2952                 if (event_bond->send_peer_notif)
2953                         event_bond->send_peer_notif--;
2954                 break;
2955         default:
2956                 break;
2957         }
2958
2959         return NOTIFY_DONE;
2960 }
2961
2962 static int bond_slave_netdev_event(unsigned long event,
2963                                    struct net_device *slave_dev)
2964 {
2965         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
2966         struct bonding *bond;
2967         struct net_device *bond_dev;
2968
2969         /* A netdev event can be generated while enslaving a device
2970          * before netdev_rx_handler_register is called in which case
2971          * slave will be NULL
2972          */
2973         if (!slave)
2974                 return NOTIFY_DONE;
2975         bond_dev = slave->bond->dev;
2976         bond = slave->bond;
2977         primary = rtnl_dereference(bond->primary_slave);
2978
2979         switch (event) {
2980         case NETDEV_UNREGISTER:
2981                 if (bond_dev->type != ARPHRD_ETHER)
2982                         bond_release_and_destroy(bond_dev, slave_dev);
2983                 else
2984                         bond_release(bond_dev, slave_dev);
2985                 break;
2986         case NETDEV_UP:
2987         case NETDEV_CHANGE:
2988                 bond_update_speed_duplex(slave);
2989                 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2990                         bond_3ad_adapter_speed_duplex_changed(slave);
2991                 /* Fallthrough */
2992         case NETDEV_DOWN:
2993                 /* Refresh slave-array if applicable!
2994                  * If the setup does not use miimon or arpmon (mode-specific!),
2995                  * then these events will not cause the slave-array to be
2996                  * refreshed. This will cause xmit to use a slave that is not
2997                  * usable. Avoid such situation by refeshing the array at these
2998                  * events. If these (miimon/arpmon) parameters are configured
2999                  * then array gets refreshed twice and that should be fine!
3000                  */
3001                 if (bond_mode_uses_xmit_hash(bond))
3002                         bond_update_slave_arr(bond, NULL);
3003                 break;
3004         case NETDEV_CHANGEMTU:
3005                 /* TODO: Should slaves be allowed to
3006                  * independently alter their MTU?  For
3007                  * an active-backup bond, slaves need
3008                  * not be the same type of device, so
3009                  * MTUs may vary.  For other modes,
3010                  * slaves arguably should have the
3011                  * same MTUs. To do this, we'd need to
3012                  * take over the slave's change_mtu
3013                  * function for the duration of their
3014                  * servitude.
3015                  */
3016                 break;
3017         case NETDEV_CHANGENAME:
3018                 /* we don't care if we don't have primary set */
3019                 if (!bond_uses_primary(bond) ||
3020                     !bond->params.primary[0])
3021                         break;
3022
3023                 if (slave == primary) {
3024                         /* slave's name changed - he's no longer primary */
3025                         RCU_INIT_POINTER(bond->primary_slave, NULL);
3026                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3027                         /* we have a new primary slave */
3028                         rcu_assign_pointer(bond->primary_slave, slave);
3029                 } else { /* we didn't change primary - exit */
3030                         break;
3031                 }
3032
3033                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3034                             primary ? slave_dev->name : "none");
3035
3036                 block_netpoll_tx();
3037                 bond_select_active_slave(bond);
3038                 unblock_netpoll_tx();
3039                 break;
3040         case NETDEV_FEAT_CHANGE:
3041                 bond_compute_features(bond);
3042                 break;
3043         case NETDEV_RESEND_IGMP:
3044                 /* Propagate to master device */
3045                 call_netdevice_notifiers(event, slave->bond->dev);
3046                 break;
3047         default:
3048                 break;
3049         }
3050
3051         return NOTIFY_DONE;
3052 }
3053
3054 /* bond_netdev_event: handle netdev notifier chain events.
3055  *
3056  * This function receives events for the netdev chain.  The caller (an
3057  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3058  * locks for us to safely manipulate the slave devices (RTNL lock,
3059  * dev_probe_lock).
3060  */
3061 static int bond_netdev_event(struct notifier_block *this,
3062                              unsigned long event, void *ptr)
3063 {
3064         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3065
3066         netdev_dbg(event_dev, "event: %lx\n", event);
3067
3068         if (!(event_dev->priv_flags & IFF_BONDING))
3069                 return NOTIFY_DONE;
3070
3071         if (event_dev->flags & IFF_MASTER) {
3072                 netdev_dbg(event_dev, "IFF_MASTER\n");
3073                 return bond_master_netdev_event(event, event_dev);
3074         }
3075
3076         if (event_dev->flags & IFF_SLAVE) {
3077                 netdev_dbg(event_dev, "IFF_SLAVE\n");
3078                 return bond_slave_netdev_event(event, event_dev);
3079         }
3080
3081         return NOTIFY_DONE;
3082 }
3083
3084 static struct notifier_block bond_netdev_notifier = {
3085         .notifier_call = bond_netdev_event,
3086 };
3087
3088 /*---------------------------- Hashing Policies -----------------------------*/
3089
3090 /* L2 hash helper */
3091 static inline u32 bond_eth_hash(struct sk_buff *skb)
3092 {
3093         struct ethhdr *ep, hdr_tmp;
3094
3095         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3096         if (ep)
3097                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3098         return 0;
3099 }
3100
3101 /* Extract the appropriate headers based on bond's xmit policy */
3102 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3103                               struct flow_keys *fk)
3104 {
3105         const struct ipv6hdr *iph6;
3106         const struct iphdr *iph;
3107         int noff, proto = -1;
3108
3109         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3110                 return skb_flow_dissect_flow_keys(skb, fk, 0);
3111
3112         fk->ports.ports = 0;
3113         noff = skb_network_offset(skb);
3114         if (skb->protocol == htons(ETH_P_IP)) {
3115                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3116                         return false;
3117                 iph = ip_hdr(skb);
3118                 iph_to_flow_copy_v4addrs(fk, iph);
3119                 noff += iph->ihl << 2;
3120                 if (!ip_is_fragment(iph))
3121                         proto = iph->protocol;
3122         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3123                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3124                         return false;
3125                 iph6 = ipv6_hdr(skb);
3126                 iph_to_flow_copy_v6addrs(fk, iph6);
3127                 noff += sizeof(*iph6);
3128                 proto = iph6->nexthdr;
3129         } else {
3130                 return false;
3131         }
3132         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3133                 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3134
3135         return true;
3136 }
3137
3138 /**
3139  * bond_xmit_hash - generate a hash value based on the xmit policy
3140  * @bond: bonding device
3141  * @skb: buffer to use for headers
3142  *
3143  * This function will extract the necessary headers from the skb buffer and use
3144  * them to generate a hash based on the xmit_policy set in the bonding device
3145  */
3146 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3147 {
3148         struct flow_keys flow;
3149         u32 hash;
3150
3151         if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3152             skb->l4_hash)
3153                 return skb->hash;
3154
3155         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3156             !bond_flow_dissect(bond, skb, &flow))
3157                 return bond_eth_hash(skb);
3158
3159         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3160             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3161                 hash = bond_eth_hash(skb);
3162         else
3163                 hash = (__force u32)flow.ports.ports;
3164         hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3165                 (__force u32)flow_get_u32_src(&flow);
3166         hash ^= (hash >> 16);
3167         hash ^= (hash >> 8);
3168
3169         return hash;
3170 }
3171
3172 /*-------------------------- Device entry points ----------------------------*/
3173
3174 static void bond_work_init_all(struct bonding *bond)
3175 {
3176         INIT_DELAYED_WORK(&bond->mcast_work,
3177                           bond_resend_igmp_join_requests_delayed);
3178         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3179         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3180         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3181                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3182         else
3183                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3184         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3185         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3186 }
3187
3188 static void bond_work_cancel_all(struct bonding *bond)
3189 {
3190         cancel_delayed_work_sync(&bond->mii_work);
3191         cancel_delayed_work_sync(&bond->arp_work);
3192         cancel_delayed_work_sync(&bond->alb_work);
3193         cancel_delayed_work_sync(&bond->ad_work);
3194         cancel_delayed_work_sync(&bond->mcast_work);
3195         cancel_delayed_work_sync(&bond->slave_arr_work);
3196 }
3197
3198 static int bond_open(struct net_device *bond_dev)
3199 {
3200         struct bonding *bond = netdev_priv(bond_dev);
3201         struct list_head *iter;
3202         struct slave *slave;
3203
3204         /* reset slave->backup and slave->inactive */
3205         if (bond_has_slaves(bond)) {
3206                 bond_for_each_slave(bond, slave, iter) {
3207                         if (bond_uses_primary(bond) &&
3208                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3209                                 bond_set_slave_inactive_flags(slave,
3210                                                               BOND_SLAVE_NOTIFY_NOW);
3211                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3212                                 bond_set_slave_active_flags(slave,
3213                                                             BOND_SLAVE_NOTIFY_NOW);
3214                         }
3215                 }
3216         }
3217
3218         bond_work_init_all(bond);
3219
3220         if (bond_is_lb(bond)) {
3221                 /* bond_alb_initialize must be called before the timer
3222                  * is started.
