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 static void bond_netdev_notify_work(struct work_struct *work);
 220
 221 /*---------------------------- General routines -----------------------------*/
 222
 223 const char *bond_mode_name(int mode)
 224 {
 225         static const char *names[] = {
 226                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 227                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 228                 [BOND_MODE_XOR] = "load balancing (xor)",
 229                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 230                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 231                 [BOND_MODE_TLB] = "transmit load balancing",
 232                 [BOND_MODE_ALB] = "adaptive load balancing",
 233         };
 234
 235         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 236                 return "unknown";
 237
 238         return names[mode];
 239 }
 240
 241 /*---------------------------------- VLAN -----------------------------------*/
 242
 243 /**
 244  * bond_dev_queue_xmit - Prepare skb for xmit.
 245  *
 246  * @bond: bond device that got this skb for tx.
 247  * @skb: hw accel VLAN tagged skb to transmit
 248  * @slave_dev: slave that is supposed to xmit this skbuff
 249  */
 250 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 251                         struct net_device *slave_dev)
 252 {
 253         skb->dev = slave_dev;
 254
 255         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 256                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 257         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 258
 259         if (unlikely(netpoll_tx_running(bond->dev)))
 260                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 261         else
 262                 dev_queue_xmit(skb);
 263 }
 264
 265 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 266  * We don't protect the slave list iteration with a lock because:
 267  * a. This operation is performed in IOCTL context,
 268  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 269  * c. Holding a lock with BH disabled while directly calling a base driver
 270  *    entry point is generally a BAD idea.
 271  *
 272  * The design of synchronization/protection for this operation in the 8021q
 273  * module is good for one or more VLAN devices over a single physical device
 274  * and cannot be extended for a teaming solution like bonding, so there is a
 275  * potential race condition here where a net device from the vlan group might
 276  * be referenced (either by a base driver or the 8021q code) while it is being
 277  * removed from the system. However, it turns out we're not making matters
 278  * worse, and if it works for regular VLAN usage it will work here too.
 279 */
 280
 281 /**
 282  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 283  * @bond_dev: bonding net device that got called
 284  * @vid: vlan id being added
 285  */
 286 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 287                                 __be16 proto, u16 vid)
 288 {
 289         struct bonding *bond = netdev_priv(bond_dev);
 290         struct slave *slave, *rollback_slave;
 291         struct list_head *iter;
 292         int res;
 293
 294         bond_for_each_slave(bond, slave, iter) {
 295                 res = vlan_vid_add(slave->dev, proto, vid);
 296                 if (res)
 297                         goto unwind;
 298         }
 299
 300         return 0;
 301
 302 unwind:
 303         /* unwind to the slave that failed */
 304         bond_for_each_slave(bond, rollback_slave, iter) {
 305                 if (rollback_slave == slave)
 306                         break;
 307
 308                 vlan_vid_del(rollback_slave->dev, proto, vid);
 309         }
 310
 311         return res;
 312 }
 313
 314 /**
 315  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 316  * @bond_dev: bonding net device that got called
 317  * @vid: vlan id being removed
 318  */
 319 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 320                                  __be16 proto, u16 vid)
 321 {
 322         struct bonding *bond = netdev_priv(bond_dev);
 323         struct list_head *iter;
 324         struct slave *slave;
 325
 326         bond_for_each_slave(bond, slave, iter)
 327                 vlan_vid_del(slave->dev, proto, vid);
 328
 329         if (bond_is_lb(bond))
 330                 bond_alb_clear_vlan(bond, vid);
 331
 332         return 0;
 333 }
 334
 335 /*------------------------------- Link status -------------------------------*/
 336
 337 /* Set the carrier state for the master according to the state of its
 338  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 339  * do special 802.3ad magic.
 340  *
 341  * Returns zero if carrier state does not change, nonzero if it does.
 342  */
 343 int bond_set_carrier(struct bonding *bond)
 344 {
 345         struct list_head *iter;
 346         struct slave *slave;
 347
 348         if (!bond_has_slaves(bond))
 349                 goto down;
 350
 351         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 352                 return bond_3ad_set_carrier(bond);
 353
 354         bond_for_each_slave(bond, slave, iter) {
 355                 if (slave->link == BOND_LINK_UP) {
 356                         if (!netif_carrier_ok(bond->dev)) {
 357                                 netif_carrier_on(bond->dev);
 358                                 return 1;
 359                         }
 360                         return 0;
 361                 }
 362         }
 363
 364 down:
 365         if (netif_carrier_ok(bond->dev)) {
 366                 netif_carrier_off(bond->dev);
 367                 return 1;
 368         }
 369         return 0;
 370 }
 371
 372 /* Get link speed and duplex from the slave's base driver
 373  * using ethtool. If for some reason the call fails or the
 374  * values are invalid, set speed and duplex to -1,
 375  * and return.
 376  */
 377 static void bond_update_speed_duplex(struct slave *slave)
 378 {
 379         struct net_device *slave_dev = slave->dev;
 380         struct ethtool_cmd ecmd;
 381         u32 slave_speed;
 382         int res;
 383
 384         slave->speed = SPEED_UNKNOWN;
 385         slave->duplex = DUPLEX_UNKNOWN;
 386
 387         res = __ethtool_get_settings(slave_dev, &ecmd);
 388         if (res < 0)
 389                 return;
 390
 391         slave_speed = ethtool_cmd_speed(&ecmd);
 392         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 393                 return;
 394
 395         switch (ecmd.duplex) {
 396         case DUPLEX_FULL:
 397         case DUPLEX_HALF:
 398                 break;
 399         default:
 400                 return;
 401         }
 402
 403         slave->speed = slave_speed;
 404         slave->duplex = ecmd.duplex;
 405
 406         return;
 407 }
 408
 409 const char *bond_slave_link_status(s8 link)
 410 {
 411         switch (link) {
 412         case BOND_LINK_UP:
 413                 return "up";
 414         case BOND_LINK_FAIL:
 415                 return "going down";
 416         case BOND_LINK_DOWN:
 417                 return "down";
 418         case BOND_LINK_BACK:
 419                 return "going back";
 420         default:
 421                 return "unknown";
 422         }
 423 }
 424
 425 /* if <dev> supports MII link status reporting, check its link status.
 426  *
 427  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 428  * depending upon the setting of the use_carrier parameter.
 429  *
 430  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 431  * can't tell and just pretend it is), or 0, meaning that the link is
 432  * down.
 433  *
 434  * If reporting is non-zero, instead of faking link up, return -1 if
 435  * both ETHTOOL and MII ioctls fail (meaning the device does not
 436  * support them).  If use_carrier is set, return whatever it says.
 437  * It'd be nice if there was a good way to tell if a driver supports
 438  * netif_carrier, but there really isn't.
 439  */
 440 static int bond_check_dev_link(struct bonding *bond,
 441                                struct net_device *slave_dev, int reporting)
 442 {
 443         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 444         int (*ioctl)(struct net_device *, struct ifreq *, int);
 445         struct ifreq ifr;
 446         struct mii_ioctl_data *mii;
 447
 448         if (!reporting && !netif_running(slave_dev))
 449                 return 0;
 450
 451         if (bond->params.use_carrier)
 452                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 453
 454         /* Try to get link status using Ethtool first. */
 455         if (slave_dev->ethtool_ops->get_link)
 456                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 457                         BMSR_LSTATUS : 0;
 458
 459         /* Ethtool can't be used, fallback to MII ioctls. */
 460         ioctl = slave_ops->ndo_do_ioctl;
 461         if (ioctl) {
 462                 /* TODO: set pointer to correct ioctl on a per team member
 463                  *       bases to make this more efficient. that is, once
 464                  *       we determine the correct ioctl, we will always
 465                  *       call it and not the others for that team
 466                  *       member.