3223                  */
3224                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3225                         return -ENOMEM;
3226                 if (bond->params.tlb_dynamic_lb)
3227                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3228         }
3229
3230         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3231                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3232
3233         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3234                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3235                 bond->recv_probe = bond_arp_rcv;
3236         }
3237
3238         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3239                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3240                 /* register to receive LACPDUs */
3241                 bond->recv_probe = bond_3ad_lacpdu_recv;
3242                 bond_3ad_initiate_agg_selection(bond, 1);
3243         }
3244
3245         if (bond_mode_uses_xmit_hash(bond))
3246                 bond_update_slave_arr(bond, NULL);
3247
3248         return 0;
3249 }
3250
3251 static int bond_close(struct net_device *bond_dev)
3252 {
3253         struct bonding *bond = netdev_priv(bond_dev);
3254
3255         bond_work_cancel_all(bond);
3256         bond->send_peer_notif = 0;
3257         if (bond_is_lb(bond))
3258                 bond_alb_deinitialize(bond);
3259         bond->recv_probe = NULL;
3260
3261         return 0;
3262 }
3263
3264 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3265  * that some drivers can provide 32bit values only.
3266  */
3267 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3268                             const struct rtnl_link_stats64 *_new,
3269                             const struct rtnl_link_stats64 *_old)
3270 {
3271         const u64 *new = (const u64 *)_new;
3272         const u64 *old = (const u64 *)_old;
3273         u64 *res = (u64 *)_res;
3274         int i;
3275
3276         for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3277                 u64 nv = new[i];
3278                 u64 ov = old[i];
3279
3280                 /* detects if this particular field is 32bit only */
3281                 if (((nv | ov) >> 32) == 0)
3282                         res[i] += (u32)nv - (u32)ov;
3283                 else
3284                         res[i] += nv - ov;
3285         }
3286 }
3287
3288 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3289                                                 struct rtnl_link_stats64 *stats)
3290 {
3291         struct bonding *bond = netdev_priv(bond_dev);
3292         struct rtnl_link_stats64 temp;
3293         struct list_head *iter;
3294         struct slave *slave;
3295
3296         spin_lock(&bond->stats_lock);
3297         memcpy(stats, &bond->bond_stats, sizeof(*stats));
3298
3299         rcu_read_lock();
3300         bond_for_each_slave_rcu(bond, slave, iter) {
3301                 const struct rtnl_link_stats64 *new =
3302                         dev_get_stats(slave->dev, &temp);
3303
3304                 bond_fold_stats(stats, new, &slave->slave_stats);
3305
3306                 /* save off the slave stats for the next run */
3307                 memcpy(&slave->slave_stats, new, sizeof(*new));
3308         }
3309         rcu_read_unlock();
3310
3311         memcpy(&bond->bond_stats, stats, sizeof(*stats));
3312         spin_unlock(&bond->stats_lock);
3313
3314         return stats;
3315 }
3316
3317 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3318 {
3319         struct bonding *bond = netdev_priv(bond_dev);
3320         struct net_device *slave_dev = NULL;
3321         struct ifbond k_binfo;
3322         struct ifbond __user *u_binfo = NULL;
3323         struct ifslave k_sinfo;
3324         struct ifslave __user *u_sinfo = NULL;
3325         struct mii_ioctl_data *mii = NULL;
3326         struct bond_opt_value newval;
3327         struct net *net;
3328         int res = 0;
3329
3330         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3331
3332         switch (cmd) {
3333         case SIOCGMIIPHY:
3334                 mii = if_mii(ifr);
3335                 if (!mii)
3336                         return -EINVAL;
3337
3338                 mii->phy_id = 0;
3339                 /* Fall Through */
3340         case SIOCGMIIREG:
3341                 /* We do this again just in case we were called by SIOCGMIIREG
3342                  * instead of SIOCGMIIPHY.
3343                  */
3344                 mii = if_mii(ifr);
3345                 if (!mii)
3346                         return -EINVAL;
3347
3348                 if (mii->reg_num == 1) {
3349                         mii->val_out = 0;
3350                         if (netif_carrier_ok(bond->dev))
3351                                 mii->val_out = BMSR_LSTATUS;
3352                 }
3353
3354                 return 0;
3355         case BOND_INFO_QUERY_OLD:
3356         case SIOCBONDINFOQUERY:
3357                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3358
3359                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3360                         return -EFAULT;
3361
3362                 res = bond_info_query(bond_dev, &k_binfo);
3363                 if (res == 0 &&
3364                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3365                         return -EFAULT;
3366
3367                 return res;
3368         case BOND_SLAVE_INFO_QUERY_OLD:
3369         case SIOCBONDSLAVEINFOQUERY:
3370                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3371
3372                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3373                         return -EFAULT;
3374
3375                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3376                 if (res == 0 &&
3377                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3378                         return -EFAULT;
3379
3380                 return res;
3381         default:
3382                 break;
3383         }
3384
3385         net = dev_net(bond_dev);
3386
3387         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3388                 return -EPERM;
3389
3390         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3391
3392         netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3393
3394         if (!slave_dev)
3395                 return -ENODEV;
3396
3397         netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3398         switch (cmd) {
3399         case BOND_ENSLAVE_OLD:
3400         case SIOCBONDENSLAVE:
3401                 res = bond_enslave(bond_dev, slave_dev);
3402                 break;
3403         case BOND_RELEASE_OLD:
3404         case SIOCBONDRELEASE:
3405                 res = bond_release(bond_dev, slave_dev);
3406                 break;
3407         case BOND_SETHWADDR_OLD:
3408         case SIOCBONDSETHWADDR:
3409                 bond_set_dev_addr(bond_dev, slave_dev);
3410                 res = 0;
3411                 break;
3412         case BOND_CHANGE_ACTIVE_OLD:
3413         case SIOCBONDCHANGEACTIVE:
3414                 bond_opt_initstr(&newval, slave_dev->name);
3415                 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3416                 break;
3417         default:
3418                 res = -EOPNOTSUPP;
3419         }
3420
3421         return res;
3422 }
3423
3424 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3425 {
3426         struct bonding *bond = netdev_priv(bond_dev);
3427
3428         if (change & IFF_PROMISC)
3429                 bond_set_promiscuity(bond,
3430                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3431
3432         if (change & IFF_ALLMULTI)
3433                 bond_set_allmulti(bond,
3434                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3435 }
3436
3437 static void bond_set_rx_mode(struct net_device *bond_dev)
3438 {
3439         struct bonding *bond = netdev_priv(bond_dev);
3440         struct list_head *iter;
3441         struct slave *slave;
3442
3443         rcu_read_lock();
3444         if (bond_uses_primary(bond)) {
3445                 slave = rcu_dereference(bond->curr_active_slave);
3446                 if (slave) {
3447                         dev_uc_sync(slave->dev, bond_dev);
3448                         dev_mc_sync(slave->dev, bond_dev);
3449                 }
3450         } else {
3451                 bond_for_each_slave_rcu(bond, slave, iter) {
3452                         dev_uc_sync_multiple(slave->dev, bond_dev);
3453                         dev_mc_sync_multiple(slave->dev, bond_dev);
3454                 }
3455         }
3456         rcu_read_unlock();
3457 }
3458
3459 static int bond_neigh_init(struct neighbour *n)
3460 {
3461         struct bonding *bond = netdev_priv(n->dev);
3462         const struct net_device_ops *slave_ops;
3463         struct neigh_parms parms;
3464         struct slave *slave;
3465         int ret;
3466
3467         slave = bond_first_slave(bond);
3468         if (!slave)
3469                 return 0;
3470         slave_ops = slave->dev->netdev_ops;
3471         if (!slave_ops->ndo_neigh_setup)
3472                 return 0;
3473
3474         parms.neigh_setup = NULL;
3475         parms.neigh_cleanup = NULL;
3476         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3477         if (ret)
3478                 return ret;
3479
3480         /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3481          * after the last slave has been detached.  Assumes that all slaves
3482          * utilize the same neigh_cleanup (true at this writing as only user
3483          * is ipoib).
3484          */
3485         n->parms->neigh_cleanup = parms.neigh_cleanup;
3486
3487         if (!parms.neigh_setup)
3488                 return 0;
3489
3490         return parms.neigh_setup(n);
3491 }
3492
3493 /* The bonding ndo_neigh_setup is called at init time beofre any
3494  * slave exists. So we must declare proxy setup function which will
3495  * be used at run time to resolve the actual slave neigh param setup.
3496  *
3497  * It's also called by master devices (such as vlans) to setup their
3498  * underlying devices. In that case - do nothing, we're already set up from
3499  * our init.
3500  */
3501 static int bond_neigh_setup(struct net_device *dev,
3502                             struct neigh_parms *parms)
3503 {
3504         /* modify only our neigh_parms */
3505         if (parms->dev == dev)
3506                 parms->neigh_setup = bond_neigh_init;
3507
3508         return 0;
3509 }
3510
3511 /* Change the MTU of all of a master's slaves to match the master */
3512 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3513 {
3514         struct bonding *bond = netdev_priv(bond_dev);
3515         struct slave *slave, *rollback_slave;
3516         struct list_head *iter;
3517         int res = 0;
3518
3519         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3520
3521         bond_for_each_slave(bond, slave, iter) {
3522                 netdev_dbg(bond_dev, "s %p c_m %p\n",
3523                            slave, slave->dev->netdev_ops->ndo_change_mtu);
3524
3525                 res = dev_set_mtu(slave->dev, new_mtu);
3526
3527                 if (res) {
3528                         /* If we failed to set the slave's mtu to the new value
3529                          * we must abort the operation even in ACTIVE_BACKUP
3530                          * mode, because if we allow the backup slaves to have
3531                          * different mtu values than the active slave we'll
3532                          * need to change their mtu when doing a failover. That
3533                          * means changing their mtu from timer context, which
3534                          * is probably not a good idea.
3535                          */
3536                         netdev_dbg(bond_dev, "err %d %s\n", res,
3537                                    slave->dev->name);
3538                         goto unwind;
3539                 }
3540         }
3541
3542         bond_dev->mtu = new_mtu;
3543
3544         return 0;
3545
3546 unwind:
3547         /* unwind from head to the slave that failed */
3548         bond_for_each_slave(bond, rollback_slave, iter) {
3549                 int tmp_res;
3550
3551                 if (rollback_slave == slave)
3552                         break;
3553
3554                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3555                 if (tmp_res) {
3556                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3557                                    tmp_res, rollback_slave->dev->name);
3558                 }
3559         }
3560
3561         return res;
3562 }
3563
3564 /* Change HW address
3565  *
3566  * Note that many devices must be down to change the HW address, and
3567  * downing the master releases all slaves.  We can make bonds full of
3568  * bonding devices to test this, however.
3569  */
3570 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3571 {
3572         struct bonding *bond = netdev_priv(bond_dev);
3573         struct slave *slave, *rollback_slave;
3574         struct sockaddr *sa = addr, tmp_sa;
3575         struct list_head *iter;
3576         int res = 0;
3577
3578         if (BOND_MODE(bond) == BOND_MODE_ALB)
3579                 return bond_alb_set_mac_address(bond_dev, addr);
3580
3581
3582         netdev_dbg(bond_dev, "bond=%p\n", bond);
3583
3584         /* If fail_over_mac is enabled, do nothing and return success.
3585          * Returning an error causes ifenslave to fail.
3586          */
3587         if (bond->params.fail_over_mac &&
3588             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3589                 return 0;
3590
3591         if (!is_valid_ether_addr(sa->sa_data))
3592                 return -EADDRNOTAVAIL;
3593
3594         bond_for_each_slave(bond, slave, iter) {
3595                 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3596                 res = dev_set_mac_address(slave->dev, addr);
3597                 if (res) {
3598                         /* TODO: consider downing the slave
3599                          * and retry ?
3600                          * User should expect communications
3601                          * breakage anyway until ARP finish
3602                          * updating, so...
3603                          */
3604                         netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3605                         goto unwind;
3606                 }
3607         }
3608
3609         /* success */
3610         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3611         return 0;
3612
3613 unwind:
3614         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3615         tmp_sa.sa_family = bond_dev->type;
3616
3617         /* unwind from head to the slave that failed */
3618         bond_for_each_slave(bond, rollback_slave, iter) {
3619                 int tmp_res;
3620
3621                 if (rollback_slave == slave)
3622                         break;
3623
3624                 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3625                 if (tmp_res) {
3626                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3627                                    tmp_res, rollback_slave->dev->name);
3628                 }
3629         }
3630
3631         return res;
3632 }
3633
3634 /**
3635  * bond_xmit_slave_id - transmit skb through slave with slave_id
3636  * @bond: bonding device that is transmitting
3637  * @skb: buffer to transmit
3638  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3639  *
3640  * This function tries to transmit through slave with slave_id but in case
3641  * it fails, it tries to find the first available slave for transmission.