 467                  */
 468
 469                 /* We cannot assume that SIOCGMIIPHY will also read a
 470                  * register; not all network drivers (e.g., e100)
 471                  * support that.
 472                  */
 473
 474                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 475                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 476                 mii = if_mii(&ifr);
 477                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 478                         mii->reg_num = MII_BMSR;
 479                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 480                                 return mii->val_out & BMSR_LSTATUS;
 481                 }
 482         }
 483
 484         /* If reporting, report that either there's no dev->do_ioctl,
 485          * or both SIOCGMIIREG and get_link failed (meaning that we
 486          * cannot report link status).  If not reporting, pretend
 487          * we're ok.
 488          */
 489         return reporting ? -1 : BMSR_LSTATUS;
 490 }
 491
 492 /*----------------------------- Multicast list ------------------------------*/
 493
 494 /* Push the promiscuity flag down to appropriate slaves */
 495 static int bond_set_promiscuity(struct bonding *bond, int inc)
 496 {
 497         struct list_head *iter;
 498         int err = 0;
 499
 500         if (bond_uses_primary(bond)) {
 501                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 502
 503                 if (curr_active)
 504                         err = dev_set_promiscuity(curr_active->dev, inc);
 505         } else {
 506                 struct slave *slave;
 507
 508                 bond_for_each_slave(bond, slave, iter) {
 509                         err = dev_set_promiscuity(slave->dev, inc);
 510                         if (err)
 511                                 return err;
 512                 }
 513         }
 514         return err;
 515 }
 516
 517 /* Push the allmulti flag down to all slaves */
 518 static int bond_set_allmulti(struct bonding *bond, int inc)
 519 {
 520         struct list_head *iter;
 521         int err = 0;
 522
 523         if (bond_uses_primary(bond)) {
 524                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 525
 526                 if (curr_active)
 527                         err = dev_set_allmulti(curr_active->dev, inc);
 528         } else {
 529                 struct slave *slave;
 530
 531                 bond_for_each_slave(bond, slave, iter) {
 532                         err = dev_set_allmulti(slave->dev, inc);
 533                         if (err)
 534                                 return err;
 535                 }
 536         }
 537         return err;
 538 }
 539
 540 /* Retrieve the list of registered multicast addresses for the bonding
 541  * device and retransmit an IGMP JOIN request to the current active
 542  * slave.
 543  */
 544 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 545 {
 546         struct bonding *bond = container_of(work, struct bonding,
 547                                             mcast_work.work);
 548
 549         if (!rtnl_trylock()) {
 550                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 551                 return;
 552         }
 553         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 554
 555         if (bond->igmp_retrans > 1) {
 556                 bond->igmp_retrans--;
 557                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 558         }
 559         rtnl_unlock();
 560 }
 561
 562 /* Flush bond's hardware addresses from slave */
 563 static void bond_hw_addr_flush(struct net_device *bond_dev,
 564                                struct net_device *slave_dev)
 565 {
 566         struct bonding *bond = netdev_priv(bond_dev);
 567
 568         dev_uc_unsync(slave_dev, bond_dev);
 569         dev_mc_unsync(slave_dev, bond_dev);
 570
 571         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 572                 /* del lacpdu mc addr from mc list */
 573                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 574
 575                 dev_mc_del(slave_dev, lacpdu_multicast);
 576         }
 577 }
 578
 579 /*--------------------------- Active slave change ---------------------------*/
 580
 581 /* Update the hardware address list and promisc/allmulti for the new and
 582  * old active slaves (if any).  Modes that are not using primary keep all
 583  * slaves up date at all times; only the modes that use primary need to call
 584  * this function to swap these settings during a failover.
 585  */
 586 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 587                               struct slave *old_active)
 588 {
 589         if (old_active) {
 590                 if (bond->dev->flags & IFF_PROMISC)
 591                         dev_set_promiscuity(old_active->dev, -1);
 592
 593                 if (bond->dev->flags & IFF_ALLMULTI)
 594                         dev_set_allmulti(old_active->dev, -1);
 595
 596                 bond_hw_addr_flush(bond->dev, old_active->dev);
 597         }
 598
 599         if (new_active) {
 600                 /* FIXME: Signal errors upstream. */
 601                 if (bond->dev->flags & IFF_PROMISC)
 602                         dev_set_promiscuity(new_active->dev, 1);
 603
 604                 if (bond->dev->flags & IFF_ALLMULTI)
 605                         dev_set_allmulti(new_active->dev, 1);
 606
 607                 netif_addr_lock_bh(bond->dev);
 608                 dev_uc_sync(new_active->dev, bond->dev);
 609                 dev_mc_sync(new_active->dev, bond->dev);
 610                 netif_addr_unlock_bh(bond->dev);
 611         }
 612 }
 613
 614 /**
 615  * bond_set_dev_addr - clone slave's address to bond
 616  * @bond_dev: bond net device
 617  * @slave_dev: slave net device
 618  *
 619  * Should be called with RTNL held.
 620  */
 621 static void bond_set_dev_addr(struct net_device *bond_dev,
 622                               struct net_device *slave_dev)
 623 {
 624         netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 625                    bond_dev, slave_dev, slave_dev->addr_len);
 626         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 627         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 628         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 629 }
 630
 631 static struct slave *bond_get_old_active(struct bonding *bond,
 632                                          struct slave *new_active)
 633 {
 634         struct slave *slave;
 635         struct list_head *iter;
 636
 637         bond_for_each_slave(bond, slave, iter) {
 638                 if (slave == new_active)
 639                         continue;
 640
 641                 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
 642                         return slave;
 643         }
 644
 645         return NULL;
 646 }
 647
 648 /* bond_do_fail_over_mac
 649  *
 650  * Perform special MAC address swapping for fail_over_mac settings
 651  *
 652  * Called with RTNL
 653  */
 654 static void bond_do_fail_over_mac(struct bonding *bond,
 655                                   struct slave *new_active,
 656                                   struct slave *old_active)
 657 {
 658         u8 tmp_mac[ETH_ALEN];
 659         struct sockaddr saddr;
 660         int rv;
 661
 662         switch (bond->params.fail_over_mac) {
 663         case BOND_FOM_ACTIVE:
 664                 if (new_active)
 665                         bond_set_dev_addr(bond->dev, new_active->dev);
 666                 break;
 667         case BOND_FOM_FOLLOW:
 668                 /* if new_active && old_active, swap them
 669                  * if just old_active, do nothing (going to no active slave)
 670                  * if just new_active, set new_active to bond's MAC
 671                  */
 672                 if (!new_active)
 673                         return;
 674
 675                 if (!old_active)
 676                         old_active = bond_get_old_active(bond, new_active);
 677
 678                 if (old_active) {
 679                         ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
 680                         ether_addr_copy(saddr.sa_data,
 681                                         old_active->dev->dev_addr);
 682                         saddr.sa_family = new_active->dev->type;
 683                 } else {
 684                         ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
 685                         saddr.sa_family = bond->dev->type;
 686                 }
 687
 688                 rv = dev_set_mac_address(new_active->dev, &saddr);
 689                 if (rv) {
 690                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 691                                    -rv, new_active->dev->name);
 692                         goto out;
 693                 }
 694
 695                 if (!old_active)
 696                         goto out;