3642  * The skb is consumed in all cases, thus the function is void.
3643  */
3644 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3645 {
3646         struct list_head *iter;
3647         struct slave *slave;
3648         int i = slave_id;
3649
3650         /* Here we start from the slave with slave_id */
3651         bond_for_each_slave_rcu(bond, slave, iter) {
3652                 if (--i < 0) {
3653                         if (bond_slave_can_tx(slave)) {
3654                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3655                                 return;
3656                         }
3657                 }
3658         }
3659
3660         /* Here we start from the first slave up to slave_id */
3661         i = slave_id;
3662         bond_for_each_slave_rcu(bond, slave, iter) {
3663                 if (--i < 0)
3664                         break;
3665                 if (bond_slave_can_tx(slave)) {
3666                         bond_dev_queue_xmit(bond, skb, slave->dev);
3667                         return;
3668                 }
3669         }
3670         /* no slave that can tx has been found */
3671         bond_tx_drop(bond->dev, skb);
3672 }
3673
3674 /**
3675  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3676  * @bond: bonding device to use
3677  *
3678  * Based on the value of the bonding device's packets_per_slave parameter
3679  * this function generates a slave id, which is usually used as the next
3680  * slave to transmit through.
3681  */
3682 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3683 {
3684         u32 slave_id;
3685         struct reciprocal_value reciprocal_packets_per_slave;
3686         int packets_per_slave = bond->params.packets_per_slave;
3687
3688         switch (packets_per_slave) {
3689         case 0:
3690                 slave_id = prandom_u32();
3691                 break;
3692         case 1:
3693                 slave_id = bond->rr_tx_counter;
3694                 break;
3695         default:
3696                 reciprocal_packets_per_slave =
3697                         bond->params.reciprocal_packets_per_slave;
3698                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3699                                              reciprocal_packets_per_slave);
3700                 break;
3701         }
3702         bond->rr_tx_counter++;
3703
3704         return slave_id;
3705 }
3706
3707 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3708 {
3709         struct bonding *bond = netdev_priv(bond_dev);
3710         struct iphdr *iph = ip_hdr(skb);
3711         struct slave *slave;
3712         u32 slave_id;
3713
3714         /* Start with the curr_active_slave that joined the bond as the
3715          * default for sending IGMP traffic.  For failover purposes one
3716          * needs to maintain some consistency for the interface that will
3717          * send the join/membership reports.  The curr_active_slave found
3718          * will send all of this type of traffic.
3719          */
3720         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3721                 slave = rcu_dereference(bond->curr_active_slave);
3722                 if (slave)
3723                         bond_dev_queue_xmit(bond, skb, slave->dev);
3724                 else
3725                         bond_xmit_slave_id(bond, skb, 0);
3726         } else {
3727                 int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3728
3729                 if (likely(slave_cnt)) {
3730                         slave_id = bond_rr_gen_slave_id(bond);
3731                         bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3732                 } else {
3733                         bond_tx_drop(bond_dev, skb);
3734                 }
3735         }
3736
3737         return NETDEV_TX_OK;
3738 }
3739
3740 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3741  * the bond has a usable interface.
3742  */
3743 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3744 {
3745         struct bonding *bond = netdev_priv(bond_dev);
3746         struct slave *slave;
3747
3748         slave = rcu_dereference(bond->curr_active_slave);
3749         if (slave)
3750                 bond_dev_queue_xmit(bond, skb, slave->dev);
3751         else
3752                 bond_tx_drop(bond_dev, skb);
3753
3754         return NETDEV_TX_OK;
3755 }
3756
3757 /* Use this to update slave_array when (a) it's not appropriate to update
3758  * slave_array right away (note that update_slave_array() may sleep)
3759  * and / or (b) RTNL is not held.
3760  */
3761 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3762 {
3763         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3764 }
3765
3766 /* Slave array work handler. Holds only RTNL */
3767 static void bond_slave_arr_handler(struct work_struct *work)
3768 {
3769         struct bonding *bond = container_of(work, struct bonding,
3770                                             slave_arr_work.work);
3771         int ret;
3772
3773         if (!rtnl_trylock())
3774                 goto err;
3775
3776         ret = bond_update_slave_arr(bond, NULL);
3777         rtnl_unlock();
3778         if (ret) {
3779                 pr_warn_ratelimited("Failed to update slave array from WT\n");
3780                 goto err;
3781         }
3782         return;
3783
3784 err:
3785         bond_slave_arr_work_rearm(bond, 1);
3786 }
3787
3788 /* Build the usable slaves array in control path for modes that use xmit-hash
3789  * to determine the slave interface -
3790  * (a) BOND_MODE_8023AD
3791  * (b) BOND_MODE_XOR
3792  * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3793  *
3794  * The caller is expected to hold RTNL only and NO other lock!
3795  */
3796 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3797 {
3798         struct slave *slave;
3799         struct list_head *iter;
3800         struct bond_up_slave *new_arr, *old_arr;
3801         int agg_id = 0;
3802         int ret = 0;
3803
3804 #ifdef CONFIG_LOCKDEP
3805         WARN_ON(lockdep_is_held(&bond->mode_lock));
3806 #endif
3807
3808         new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3809                           GFP_KERNEL);
3810         if (!new_arr) {
3811                 ret = -ENOMEM;
3812                 pr_err("Failed to build slave-array.\n");
3813                 goto out;
3814         }
3815         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3816                 struct ad_info ad_info;
3817
3818                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3819                         pr_debug("bond_3ad_get_active_agg_info failed\n");
3820                         kfree_rcu(new_arr, rcu);
3821                         /* No active aggragator means it's not safe to use
3822                          * the previous array.
3823                          */
3824                         old_arr = rtnl_dereference(bond->slave_arr);
3825                         if (old_arr) {
3826                                 RCU_INIT_POINTER(bond->slave_arr, NULL);
3827                                 kfree_rcu(old_arr, rcu);
3828                         }
3829                         goto out;
3830                 }
3831                 agg_id = ad_info.aggregator_id;
3832         }
3833         bond_for_each_slave(bond, slave, iter) {
3834                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3835                         struct aggregator *agg;
3836
3837                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
3838                         if (!agg || agg->aggregator_identifier != agg_id)
3839                                 continue;
3840                 }
3841                 if (!bond_slave_can_tx(slave))
3842                         continue;
3843                 if (skipslave == slave)
3844                         continue;
3845                 new_arr->arr[new_arr->count++] = slave;
3846         }
3847
3848         old_arr = rtnl_dereference(bond->slave_arr);
3849         rcu_assign_pointer(bond->slave_arr, new_arr);
3850         if (old_arr)
3851                 kfree_rcu(old_arr, rcu);
3852 out:
3853         if (ret != 0 && skipslave) {
3854                 int idx;
3855
3856                 /* Rare situation where caller has asked to skip a specific
3857                  * slave but allocation failed (most likely!). BTW this is
3858                  * only possible when the call is initiated from
3859                  * __bond_release_one(). In this situation; overwrite the
3860                  * skipslave entry in the array with the last entry from the
3861                  * array to avoid a situation where the xmit path may choose
3862                  * this to-be-skipped slave to send a packet out.
3863                  */
3864                 old_arr = rtnl_dereference(bond->slave_arr);
3865                 for (idx = 0; idx < old_arr->count; idx++) {
3866                         if (skipslave == old_arr->arr[idx]) {
3867                                 old_arr->arr[idx] =
3868                                     old_arr->arr[old_arr->count-1];
3869                                 old_arr->count--;
3870                                 break;
3871                         }
3872                 }
3873         }
3874         return ret;
3875 }
3876
3877 /* Use this Xmit function for 3AD as well as XOR modes. The current
3878  * usable slave array is formed in the control path. The xmit function
3879  * just calculates hash and sends the packet out.
3880  */
3881 static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3882 {
3883         struct bonding *bond = netdev_priv(dev);
3884         struct slave *slave;
3885         struct bond_up_slave *slaves;
3886         unsigned int count;
3887
3888         slaves = rcu_dereference(bond->slave_arr);
3889         count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3890         if (likely(count)) {
3891                 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3892                 bond_dev_queue_xmit(bond, skb, slave->dev);
3893         } else {
3894                 bond_tx_drop(dev, skb);
3895         }
3896
3897         return NETDEV_TX_OK;
3898 }
3899
3900 /* in broadcast mode, we send everything to all usable interfaces. */
3901 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3902 {
3903         struct bonding *bond = netdev_priv(bond_dev);
3904         struct slave *slave = NULL;
3905         struct list_head *iter;
3906
3907         bond_for_each_slave_rcu(bond, slave, iter) {
3908                 if (bond_is_last_slave(bond, slave))
3909                         break;
3910                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3911                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3912
3913                         if (!skb2) {
3914                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3915                                                     bond_dev->name, __func__);
3916                                 continue;
3917                         }
3918                         bond_dev_queue_xmit(bond, skb2, slave->dev);
3919                 }
3920         }
3921         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3922                 bond_dev_queue_xmit(bond, skb, slave->dev);
3923         else
3924                 bond_tx_drop(bond_dev, skb);
3925
3926         return NETDEV_TX_OK;
3927 }
3928
3929 /*------------------------- Device initialization ---------------------------*/
3930
3931 /* Lookup the slave that corresponds to a qid */
3932 static inline int bond_slave_override(struct bonding *bond,
3933                                       struct sk_buff *skb)
3934 {
3935         struct slave *slave = NULL;
3936         struct list_head *iter;
3937
3938         if (!skb->queue_mapping)
3939                 return 1;
3940
3941         /* Find out if any slaves have the same mapping as this skb. */
3942         bond_for_each_slave_rcu(bond, slave, iter) {
3943                 if (slave->queue_id == skb->queue_mapping) {
3944                         if (bond_slave_is_up(slave) &&
3945                             slave->link == BOND_LINK_UP) {
3946                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3947                                 return 0;
3948                         }
3949                         /* If the slave isn't UP, use default transmit policy. */
3950                         break;
3951                 }
3952         }
3953
3954         return 1;
3955 }
3956
3957
3958 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3959                              void *accel_priv, select_queue_fallback_t fallback)
3960 {
3961         /* This helper function exists to help dev_pick_tx get the correct
3962          * destination queue.  Using a helper function skips a call to
3963          * skb_tx_hash and will put the skbs in the queue we expect on their
3964          * way down to the bonding driver.
3965          */
3966         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3967
3968         /* Save the original txq to restore before passing to the driver */
3969         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3970
3971         if (unlikely(txq >= dev->real_num_tx_queues)) {
3972                 do {
3973                         txq -= dev->real_num_tx_queues;
3974                 } while (txq >= dev->real_num_tx_queues);
3975         }
3976         return txq;
3977 }
3978
3979 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3980 {
3981         struct bonding *bond = netdev_priv(dev);
3982
3983         if (bond_should_override_tx_queue(bond) &&
3984             !bond_slave_override(bond, skb))
3985                 return NETDEV_TX_OK;
3986
3987         switch (BOND_MODE(bond)) {
3988         case BOND_MODE_ROUNDROBIN:
3989                 return bond_xmit_roundrobin(skb, dev);
3990         case BOND_MODE_ACTIVEBACKUP:
3991                 return bond_xmit_activebackup(skb, dev);
3992         case BOND_MODE_8023AD:
3993         case BOND_MODE_XOR:
3994                 return bond_3ad_xor_xmit(skb, dev);
3995         case BOND_MODE_BROADCAST:
3996                 return bond_xmit_broadcast(skb, dev);
3997         case BOND_MODE_ALB:
3998                 return bond_alb_xmit(skb, dev);
3999         case BOND_MODE_TLB:
4000                 return bond_tlb_xmit(skb, dev);
4001         default:
4002                 /* Should never happen, mode already checked */
4003                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4004                 WARN_ON_ONCE(1);
4005                 bond_tx_drop(dev, skb);
4006                 return NETDEV_TX_OK;
4007         }
4008 }
4009
4010 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4011 {
4012         struct bonding *bond = netdev_priv(dev);
4013         netdev_tx_t ret = NETDEV_TX_OK;
4014
4015         /* If we risk deadlock from transmitting this in the
4016          * netpoll path, tell netpoll to queue the frame for later tx
4017          */
4018         if (unlikely(is_netpoll_tx_blocked(dev)))
4019                 return NETDEV_TX_BUSY;
4020
4021         rcu_read_lock();
4022         if (bond_has_slaves(bond))
4023                 ret = __bond_start_xmit(skb, dev);
4024         else
4025                 bond_tx_drop(dev, skb);
4026         rcu_read_unlock();
4027
4028         return ret;
4029 }
4030
4031 static int bond_ethtool_get_settings(struct net_device *bond_dev,
4032                                      struct ethtool_cmd *ecmd)
4033 {
4034         struct bonding *bond = netdev_priv(bond_dev);
4035         unsigned long speed = 0;
4036         struct list_head *iter;
4037         struct slave *slave;
4038
4039         ecmd->duplex = DUPLEX_UNKNOWN;
4040         ecmd->port = PORT_OTHER;
4041
4042         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4043          * do not need to check mode.  Though link speed might not represent
4044          * the true receive or transmit bandwidth (not all modes are symmetric)
4045          * this is an accurate maximum.