 697
 698                 ether_addr_copy(saddr.sa_data, tmp_mac);
 699                 saddr.sa_family = old_active->dev->type;
 700
 701                 rv = dev_set_mac_address(old_active->dev, &saddr);
 702                 if (rv)
 703                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 704                                    -rv, new_active->dev->name);
 705 out:
 706                 break;
 707         default:
 708                 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
 709                            bond->params.fail_over_mac);
 710                 break;
 711         }
 712
 713 }
 714
 715 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
 716 {
 717         struct slave *prim = rtnl_dereference(bond->primary_slave);
 718         struct slave *curr = rtnl_dereference(bond->curr_active_slave);
 719
 720         if (!prim || prim->link != BOND_LINK_UP) {
 721                 if (!curr || curr->link != BOND_LINK_UP)
 722                         return NULL;
 723                 return curr;
 724         }
 725
 726         if (bond->force_primary) {
 727                 bond->force_primary = false;
 728                 return prim;
 729         }
 730
 731         if (!curr || curr->link != BOND_LINK_UP)
 732                 return prim;
 733
 734         /* At this point, prim and curr are both up */
 735         switch (bond->params.primary_reselect) {
 736         case BOND_PRI_RESELECT_ALWAYS:
 737                 return prim;
 738         case BOND_PRI_RESELECT_BETTER:
 739                 if (prim->speed < curr->speed)
 740                         return curr;
 741                 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
 742                         return curr;
 743                 return prim;
 744         case BOND_PRI_RESELECT_FAILURE:
 745                 return curr;
 746         default:
 747                 netdev_err(bond->dev, "impossible primary_reselect %d\n",
 748                            bond->params.primary_reselect);
 749                 return curr;
 750         }
 751 }
 752
 753 /**
 754  * bond_find_best_slave - select the best available slave to be the active one
 755  * @bond: our bonding struct
 756  */
 757 static struct slave *bond_find_best_slave(struct bonding *bond)
 758 {
 759         struct slave *slave, *bestslave = NULL;
 760         struct list_head *iter;
 761         int mintime = bond->params.updelay;
 762
 763         slave = bond_choose_primary_or_current(bond);
 764         if (slave)
 765                 return slave;
 766
 767         bond_for_each_slave(bond, slave, iter) {
 768                 if (slave->link == BOND_LINK_UP)
 769                         return slave;
 770                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
 771                     slave->delay < mintime) {
 772                         mintime = slave->delay;
 773                         bestslave = slave;
 774                 }
 775         }
 776
 777         return bestslave;
 778 }
 779
 780 static bool bond_should_notify_peers(struct bonding *bond)
 781 {
 782         struct slave *slave;
 783
 784         rcu_read_lock();
 785         slave = rcu_dereference(bond->curr_active_slave);
 786         rcu_read_unlock();
 787
 788         netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
 789                    slave ? slave->dev->name : "NULL");
 790
 791         if (!slave || !bond->send_peer_notif ||
 792             !netif_carrier_ok(bond->dev) ||
 793             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 794                 return false;
 795
 796         return true;
 797 }
 798
 799 /**
 800  * change_active_interface - change the active slave into the specified one
 801  * @bond: our bonding struct
 802  * @new: the new slave to make the active one
 803  *
 804  * Set the new slave to the bond's settings and unset them on the old
 805  * curr_active_slave.
 806  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 807  *
 808  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 809  * because it is apparently the best available slave we have, even though its
 810  * updelay hasn't timed out yet.
 811  *
 812  * Caller must hold RTNL.
 813  */
 814 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 815 {
 816         struct slave *old_active;
 817
 818         ASSERT_RTNL();
 819
 820         old_active = rtnl_dereference(bond->curr_active_slave);
 821
 822         if (old_active == new_active)
 823                 return;
 824
 825         if (new_active) {
 826                 new_active->last_link_up = jiffies;
 827
 828                 if (new_active->link == BOND_LINK_BACK) {
 829                         if (bond_uses_primary(bond)) {
 830                                 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
 831                                             new_active->dev->name,
 832                                             (bond->params.updelay - new_active->delay) * bond->params.miimon);
 833                         }
 834
 835                         new_active->delay = 0;
 836                         bond_set_slave_link_state(new_active, BOND_LINK_UP);
 837
 838                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 839                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 840
 841                         if (bond_is_lb(bond))
 842                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 843                 } else {
 844                         if (bond_uses_primary(bond)) {
 845                                 netdev_info(bond->dev, "making interface %s the new active one\n",
 846                                             new_active->dev->name);
 847                         }
 848                 }
 849         }
 850
 851         if (bond_uses_primary(bond))
 852                 bond_hw_addr_swap(bond, new_active, old_active);
 853
 854         if (bond_is_lb(bond)) {
 855                 bond_alb_handle_active_change(bond, new_active);
 856                 if (old_active)
 857                         bond_set_slave_inactive_flags(old_active,
 858                                                       BOND_SLAVE_NOTIFY_NOW);
 859                 if (new_active)
 860                         bond_set_slave_active_flags(new_active,
 861                                                     BOND_SLAVE_NOTIFY_NOW);
 862         } else {
 863                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 864         }
 865
 866         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
 867                 if (old_active)
 868                         bond_set_slave_inactive_flags(old_active,
 869                                                       BOND_SLAVE_NOTIFY_NOW);
 870
 871                 if (new_active) {
 872                         bool should_notify_peers = false;
 873
 874                         bond_set_slave_active_flags(new_active,
 875                                                     BOND_SLAVE_NOTIFY_NOW);
 876
 877                         if (bond->params.fail_over_mac)
 878                                 bond_do_fail_over_mac(bond, new_active,
 879                                                       old_active);
 880
 881                         if (netif_running(bond->dev)) {
 882                                 bond->send_peer_notif =
 883                                         bond->params.num_peer_notif;
 884                                 should_notify_peers =
 885                                         bond_should_notify_peers(bond);
 886                         }
 887
 888                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 889                         if (should_notify_peers)
 890                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 891                                                          bond->dev);
 892                 }
 893         }
 894
 895         /* resend IGMP joins since active slave has changed or
 896          * all were sent on curr_active_slave.
 897          * resend only if bond is brought up with the affected
 898          * bonding modes and the retransmission is enabled
 899          */
 900         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 901             ((bond_uses_primary(bond) && new_active) ||
 902              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
 903                 bond->igmp_retrans = bond->params.resend_igmp;
 904                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 905         }
 906 }
 907
 908 /**
 909  * bond_select_active_slave - select a new active slave, if needed
 910  * @bond: our bonding struct
 911  *
 912  * This functions should be called when one of the following occurs:
 913  * - The old curr_active_slave has been released or lost its link.
 914  * - The primary_slave has got its link back.
 915  * - A slave has got its link back and there's no old curr_active_slave.
 916  *
 917  * Caller must hold RTNL.
 918  */
 919 void bond_select_active_slave(struct bonding *bond)
 920 {
 921         struct slave *best_slave;
 922         int rv;
 923
 924         ASSERT_RTNL();
 925
 926         best_slave = bond_find_best_slave(bond);
 927         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
 928                 bond_change_active_slave(bond, best_slave);
 929                 rv = bond_set_carrier(bond);
 930                 if (!rv)
 931                         return;
 932
 933                 if (netif_carrier_ok(bond->dev)) {
 934                         netdev_info(bond->dev, "first active interface up!\n");
 935                 } else {
 936                         netdev_info(bond->dev, "now running without any active interface!\n");
 937                 }
 938         }
 939 }
 940
 941 #ifdef CONFIG_NET_POLL_CONTROLLER
 942 static inline int slave_enable_netpoll(struct slave *slave)
 943 {
 944         struct netpoll *np;
 945         int err = 0;
 946
 947         np = kzalloc(sizeof(*np), GFP_KERNEL);
 948         err = -ENOMEM;
 949         if (!np)
 950                 goto out;
 951
 952         err = __netpoll_setup(np, slave->dev);
 953         if (err) {
 954                 kfree(np);
 955                 goto out;
 956         }
 957         slave->np = np;
 958 out:
 959         return err;
 960 }
 961 static inline void slave_disable_netpoll(struct slave *slave)
 962 {
 963         struct netpoll *np = slave->np;
 964
 965         if (!np)
 966                 return;
 967
 968         slave->np = NULL;
 969         __netpoll_free_async(np);
 970 }
 971
 972 static void bond_poll_controller(struct net_device *bond_dev)
 973 {
 974         struct bonding *bond = netdev_priv(bond_dev);
 975         struct slave *slave = NULL;
 976         struct list_head *iter;
 977         struct ad_info ad_info;
 978         struct netpoll_info *ni;
 979         const struct net_device_ops *ops;
 980
 981         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 982                 if (bond_3ad_get_active_agg_info(bond, &ad_info))
 983                         return;
 984
 985         bond_for_each_slave_rcu(bond, slave, iter) {
 986                 ops = slave->dev->netdev_ops;
 987                 if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller)
 988                         continue;
 989
 990                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 991                         struct aggregator *agg =
 992                             SLAVE_AD_INFO(slave)->port.aggregator;
 993
 994                         if (agg &&
 995                             agg->aggregator_identifier != ad_info.aggregator_id)
 996                                 continue;
 997                 }
 998
 999                 ni = rcu_dereference_bh(slave->dev->npinfo);
1000                 if (down_trylock(&ni->dev_lock))
1001                         continue;
1002                 ops->ndo_poll_controller(slave->dev);
1003                 up(&ni->dev_lock);
1004         }
1005 }
1006
1007 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1008 {
1009         struct bonding *bond = netdev_priv(bond_dev);
1010         struct list_head *iter;
1011         struct slave *slave;
1012
1013         bond_for_each_slave(bond, slave, iter)
1014                 if (bond_slave_is_up(slave))
1015                         slave_disable_netpoll(slave);
1016 }
1017
1018 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1019 {
1020         struct bonding *bond = netdev_priv(dev);
1021         struct list_head *iter;
1022         struct slave *slave;
1023         int err = 0;
1024
1025         bond_for_each_slave(bond, slave, iter) {
1026                 err = slave_enable_netpoll(slave);
1027                 if (err) {
1028                         bond_netpoll_cleanup(dev);
1029                         break;
1030                 }
1031         }
1032         return err;
1033 }
1034 #else
1035 static inline int slave_enable_netpoll(struct slave *slave)
1036 {
1037         return 0;
1038 }
1039 static inline void slave_disable_netpoll(struct slave *slave)
1040 {
1041 }
1042 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1043 {
1044 }
1045 #endif
1046
1047 /*---------------------------------- IOCTL ----------------------------------*/
1048
1049 static netdev_features_t bond_fix_features(struct net_device *dev,
1050                                            netdev_features_t features)
1051 {
1052         struct bonding *bond = netdev_priv(dev);
1053         struct list_head *iter;
1054         netdev_features_t mask;
1055         struct slave *slave;
1056
1057         mask = features;
1058
1059         features &= ~NETIF_F_ONE_FOR_ALL;
1060         features |= NETIF_F_ALL_FOR_ALL;
1061
1062         bond_for_each_slave(bond, slave, iter) {
1063                 features = netdev_increment_features(features,
1064                                                      slave->dev->features,
1065                                                      mask);
1066         }
1067         features = netdev_add_tso_features(features, mask);
1068
1069         return features;
1070 }
1071
1072 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1073                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1074                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1075
1076 #define BOND_ENC_FEATURES       (NETIF_F_ALL_CSUM | NETIF_F_SG | NETIF_F_RXCSUM |\
1077                                  NETIF_F_ALL_TSO)
1078
1079 static void bond_compute_features(struct bonding *bond)
1080 {
1081         unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1082                                         IFF_XMIT_DST_RELEASE_PERM;
1083         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1084         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1085         struct net_device *bond_dev = bond->dev;
1086         struct list_head *iter;
1087         struct slave *slave;
1088         unsigned short max_hard_header_len = ETH_HLEN;
1089         unsigned int gso_max_size = GSO_MAX_SIZE;
1090         u16 gso_max_segs = GSO_MAX_SEGS;
1091
1092         if (!bond_has_slaves(bond))
1093                 goto done;
1094         vlan_features &= NETIF_F_ALL_FOR_ALL;
1095
1096         bond_for_each_slave(bond, slave, iter) {
1097                 vlan_features = netdev_increment_features(vlan_features,
1098                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1099
1100                 enc_features = netdev_increment_features(enc_features,
1101                                                          slave->dev->hw_enc_features,
1102                                                          BOND_ENC_FEATURES);
1103                 dst_release_flag &= slave->dev->priv_flags;
1104                 if (slave->dev->hard_header_len > max_hard_header_len)
1105                         max_hard_header_len = slave->dev->hard_header_len;
1106
1107                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1108                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1109         }
1110
1111 done:
1112         bond_dev->vlan_features = vlan_features;
1113         bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
1114         bond_dev->hard_header_len = max_hard_header_len;
1115         bond_dev->gso_max_segs = gso_max_segs;
1116         netif_set_gso_max_size(bond_dev, gso_max_size);
1117
1118         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1119         if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1120             dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1121                 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1122
1123         netdev_change_features(bond_dev);
1124 }
1125
1126 static void bond_setup_by_slave(struct net_device *bond_dev,
1127                                 struct net_device *slave_dev)
1128 {
1129         bond_dev->header_ops        = slave_dev->header_ops;
1130
1131         bond_dev->type              = slave_dev->type;
1132         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1133         bond_dev->addr_len          = slave_dev->addr_len;
1134
1135         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1136                 slave_dev->addr_len);
1137 }
1138
1139 /* On bonding slaves other than the currently active slave, suppress
1140  * duplicates except for alb non-mcast/bcast.