4046          */
4047         bond_for_each_slave(bond, slave, iter) {
4048                 if (bond_slave_can_tx(slave)) {
4049                         if (slave->speed != SPEED_UNKNOWN)
4050                                 speed += slave->speed;
4051                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
4052                             slave->duplex != DUPLEX_UNKNOWN)
4053                                 ecmd->duplex = slave->duplex;
4054                 }
4055         }
4056         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
4057
4058         return 0;
4059 }
4060
4061 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4062                                      struct ethtool_drvinfo *drvinfo)
4063 {
4064         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4065         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4066         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4067                  BOND_ABI_VERSION);
4068 }
4069
4070 static const struct ethtool_ops bond_ethtool_ops = {
4071         .get_drvinfo            = bond_ethtool_get_drvinfo,
4072         .get_settings           = bond_ethtool_get_settings,
4073         .get_link               = ethtool_op_get_link,
4074 };
4075
4076 static const struct net_device_ops bond_netdev_ops = {
4077         .ndo_init               = bond_init,
4078         .ndo_uninit             = bond_uninit,
4079         .ndo_open               = bond_open,
4080         .ndo_stop               = bond_close,
4081         .ndo_start_xmit         = bond_start_xmit,
4082         .ndo_select_queue       = bond_select_queue,
4083         .ndo_get_stats64        = bond_get_stats,
4084         .ndo_do_ioctl           = bond_do_ioctl,
4085         .ndo_change_rx_flags    = bond_change_rx_flags,
4086         .ndo_set_rx_mode        = bond_set_rx_mode,
4087         .ndo_change_mtu         = bond_change_mtu,
4088         .ndo_set_mac_address    = bond_set_mac_address,
4089         .ndo_neigh_setup        = bond_neigh_setup,
4090         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4091         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4092 #ifdef CONFIG_NET_POLL_CONTROLLER
4093         .ndo_netpoll_setup      = bond_netpoll_setup,
4094         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4095         .ndo_poll_controller    = bond_poll_controller,
4096 #endif
4097         .ndo_add_slave          = bond_enslave,
4098         .ndo_del_slave          = bond_release,
4099         .ndo_fix_features       = bond_fix_features,
4100         .ndo_bridge_setlink     = switchdev_port_bridge_setlink,
4101         .ndo_bridge_getlink     = switchdev_port_bridge_getlink,
4102         .ndo_bridge_dellink     = switchdev_port_bridge_dellink,
4103         .ndo_fdb_add            = switchdev_port_fdb_add,
4104         .ndo_fdb_del            = switchdev_port_fdb_del,
4105         .ndo_fdb_dump           = switchdev_port_fdb_dump,
4106         .ndo_features_check     = passthru_features_check,
4107 };
4108
4109 static const struct device_type bond_type = {
4110         .name = "bond",
4111 };
4112
4113 static void bond_destructor(struct net_device *bond_dev)
4114 {
4115         struct bonding *bond = netdev_priv(bond_dev);
4116         if (bond->wq)
4117                 destroy_workqueue(bond->wq);
4118         free_netdev(bond_dev);
4119 }
4120
4121 void bond_setup(struct net_device *bond_dev)
4122 {
4123         struct bonding *bond = netdev_priv(bond_dev);
4124
4125         spin_lock_init(&bond->mode_lock);
4126         spin_lock_init(&bond->stats_lock);
4127         bond->params = bonding_defaults;
4128
4129         /* Initialize pointers */
4130         bond->dev = bond_dev;
4131
4132         /* Initialize the device entry points */
4133         ether_setup(bond_dev);
4134         bond_dev->netdev_ops = &bond_netdev_ops;
4135         bond_dev->ethtool_ops = &bond_ethtool_ops;
4136
4137         bond_dev->destructor = bond_destructor;
4138
4139         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4140
4141         /* Initialize the device options */
4142         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4143         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4144         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4145
4146         /* don't acquire bond device's netif_tx_lock when transmitting */
4147         bond_dev->features |= NETIF_F_LLTX;
4148
4149         /* By default, we declare the bond to be fully
4150          * VLAN hardware accelerated capable. Special
4151          * care is taken in the various xmit functions
4152          * when there are slaves that are not hw accel
4153          * capable
4154          */
4155
4156         /* Don't allow bond devices to change network namespaces. */
4157         bond_dev->features |= NETIF_F_NETNS_LOCAL;
4158
4159         bond_dev->hw_features = BOND_VLAN_FEATURES |
4160                                 NETIF_F_HW_VLAN_CTAG_TX |
4161                                 NETIF_F_HW_VLAN_CTAG_RX |
4162                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4163
4164         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4165         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4166         bond_dev->features |= bond_dev->hw_features;
4167 }
4168
4169 /* Destroy a bonding device.
4170  * Must be under rtnl_lock when this function is called.
4171  */
4172 static void bond_uninit(struct net_device *bond_dev)
4173 {
4174         struct bonding *bond = netdev_priv(bond_dev);
4175         struct list_head *iter;
4176         struct slave *slave;
4177         struct bond_up_slave *arr;
4178
4179         bond_netpoll_cleanup(bond_dev);
4180
4181         /* Release the bonded slaves */
4182         bond_for_each_slave(bond, slave, iter)
4183                 __bond_release_one(bond_dev, slave->dev, true);
4184         netdev_info(bond_dev, "Released all slaves\n");
4185
4186         arr = rtnl_dereference(bond->slave_arr);
4187         if (arr) {
4188                 RCU_INIT_POINTER(bond->slave_arr, NULL);
4189                 kfree_rcu(arr, rcu);
4190         }
4191
4192         list_del(&bond->bond_list);
4193
4194         bond_debug_unregister(bond);
4195 }
4196
4197 /*------------------------- Module initialization ---------------------------*/
4198
4199 static int bond_check_params(struct bond_params *params)
4200 {
4201         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4202         struct bond_opt_value newval;
4203         const struct bond_opt_value *valptr;
4204         int arp_all_targets_value;
4205         u16 ad_actor_sys_prio = 0;
4206         u16 ad_user_port_key = 0;
4207
4208         /* Convert string parameters. */
4209         if (mode) {
4210                 bond_opt_initstr(&newval, mode);
4211                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4212                 if (!valptr) {
4213                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4214                         return -EINVAL;
4215                 }
4216                 bond_mode = valptr->value;
4217         }
4218
4219         if (xmit_hash_policy) {
4220                 if ((bond_mode != BOND_MODE_XOR) &&
4221                     (bond_mode != BOND_MODE_8023AD) &&
4222                     (bond_mode != BOND_MODE_TLB)) {
4223                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4224                                 bond_mode_name(bond_mode));
4225                 } else {
4226                         bond_opt_initstr(&newval, xmit_hash_policy);
4227                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4228                                                 &newval);
4229                         if (!valptr) {
4230                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4231                                        xmit_hash_policy);
4232                                 return -EINVAL;
4233                         }
4234                         xmit_hashtype = valptr->value;
4235                 }
4236         }
4237
4238         if (lacp_rate) {
4239                 if (bond_mode != BOND_MODE_8023AD) {
4240                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4241                                 bond_mode_name(bond_mode));
4242                 } else {
4243                         bond_opt_initstr(&newval, lacp_rate);
4244                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4245                                                 &newval);
4246                         if (!valptr) {
4247                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4248                                        lacp_rate);
4249                                 return -EINVAL;
4250                         }
4251                         lacp_fast = valptr->value;
4252                 }
4253         }
4254
4255         if (ad_select) {
4256                 bond_opt_initstr(&newval, ad_select);
4257                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4258                                         &newval);
4259                 if (!valptr) {
4260                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4261                         return -EINVAL;
4262                 }
4263                 params->ad_select = valptr->value;
4264                 if (bond_mode != BOND_MODE_8023AD)
4265                         pr_warn("ad_select param only affects 802.3ad mode\n");
4266         } else {
4267                 params->ad_select = BOND_AD_STABLE;
4268         }
4269
4270         if (max_bonds < 0) {
4271                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4272                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4273                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4274         }
4275
4276         if (miimon < 0) {
4277                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4278                         miimon, INT_MAX);
4279                 miimon = 0;
4280         }
4281
4282         if (updelay < 0) {
4283                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4284                         updelay, INT_MAX);
4285                 updelay = 0;
4286         }
4287
4288         if (downdelay < 0) {
4289                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4290                         downdelay, INT_MAX);
4291                 downdelay = 0;
4292         }
4293
4294         if ((use_carrier != 0) && (use_carrier != 1)) {
4295                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4296                         use_carrier);
4297                 use_carrier = 1;
4298         }
4299
4300         if (num_peer_notif < 0 || num_peer_notif > 255) {
4301                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4302                         num_peer_notif);
4303                 num_peer_notif = 1;
4304         }
4305
4306         /* reset values for 802.3ad/TLB/ALB */
4307         if (!bond_mode_uses_arp(bond_mode)) {
4308                 if (!miimon) {
4309                         pr_warn("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4310                         pr_warn("Forcing miimon to 100msec\n");
4311                         miimon = BOND_DEFAULT_MIIMON;
4312                 }
4313         }
4314
4315         if (tx_queues < 1 || tx_queues > 255) {
4316                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4317                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4318                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4319         }
4320
4321         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4322                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4323                         all_slaves_active);
4324                 all_slaves_active = 0;
4325         }
4326
4327         if (resend_igmp < 0 || resend_igmp > 255) {
4328                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4329                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4330                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4331         }
4332
4333         bond_opt_initval(&newval, packets_per_slave);
4334         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4335                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4336                         packets_per_slave, USHRT_MAX);
4337                 packets_per_slave = 1;
4338         }
4339
4340         if (bond_mode == BOND_MODE_ALB) {
4341                 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",
4342                           updelay);
4343         }
4344
4345         if (!miimon) {
4346                 if (updelay || downdelay) {
4347                         /* just warn the user the up/down delay will have
4348                          * no effect since miimon is zero...