1141  */
1142 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1143                                             struct slave *slave,
1144                                             struct bonding *bond)
1145 {
1146         if (bond_is_slave_inactive(slave)) {
1147                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1148                     skb->pkt_type != PACKET_BROADCAST &&
1149                     skb->pkt_type != PACKET_MULTICAST)
1150                         return false;
1151                 return true;
1152         }
1153         return false;
1154 }
1155
1156 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1157 {
1158         struct sk_buff *skb = *pskb;
1159         struct slave *slave;
1160         struct bonding *bond;
1161         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1162                           struct slave *);
1163         int ret = RX_HANDLER_ANOTHER;
1164
1165         skb = skb_share_check(skb, GFP_ATOMIC);
1166         if (unlikely(!skb))
1167                 return RX_HANDLER_CONSUMED;
1168
1169         *pskb = skb;
1170
1171         slave = bond_slave_get_rcu(skb->dev);
1172         bond = slave->bond;
1173
1174         recv_probe = ACCESS_ONCE(bond->recv_probe);
1175         if (recv_probe) {
1176                 ret = recv_probe(skb, bond, slave);
1177                 if (ret == RX_HANDLER_CONSUMED) {
1178                         consume_skb(skb);
1179                         return ret;
1180                 }
1181         }
1182
1183         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1184                 return RX_HANDLER_EXACT;
1185         }
1186
1187         skb->dev = bond->dev;
1188
1189         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1190             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1191             skb->pkt_type == PACKET_HOST) {
1192
1193                 if (unlikely(skb_cow_head(skb,
1194                                           skb->data - skb_mac_header(skb)))) {
1195                         kfree_skb(skb);
1196                         return RX_HANDLER_CONSUMED;
1197                 }
1198                 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1199         }
1200
1201         return ret;
1202 }
1203
1204 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1205                                       struct net_device *slave_dev,
1206                                       struct slave *slave)
1207 {
1208         int err;
1209
1210         err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1211         if (err)
1212                 return err;
1213         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1214         return 0;
1215 }
1216
1217 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1218                                   struct net_device *slave_dev)
1219 {
1220         netdev_upper_dev_unlink(slave_dev, bond_dev);
1221         slave_dev->flags &= ~IFF_SLAVE;
1222         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1223 }
1224
1225 static struct slave *bond_alloc_slave(struct bonding *bond)
1226 {
1227         struct slave *slave = NULL;
1228
1229         slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1230         if (!slave)
1231                 return NULL;
1232
1233         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1234                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1235                                                GFP_KERNEL);
1236                 if (!SLAVE_AD_INFO(slave)) {
1237                         kfree(slave);
1238                         return NULL;
1239                 }
1240         }
1241         INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1242
1243         return slave;
1244 }
1245
1246 static void bond_free_slave(struct slave *slave)
1247 {
1248         struct bonding *bond = bond_get_bond_by_slave(slave);
1249
1250         cancel_delayed_work_sync(&slave->notify_work);
1251         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1252                 kfree(SLAVE_AD_INFO(slave));
1253
1254         kfree(slave);
1255 }
1256
1257 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1258 {
1259         info->bond_mode = BOND_MODE(bond);
1260         info->miimon = bond->params.miimon;
1261         info->num_slaves = bond->slave_cnt;
1262 }
1263
1264 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1265 {
1266         strcpy(info->slave_name, slave->dev->name);
1267         info->link = slave->link;
1268         info->state = bond_slave_state(slave);
1269         info->link_failure_count = slave->link_failure_count;
1270 }
1271
1272 static void bond_netdev_notify_work(struct work_struct *_work)
1273 {
1274         struct slave *slave = container_of(_work, struct slave,
1275                                            notify_work.work);
1276
1277         if (rtnl_trylock()) {
1278                 struct netdev_bonding_info binfo;
1279
1280                 bond_fill_ifslave(slave, &binfo.slave);
1281                 bond_fill_ifbond(slave->bond, &binfo.master);
1282                 netdev_bonding_info_change(slave->dev, &binfo);
1283                 rtnl_unlock();
1284         } else {
1285                 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1286         }
1287 }
1288
1289 void bond_queue_slave_event(struct slave *slave)
1290 {
1291         queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1292 }
1293
1294 /* enslave device <slave> to bond device <master> */
1295 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1296 {
1297         struct bonding *bond = netdev_priv(bond_dev);
1298         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1299         struct slave *new_slave = NULL, *prev_slave;
1300         struct sockaddr addr;
1301         int link_reporting;
1302         int res = 0, i;
1303
1304         if (!bond->params.use_carrier &&
1305             slave_dev->ethtool_ops->get_link == NULL &&
1306             slave_ops->ndo_do_ioctl == NULL) {
1307                 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1308                             slave_dev->name);
1309         }
1310
1311         /* already in-use? */
1312         if (netdev_is_rx_handler_busy(slave_dev)) {
1313                 netdev_err(bond_dev,
1314                            "Error: Device is in use and cannot be enslaved\n");
1315                 return -EBUSY;
1316         }
1317
1318         if (bond_dev == slave_dev) {
1319                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1320                 return -EPERM;
1321         }
1322
1323         /* vlan challenged mutual exclusion */
1324         /* no need to lock since we're protected by rtnl_lock */
1325         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1326                 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1327                            slave_dev->name);
1328                 if (vlan_uses_dev(bond_dev)) {
1329                         netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1330                                    slave_dev->name, bond_dev->name);
1331                         return -EPERM;
1332                 } else {
1333                         netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1334                                     slave_dev->name, slave_dev->name,
1335                                     bond_dev->name);
1336                 }
1337         } else {
1338                 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1339                            slave_dev->name);
1340         }
1341
1342         /* Old ifenslave binaries are no longer supported.  These can
1343          * be identified with moderate accuracy by the state of the slave:
1344          * the current ifenslave will set the interface down prior to
1345          * enslaving it; the old ifenslave will not.
1346          */
1347         if ((slave_dev->flags & IFF_UP)) {
1348                 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1349                            slave_dev->name);
1350                 res = -EPERM;
1351                 goto err_undo_flags;
1352         }
1353
1354         /* set bonding device ether type by slave - bonding netdevices are
1355          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1356          * there is a need to override some of the type dependent attribs/funcs.
1357          *
1358          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1359          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1360          */
1361         if (!bond_has_slaves(bond)) {
1362                 if (bond_dev->type != slave_dev->type) {
1363                         netdev_dbg(bond_dev, "change device type from %d to %d\n",
1364                                    bond_dev->type, slave_dev->type);
1365
1366                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1367                                                        bond_dev);
1368                         res = notifier_to_errno(res);
1369                         if (res) {
1370                                 netdev_err(bond_dev, "refused to change device type\n");
1371                                 res = -EBUSY;
1372                                 goto err_undo_flags;
1373                         }
1374
1375                         /* Flush unicast and multicast addresses */
1376                         dev_uc_flush(bond_dev);
1377                         dev_mc_flush(bond_dev);
1378
1379                         if (slave_dev->type != ARPHRD_ETHER)
1380                                 bond_setup_by_slave(bond_dev, slave_dev);
1381                         else {
1382                                 ether_setup(bond_dev);
1383                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1384                         }
1385
1386                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1387                                                  bond_dev);
1388                 }
1389         } else if (bond_dev->type != slave_dev->type) {
1390                 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1391                            slave_dev->name, slave_dev->type, bond_dev->type);
1392                 res = -EINVAL;
1393                 goto err_undo_flags;
1394         }
1395
1396         if (slave_ops->ndo_set_mac_address == NULL) {
1397                 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1398                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1399                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1400                         if (!bond_has_slaves(bond)) {
1401                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1402                                 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1403                         } else {
1404                                 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");
1405                                 res = -EOPNOTSUPP;
1406                                 goto err_undo_flags;
1407                         }
1408                 }
1409         }
1410
1411         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1412
1413         /* If this is the first slave, then we need to set the master's hardware
1414          * address to be the same as the slave's.
1415          */
1416         if (!bond_has_slaves(bond) &&
1417             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1418                 bond_set_dev_addr(bond->dev, slave_dev);
1419
1420         new_slave = bond_alloc_slave(bond);
1421         if (!new_slave) {
1422                 res = -ENOMEM;
1423                 goto err_undo_flags;
1424         }
1425
1426         new_slave->bond = bond;
1427         new_slave->dev = slave_dev;
1428         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1429          * is set via sysfs or module option if desired.
1430          */
1431         new_slave->queue_id = 0;
1432
1433         /* Save slave's original mtu and then set it to match the bond */
1434         new_slave->original_mtu = slave_dev->mtu;
1435         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1436         if (res) {
1437                 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1438                 goto err_free;
1439         }
1440
1441         /* Save slave's original ("permanent") mac address for modes
1442          * that need it, and for restoring it upon release, and then
1443          * set it to the master's address
1444          */
1445         ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1446
1447         if (!bond->params.fail_over_mac ||
1448             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1449                 /* Set slave to master's mac address.  The application already
1450                  * set the master's mac address to that of the first slave
1451                  */
1452                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1453                 addr.sa_family = slave_dev->type;
1454                 res = dev_set_mac_address(slave_dev, &addr);
1455                 if (res) {
1456                         netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1457                         goto err_restore_mtu;
1458                 }
1459         }
1460
1461         /* set slave flag before open to prevent IPv6 addrconf */
1462         slave_dev->flags |= IFF_SLAVE;
1463
1464         /* open the slave since the application closed it */
1465         res = dev_open(slave_dev);
1466         if (res) {
1467                 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1468                 goto err_restore_mac;
1469         }
1470
1471         slave_dev->priv_flags |= IFF_BONDING;
1472         /* initialize slave stats */
1473         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1474
1475         if (bond_is_lb(bond)) {
1476                 /* bond_alb_init_slave() must be called before all other stages since
1477                  * it might fail and we do not want to have to undo everything
1478                  */
1479                 res = bond_alb_init_slave(bond, new_slave);
1480                 if (res)
1481                         goto err_close;
1482         }
1483
1484         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1485         if (res) {
1486                 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1487                            slave_dev->name);
1488                 goto err_close;
1489         }
1490
1491         prev_slave = bond_last_slave(bond);
1492
1493         new_slave->delay = 0;
1494         new_slave->link_failure_count = 0;
1495
1496         bond_update_speed_duplex(new_slave);
1497
1498         new_slave->last_rx = jiffies -
1499                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1500         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1501                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1502
1503         if (bond->params.miimon && !bond->params.use_carrier) {
1504                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1505
1506                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1507                         /* miimon is set but a bonded network driver
1508                          * does not support ETHTOOL/MII and
1509                          * arp_interval is not set.  Note: if
1510                          * use_carrier is enabled, we will never go
1511                          * here (because netif_carrier is always
1512                          * supported); thus, we don't need to change
1513                          * the messages for netif_carrier.
1514                          */
1515                         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",
1516                                     slave_dev->name);
1517                 } else if (link_reporting == -1) {
1518                         /* unable get link status using mii/ethtool */
1519                         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",
1520                                     slave_dev->name);
1521                 }
1522         }
1523
1524         /* check for initial state */
1525         if (bond->params.miimon) {
1526                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1527                         if (bond->params.updelay) {
1528                                 bond_set_slave_link_state(new_slave,
1529                                                           BOND_LINK_BACK);
1530                                 new_slave->delay = bond->params.updelay;
1531                         } else {
1532                                 bond_set_slave_link_state(new_slave,
1533                                                           BOND_LINK_UP);
1534                         }
1535                 } else {
1536                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN);
1537                 }
1538         } else if (bond->params.arp_interval) {
1539                 bond_set_slave_link_state(new_slave,
1540                                           (netif_carrier_ok(slave_dev) ?
1541                                           BOND_LINK_UP : BOND_LINK_DOWN));
1542         } else {
1543                 bond_set_slave_link_state(new_slave, BOND_LINK_UP);
1544         }
1545
1546         if (new_slave->link != BOND_LINK_DOWN)
1547                 new_slave->last_link_up = jiffies;
1548         netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1549                    new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1550                    (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1551
1552         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1553                 /* if there is a primary slave, remember it */
1554                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1555                         rcu_assign_pointer(bond->primary_slave, new_slave);
1556                         bond->force_primary = true;
1557                 }
1558         }
1559
1560         switch (BOND_MODE(bond)) {
1561         case BOND_MODE_ACTIVEBACKUP:
1562                 bond_set_slave_inactive_flags(new_slave,
1563                                               BOND_SLAVE_NOTIFY_NOW);
1564                 break;
1565         case BOND_MODE_8023AD:
1566                 /* in 802.3ad mode, the internal mechanism
1567                  * will activate the slaves in the selected
1568                  * aggregator
1569                  */
1570                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1571                 /* if this is the first slave */
1572                 if (!prev_slave) {
1573                         SLAVE_AD_INFO(new_slave)->id = 1;
1574                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1575                          * can be called only after the mac address of the bond is set
1576                          */
1577                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1578                 } else {
1579                         SLAVE_AD_INFO(new_slave)->id =
1580                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1581                 }
1582
1583                 bond_3ad_bind_slave(new_slave);
1584                 break;
1585         case BOND_MODE_TLB:
1586         case BOND_MODE_ALB:
1587                 bond_set_active_slave(new_slave);
1588                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1589                 break;
1590         default:
1591                 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1592
1593                 /* always active in trunk mode */
1594                 bond_set_active_slave(new_slave);
1595
1596                 /* In trunking mode there is little meaning to curr_active_slave
1597                  * anyway (it holds no special properties of the bond device),
1598                  * so we can change it without calling change_active_interface()
1599                  */
1600                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1601                     new_slave->link == BOND_LINK_UP)
1602                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1603
1604                 break;
1605         } /* switch(bond_mode) */
1606
1607 #ifdef CONFIG_NET_POLL_CONTROLLER
1608         if (bond->dev->npinfo) {
1609                 if (slave_enable_netpoll(new_slave)) {
1610                         netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1611                         res = -EBUSY;
1612                         goto err_detach;
1613                 }
1614         }
1615 #endif
1616
1617         if (!(bond_dev->features & NETIF_F_LRO))
1618                 dev_disable_lro(slave_dev);
1619
1620         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1621                                          new_slave);
1622         if (res) {
1623                 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1624                 goto err_detach;
1625         }
1626
1627         res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1628         if (res) {
1629                 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1630                 goto err_unregister;
1631         }
1632
1633         res = bond_sysfs_slave_add(new_slave);
1634         if (res) {
1635                 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1636                 goto err_upper_unlink;
1637         }
1638
1639         /* If the mode uses primary, then the following is handled by
1640          * bond_change_active_slave().