4349                          */
4350                         pr_warn("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4351                                 updelay, downdelay);
4352                 }
4353         } else {
4354                 /* don't allow arp monitoring */
4355                 if (arp_interval) {
4356                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4357                                 miimon, arp_interval);
4358                         arp_interval = 0;
4359                 }
4360
4361                 if ((updelay % miimon) != 0) {
4362                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4363                                 updelay, miimon, (updelay / miimon) * miimon);
4364                 }
4365
4366                 updelay /= miimon;
4367
4368                 if ((downdelay % miimon) != 0) {
4369                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4370                                 downdelay, miimon,
4371                                 (downdelay / miimon) * miimon);
4372                 }
4373
4374                 downdelay /= miimon;
4375         }
4376
4377         if (arp_interval < 0) {
4378                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4379                         arp_interval, INT_MAX);
4380                 arp_interval = 0;
4381         }
4382
4383         for (arp_ip_count = 0, i = 0;
4384              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4385                 __be32 ip;
4386
4387                 /* not a complete check, but good enough to catch mistakes */
4388                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4389                     !bond_is_ip_target_ok(ip)) {
4390                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4391                                 arp_ip_target[i]);
4392                         arp_interval = 0;
4393                 } else {
4394                         if (bond_get_targets_ip(arp_target, ip) == -1)
4395                                 arp_target[arp_ip_count++] = ip;
4396                         else
4397                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4398                                         &ip);
4399                 }
4400         }
4401
4402         if (arp_interval && !arp_ip_count) {
4403                 /* don't allow arping if no arp_ip_target given... */
4404                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4405                         arp_interval);
4406                 arp_interval = 0;
4407         }
4408
4409         if (arp_validate) {
4410                 if (!arp_interval) {
4411                         pr_err("arp_validate requires arp_interval\n");
4412                         return -EINVAL;
4413                 }
4414
4415                 bond_opt_initstr(&newval, arp_validate);
4416                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4417                                         &newval);
4418                 if (!valptr) {
4419                         pr_err("Error: invalid arp_validate \"%s\"\n",
4420                                arp_validate);
4421                         return -EINVAL;
4422                 }
4423                 arp_validate_value = valptr->value;
4424         } else {
4425                 arp_validate_value = 0;
4426         }
4427
4428         arp_all_targets_value = 0;
4429         if (arp_all_targets) {
4430                 bond_opt_initstr(&newval, arp_all_targets);
4431                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4432                                         &newval);
4433                 if (!valptr) {
4434                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4435                                arp_all_targets);
4436                         arp_all_targets_value = 0;
4437                 } else {
4438                         arp_all_targets_value = valptr->value;
4439                 }
4440         }
4441
4442         if (miimon) {
4443                 pr_info("MII link monitoring set to %d ms\n", miimon);
4444         } else if (arp_interval) {
4445                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4446                                           arp_validate_value);
4447                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4448                         arp_interval, valptr->string, arp_ip_count);
4449
4450                 for (i = 0; i < arp_ip_count; i++)
4451                         pr_cont(" %s", arp_ip_target[i]);
4452
4453                 pr_cont("\n");
4454
4455         } else if (max_bonds) {
4456                 /* miimon and arp_interval not set, we need one so things
4457                  * work as expected, see bonding.txt for details
4458                  */
4459                 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");
4460         }
4461
4462         if (primary && !bond_mode_uses_primary(bond_mode)) {
4463                 /* currently, using a primary only makes sense
4464                  * in active backup, TLB or ALB modes
4465                  */
4466                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4467                         primary, bond_mode_name(bond_mode));
4468                 primary = NULL;
4469         }
4470
4471         if (primary && primary_reselect) {
4472                 bond_opt_initstr(&newval, primary_reselect);
4473                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4474                                         &newval);
4475                 if (!valptr) {
4476                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4477                                primary_reselect);
4478                         return -EINVAL;
4479                 }
4480                 primary_reselect_value = valptr->value;
4481         } else {
4482                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4483         }
4484
4485         if (fail_over_mac) {
4486                 bond_opt_initstr(&newval, fail_over_mac);
4487                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4488                                         &newval);
4489                 if (!valptr) {
4490                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4491                                fail_over_mac);
4492                         return -EINVAL;
4493                 }
4494                 fail_over_mac_value = valptr->value;
4495                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4496                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4497         } else {
4498                 fail_over_mac_value = BOND_FOM_NONE;
4499         }
4500
4501         bond_opt_initstr(&newval, "default");
4502         valptr = bond_opt_parse(
4503                         bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4504                                      &newval);
4505         if (!valptr) {
4506                 pr_err("Error: No ad_actor_sys_prio default value");
4507                 return -EINVAL;
4508         }
4509         ad_actor_sys_prio = valptr->value;
4510
4511         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4512                                 &newval);
4513         if (!valptr) {
4514                 pr_err("Error: No ad_user_port_key default value");
4515                 return -EINVAL;
4516         }
4517         ad_user_port_key = valptr->value;
4518
4519         if (lp_interval == 0) {
4520                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4521                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4522                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4523         }
4524
4525         /* fill params struct with the proper values */
4526         params->mode = bond_mode;
4527         params->xmit_policy = xmit_hashtype;
4528         params->miimon = miimon;
4529         params->num_peer_notif = num_peer_notif;
4530         params->arp_interval = arp_interval;