1641          */
1642         if (!bond_uses_primary(bond)) {
1643                 /* set promiscuity level to new slave */
1644                 if (bond_dev->flags & IFF_PROMISC) {
1645                         res = dev_set_promiscuity(slave_dev, 1);
1646                         if (res)
1647                                 goto err_sysfs_del;
1648                 }
1649
1650                 /* set allmulti level to new slave */
1651                 if (bond_dev->flags & IFF_ALLMULTI) {
1652                         res = dev_set_allmulti(slave_dev, 1);
1653                         if (res) {
1654                                 if (bond_dev->flags & IFF_PROMISC)
1655                                         dev_set_promiscuity(slave_dev, -1);
1656                                 goto err_sysfs_del;
1657                         }
1658                 }
1659
1660                 netif_addr_lock_bh(bond_dev);
1661                 dev_mc_sync_multiple(slave_dev, bond_dev);
1662                 dev_uc_sync_multiple(slave_dev, bond_dev);
1663                 netif_addr_unlock_bh(bond_dev);
1664
1665                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1666                         /* add lacpdu mc addr to mc list */
1667                         u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1668
1669                         dev_mc_add(slave_dev, lacpdu_multicast);
1670                 }
1671         }
1672
1673         bond->slave_cnt++;
1674         bond_compute_features(bond);
1675         bond_set_carrier(bond);
1676
1677         if (bond_uses_primary(bond)) {
1678                 block_netpoll_tx();
1679                 bond_select_active_slave(bond);
1680                 unblock_netpoll_tx();
1681         }
1682
1683         if (bond_mode_uses_xmit_hash(bond))
1684                 bond_update_slave_arr(bond, NULL);
1685
1686         netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1687                     slave_dev->name,
1688                     bond_is_active_slave(new_slave) ? "an active" : "a backup",
1689                     new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1690
1691         /* enslave is successful */
1692         bond_queue_slave_event(new_slave);
1693         return 0;
1694
1695 /* Undo stages on error */
1696 err_sysfs_del:
1697         bond_sysfs_slave_del(new_slave);
1698
1699 err_upper_unlink:
1700         bond_upper_dev_unlink(bond_dev, slave_dev);
1701
1702 err_unregister:
1703         netdev_rx_handler_unregister(slave_dev);
1704
1705 err_detach:
1706         vlan_vids_del_by_dev(slave_dev, bond_dev);
1707         if (rcu_access_pointer(bond->primary_slave) == new_slave)
1708                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1709         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1710                 block_netpoll_tx();
1711                 bond_change_active_slave(bond, NULL);
1712                 bond_select_active_slave(bond);
1713                 unblock_netpoll_tx();
1714         }
1715         /* either primary_slave or curr_active_slave might've changed */
1716         synchronize_rcu();
1717         slave_disable_netpoll(new_slave);
1718
1719 err_close:
1720         slave_dev->priv_flags &= ~IFF_BONDING;
1721         dev_close(slave_dev);
1722
1723 err_restore_mac:
1724         slave_dev->flags &= ~IFF_SLAVE;
1725         if (!bond->params.fail_over_mac ||
1726             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1727                 /* XXX TODO - fom follow mode needs to change master's
1728                  * MAC if this slave's MAC is in use by the bond, or at
1729                  * least print a warning.
1730                  */
1731                 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1732                 addr.sa_family = slave_dev->type;
1733                 dev_set_mac_address(slave_dev, &addr);
1734         }
1735
1736 err_restore_mtu:
1737         dev_set_mtu(slave_dev, new_slave->original_mtu);
1738
1739 err_free:
1740         bond_free_slave(new_slave);
1741
1742 err_undo_flags:
1743         /* Enslave of first slave has failed and we need to fix master's mac */
1744         if (!bond_has_slaves(bond)) {
1745                 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1746                                             slave_dev->dev_addr))
1747                         eth_hw_addr_random(bond_dev);
1748                 if (bond_dev->type != ARPHRD_ETHER) {
1749                         dev_close(bond_dev);
1750                         ether_setup(bond_dev);
1751                         bond_dev->flags |= IFF_MASTER;
1752                         bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1753                 }
1754         }
1755
1756         return res;
1757 }
1758
1759 /* Try to release the slave device <slave> from the bond device <master>
1760  * It is legal to access curr_active_slave without a lock because all the function
1761  * is RTNL-locked. If "all" is true it means that the function is being called
1762  * while destroying a bond interface and all slaves are being released.
1763  *
1764  * The rules for slave state should be:
1765  *   for Active/Backup:
1766  *     Active stays on all backups go down
1767  *   for Bonded connections:
1768  *     The first up interface should be left on and all others downed.
1769  */
1770 static int __bond_release_one(struct net_device *bond_dev,
1771                               struct net_device *slave_dev,
1772                               bool all)
1773 {
1774         struct bonding *bond = netdev_priv(bond_dev);
1775         struct slave *slave, *oldcurrent;
1776         struct sockaddr addr;
1777         int old_flags = bond_dev->flags;
1778         netdev_features_t old_features = bond_dev->features;
1779
1780         /* slave is not a slave or master is not master of this slave */
1781         if (!(slave_dev->flags & IFF_SLAVE) ||
1782             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1783                 netdev_dbg(bond_dev, "cannot release %s\n",
1784                            slave_dev->name);
1785                 return -EINVAL;
1786         }
1787
1788         block_netpoll_tx();
1789
1790         slave = bond_get_slave_by_dev(bond, slave_dev);
1791         if (!slave) {
1792                 /* not a slave of this bond */
1793                 netdev_info(bond_dev, "%s not enslaved\n",
1794                             slave_dev->name);
1795                 unblock_netpoll_tx();
1796                 return -EINVAL;
1797         }
1798
1799         bond_sysfs_slave_del(slave);
1800
1801         /* recompute stats just before removing the slave */
1802         bond_get_stats(bond->dev, &bond->bond_stats);
1803
1804         bond_upper_dev_unlink(bond_dev, slave_dev);
1805         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1806          * for this slave anymore.
1807          */
1808         netdev_rx_handler_unregister(slave_dev);
1809
1810         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1811                 bond_3ad_unbind_slave(slave);
1812
1813         if (bond_mode_uses_xmit_hash(bond))
1814                 bond_update_slave_arr(bond, slave);
1815
1816         netdev_info(bond_dev, "Releasing %s interface %s\n",
1817                     bond_is_active_slave(slave) ? "active" : "backup",
1818                     slave_dev->name);
1819
1820         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1821
1822         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1823
1824         if (!all && (!bond->params.fail_over_mac ||
1825                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1826                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1827                     bond_has_slaves(bond))
1828                         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",
1829                                     slave_dev->name, slave->perm_hwaddr,
1830                                     bond_dev->name, slave_dev->name);
1831         }
1832
1833         if (rtnl_dereference(bond->primary_slave) == slave)
1834                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1835
1836         if (oldcurrent == slave)
1837                 bond_change_active_slave(bond, NULL);
1838
1839         if (bond_is_lb(bond)) {
1840                 /* Must be called only after the slave has been
1841                  * detached from the list and the curr_active_slave
1842                  * has been cleared (if our_slave == old_current),
1843                  * but before a new active slave is selected.
1844                  */
1845                 bond_alb_deinit_slave(bond, slave);
1846         }
1847
1848         if (all) {
1849                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1850         } else if (oldcurrent == slave) {
1851                 /* Note that we hold RTNL over this sequence, so there
1852                  * is no concern that another slave add/remove event
1853                  * will interfere.
1854                  */
1855                 bond_select_active_slave(bond);
1856         }
1857
1858         if (!bond_has_slaves(bond)) {
1859                 bond_set_carrier(bond);
1860                 eth_hw_addr_random(bond_dev);
1861         }
1862
1863         unblock_netpoll_tx();
1864         synchronize_rcu();
1865         bond->slave_cnt--;
1866
1867         if (!bond_has_slaves(bond)) {
1868                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1869                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1870         }
1871
1872         bond_compute_features(bond);
1873         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1874             (old_features & NETIF_F_VLAN_CHALLENGED))
1875                 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1876                             slave_dev->name, bond_dev->name);
1877
1878         vlan_vids_del_by_dev(slave_dev, bond_dev);
1879
1880         /* If the mode uses primary, then this case was handled above by
1881          * bond_change_active_slave(..., NULL)
1882          */
1883         if (!bond_uses_primary(bond)) {
1884                 /* unset promiscuity level from slave
1885                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1886                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1887                  * value of that flag before that change, as that was the value
1888                  * when this slave was attached, so we cache at the start of the
1889                  * function and use it here. Same goes for ALLMULTI below
1890                  */
1891                 if (old_flags & IFF_PROMISC)
1892                         dev_set_promiscuity(slave_dev, -1);
1893
1894                 /* unset allmulti level from slave */
1895                 if (old_flags & IFF_ALLMULTI)
1896                         dev_set_allmulti(slave_dev, -1);
1897
1898                 bond_hw_addr_flush(bond_dev, slave_dev);
1899         }
1900
1901         slave_disable_netpoll(slave);
1902
1903         /* close slave before restoring its mac address */
1904         dev_close(slave_dev);
1905
1906         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1907             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1908                 /* restore original ("permanent") mac address */
1909                 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1910                 addr.sa_family = slave_dev->type;
1911                 dev_set_mac_address(slave_dev, &addr);
1912         }
1913
1914         dev_set_mtu(slave_dev, slave->original_mtu);
1915
1916         slave_dev->priv_flags &= ~IFF_BONDING;
1917
1918         bond_free_slave(slave);
1919
1920         return 0;
1921 }
1922
1923 /* A wrapper used because of ndo_del_link */
1924 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1925 {
1926         return __bond_release_one(bond_dev, slave_dev, false);
1927 }
1928
1929 /* First release a slave and then destroy the bond if no more slaves are left.