4531         params->arp_validate = arp_validate_value;
4532         params->arp_all_targets = arp_all_targets_value;
4533         params->updelay = updelay;
4534         params->downdelay = downdelay;
4535         params->use_carrier = use_carrier;
4536         params->lacp_fast = lacp_fast;
4537         params->primary[0] = 0;
4538         params->primary_reselect = primary_reselect_value;
4539         params->fail_over_mac = fail_over_mac_value;
4540         params->tx_queues = tx_queues;
4541         params->all_slaves_active = all_slaves_active;
4542         params->resend_igmp = resend_igmp;
4543         params->min_links = min_links;
4544         params->lp_interval = lp_interval;
4545         params->packets_per_slave = packets_per_slave;
4546         params->tlb_dynamic_lb = 1; /* Default value */
4547         params->ad_actor_sys_prio = ad_actor_sys_prio;
4548         eth_zero_addr(params->ad_actor_system);
4549         params->ad_user_port_key = ad_user_port_key;
4550         if (packets_per_slave > 0) {
4551                 params->reciprocal_packets_per_slave =
4552                         reciprocal_value(packets_per_slave);
4553         } else {
4554                 /* reciprocal_packets_per_slave is unused if
4555                  * packets_per_slave is 0 or 1, just initialize it
4556                  */
4557                 params->reciprocal_packets_per_slave =
4558                         (struct reciprocal_value) { 0 };
4559         }
4560
4561         if (primary) {
4562                 strncpy(params->primary, primary, IFNAMSIZ);
4563                 params->primary[IFNAMSIZ - 1] = 0;
4564         }
4565
4566         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4567
4568         return 0;
4569 }
4570
4571 static struct lock_class_key bonding_netdev_xmit_lock_key;
4572 static struct lock_class_key bonding_netdev_addr_lock_key;
4573 static struct lock_class_key bonding_tx_busylock_key;
4574
4575 static void bond_set_lockdep_class_one(struct net_device *dev,
4576                                        struct netdev_queue *txq,
4577                                        void *_unused)
4578 {
4579         lockdep_set_class(&txq->_xmit_lock,
4580                           &bonding_netdev_xmit_lock_key);
4581 }
4582
4583 static void bond_set_lockdep_class(struct net_device *dev)
4584 {
4585         lockdep_set_class(&dev->addr_list_lock,
4586                           &bonding_netdev_addr_lock_key);
4587         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4588         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4589 }
4590
4591 /* Called from registration process */
4592 static int bond_init(struct net_device *bond_dev)
4593 {
4594         struct bonding *bond = netdev_priv(bond_dev);
4595         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4596
4597         netdev_dbg(bond_dev, "Begin bond_init\n");
4598
4599         bond->wq = create_singlethread_workqueue(bond_dev->name);
4600         if (!bond->wq)
4601                 return -ENOMEM;
4602
4603         bond_set_lockdep_class(bond_dev);
4604
4605         list_add_tail(&bond->bond_list, &bn->dev_list);
4606
4607         bond_prepare_sysfs_group(bond);
4608
4609         bond_debug_register(bond);
4610
4611         /* Ensure valid dev_addr */
4612         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4613             bond_dev->addr_assign_type == NET_ADDR_PERM)
4614                 eth_hw_addr_random(bond_dev);
4615
4616         return 0;
4617 }
4618
4619 unsigned int bond_get_num_tx_queues(void)
4620 {
4621         return tx_queues;
4622 }
4623
4624 /* Create a new bond based on the specified name and bonding parameters.
4625  * If name is NULL, obtain a suitable "bond%d" name for us.
4626  * Caller must NOT hold rtnl_lock; we need to release it here before we
4627  * set up our sysfs entries.
4628  */
4629 int bond_create(struct net *net, const char *name)
4630 {
4631         struct net_device *bond_dev;
4632         struct bonding *bond;
4633         struct alb_bond_info *bond_info;
4634         int res;
4635
4636         rtnl_lock();
4637
4638         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4639                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
4640                                    bond_setup, tx_queues);
4641         if (!bond_dev) {
4642                 pr_err("%s: eek! can't alloc netdev!\n", name);
4643                 rtnl_unlock();
4644                 return -ENOMEM;
4645         }
4646
4647         /*
4648          * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4649          * It is set to 0 by default which is wrong.
4650          */
4651         bond = netdev_priv(bond_dev);
4652         bond_info = &(BOND_ALB_INFO(bond));
4653         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4654
4655         dev_net_set(bond_dev, net);
4656         bond_dev->rtnl_link_ops = &bond_link_ops;
4657
4658         res = register_netdevice(bond_dev);
4659
4660         netif_carrier_off(bond_dev);
4661
4662         rtnl_unlock();
4663         if (res < 0)
4664                 bond_destructor(bond_dev);
4665         return res;
4666 }
4667
4668 static int __net_init bond_net_init(struct net *net)
4669 {
4670         struct bond_net *bn = net_generic(net, bond_net_id);
4671
4672         bn->net = net;
4673         INIT_LIST_HEAD(&bn->dev_list);
4674
4675         bond_create_proc_dir(bn);
4676         bond_create_sysfs(bn);
4677
4678         return 0;
4679 }
4680
4681 static void __net_exit bond_net_exit(struct net *net)
4682 {
4683         struct bond_net *bn = net_generic(net, bond_net_id);
4684         struct bonding *bond, *tmp_bond;
4685         LIST_HEAD(list);
4686
4687         bond_destroy_sysfs(bn);
4688
4689         /* Kill off any bonds created after unregistering bond rtnl ops */
4690         rtnl_lock();
4691         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4692                 unregister_netdevice_queue(bond->dev, &list);
4693         unregister_netdevice_many(&list);
4694         rtnl_unlock();
4695
4696         bond_destroy_proc_dir(bn);
4697 }
4698
4699 static struct pernet_operations bond_net_ops = {
4700         .init = bond_net_init,
4701         .exit = bond_net_exit,
4702         .id   = &bond_net_id,
4703         .size = sizeof(struct bond_net),
4704 };
4705
4706 static int __init bonding_init(void)
4707 {
4708         int i;
4709         int res;
4710
4711         pr_info("%s", bond_version);
4712
4713         res = bond_check_params(&bonding_defaults);
4714         if (res)
4715                 goto out;
4716
4717         res = register_pernet_subsys(&bond_net_ops);
4718         if (res)
4719                 goto out;
4720
4721         res = bond_netlink_init();
4722         if (res)
4723                 goto err_link;
4724
4725         bond_create_debugfs();
4726
4727         for (i = 0; i < max_bonds; i++) {
4728                 res = bond_create(&init_net, NULL);
4729                 if (res)
4730                         goto err;
4731         }
4732
4733         register_netdevice_notifier(&bond_netdev_notifier);
4734 out:
4735         return res;
4736 err:
4737         bond_destroy_debugfs();
4738         bond_netlink_fini();
4739 err_link:
4740         unregister_pernet_subsys(&bond_net_ops);
4741         goto out;
4742
4743 }
4744
4745 static void __exit bonding_exit(void)
4746 {
4747         unregister_netdevice_notifier(&bond_netdev_notifier);
4748
4749         bond_destroy_debugfs();
4750
4751         bond_netlink_fini();
4752         unregister_pernet_subsys(&bond_net_ops);
4753
4754 #ifdef CONFIG_NET_POLL_CONTROLLER
4755         /* Make sure we don't have an imbalance on our netpoll blocking */
4756         WARN_ON(atomic_read(&netpoll_block_tx));
4757 #endif
4758 }
4759
4760 module_init(bonding_init);
4761 module_exit(bonding_exit);
4762 MODULE_LICENSE("GPL");
4763 MODULE_VERSION(DRV_VERSION);
4764 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4765 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");