1930  * Must be under rtnl_lock when this function is called.
1931  */
1932 static int  bond_release_and_destroy(struct net_device *bond_dev,
1933                                      struct net_device *slave_dev)
1934 {
1935         struct bonding *bond = netdev_priv(bond_dev);
1936         int ret;
1937
1938         ret = bond_release(bond_dev, slave_dev);
1939         if (ret == 0 && !bond_has_slaves(bond)) {
1940                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1941                 netdev_info(bond_dev, "Destroying bond %s\n",
1942                             bond_dev->name);
1943                 bond_remove_proc_entry(bond);
1944                 unregister_netdevice(bond_dev);
1945         }
1946         return ret;
1947 }
1948
1949 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1950 {
1951         struct bonding *bond = netdev_priv(bond_dev);
1952         bond_fill_ifbond(bond, info);
1953         return 0;
1954 }
1955
1956 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1957 {
1958         struct bonding *bond = netdev_priv(bond_dev);
1959         struct list_head *iter;
1960         int i = 0, res = -ENODEV;
1961         struct slave *slave;
1962
1963         bond_for_each_slave(bond, slave, iter) {
1964                 if (i++ == (int)info->slave_id) {
1965                         res = 0;
1966                         bond_fill_ifslave(slave, info);
1967                         break;
1968                 }
1969         }
1970
1971         return res;
1972 }
1973
1974 /*-------------------------------- Monitoring -------------------------------*/
1975
1976 /* called with rcu_read_lock() */
1977 static int bond_miimon_inspect(struct bonding *bond)
1978 {
1979         int link_state, commit = 0;
1980         struct list_head *iter;
1981         struct slave *slave;
1982         bool ignore_updelay;
1983
1984         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
1985
1986         bond_for_each_slave_rcu(bond, slave, iter) {
1987                 slave->new_link = BOND_LINK_NOCHANGE;
1988
1989                 link_state = bond_check_dev_link(bond, slave->dev, 0);
1990
1991                 switch (slave->link) {
1992                 case BOND_LINK_UP:
1993                         if (link_state)
1994                                 continue;
1995
1996                         bond_set_slave_link_state(slave, BOND_LINK_FAIL);
1997                         slave->delay = bond->params.downdelay;
1998                         if (slave->delay) {
1999                                 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2000                                             (BOND_MODE(bond) ==
2001                                              BOND_MODE_ACTIVEBACKUP) ?
2002                                              (bond_is_active_slave(slave) ?
2003                                               "active " : "backup ") : "",
2004                                             slave->dev->name,
2005                                             bond->params.downdelay * bond->params.miimon);
2006                         }
2007                         /*FALLTHRU*/
2008                 case BOND_LINK_FAIL:
2009                         if (link_state) {
2010                                 /* recovered before downdelay expired */
2011                                 bond_set_slave_link_state(slave, BOND_LINK_UP);
2012                                 slave->last_link_up = jiffies;
2013                                 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2014                                             (bond->params.downdelay - slave->delay) *
2015                                             bond->params.miimon,
2016                                             slave->dev->name);
2017                                 continue;
2018                         }
2019
2020                         if (slave->delay <= 0) {
2021                                 slave->new_link = BOND_LINK_DOWN;
2022                                 commit++;
2023                                 continue;
2024                         }
2025
2026                         slave->delay--;
2027                         break;
2028
2029                 case BOND_LINK_DOWN:
2030                         if (!link_state)
2031                                 continue;
2032
2033                         bond_set_slave_link_state(slave, BOND_LINK_BACK);
2034                         slave->delay = bond->params.updelay;
2035
2036                         if (slave->delay) {
2037                                 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2038                                             slave->dev->name,
2039                                             ignore_updelay ? 0 :
2040                                             bond->params.updelay *
2041                                             bond->params.miimon);
2042                         }
2043                         /*FALLTHRU*/
2044                 case BOND_LINK_BACK:
2045                         if (!link_state) {
2046                                 bond_set_slave_link_state(slave,
2047                                                           BOND_LINK_DOWN);
2048                                 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2049                                             (bond->params.updelay - slave->delay) *
2050                                             bond->params.miimon,
2051                                             slave->dev->name);
2052
2053                                 continue;
2054                         }
2055
2056                         if (ignore_updelay)
2057                                 slave->delay = 0;
2058
2059                         if (slave->delay <= 0) {
2060                                 slave->new_link = BOND_LINK_UP;
2061                                 commit++;
2062                                 ignore_updelay = false;
2063                                 continue;
2064                         }
2065
2066                         slave->delay--;
2067                         break;
2068                 }
2069         }
2070
2071         return commit;
2072 }
2073
2074 static void bond_miimon_commit(struct bonding *bond)
2075 {
2076         struct list_head *iter;
2077         struct slave *slave, *primary;
2078
2079         bond_for_each_slave(bond, slave, iter) {
2080                 switch (slave->new_link) {
2081                 case BOND_LINK_NOCHANGE:
2082                         continue;
2083
2084                 case BOND_LINK_UP:
2085                         bond_set_slave_link_state(slave, BOND_LINK_UP);
2086                         slave->last_link_up = jiffies;
2087
2088                         primary = rtnl_dereference(bond->primary_slave);
2089                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2090                                 /* prevent it from being the active one */
2091                                 bond_set_backup_slave(slave);
2092                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2093                                 /* make it immediately active */
2094                                 bond_set_active_slave(slave);
2095                         } else if (slave != primary) {
2096                                 /* prevent it from being the active one */
2097                                 bond_set_backup_slave(slave);
2098                         }
2099
2100                         netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2101                                     slave->dev->name,
2102                                     slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2103                                     slave->duplex ? "full" : "half");
2104
2105                         /* notify ad that the link status has changed */
2106                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2107                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2108
2109                         if (bond_is_lb(bond))
2110                                 bond_alb_handle_link_change(bond, slave,
2111                                                             BOND_LINK_UP);
2112
2113                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2114                                 bond_update_slave_arr(bond, NULL);
2115
2116                         if (!bond->curr_active_slave || slave == primary)
2117                                 goto do_failover;
2118
2119                         continue;
2120
2121                 case BOND_LINK_DOWN:
2122                         if (slave->link_failure_count < UINT_MAX)
2123                                 slave->link_failure_count++;
2124
2125                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2126
2127                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2128                             BOND_MODE(bond) == BOND_MODE_8023AD)
2129                                 bond_set_slave_inactive_flags(slave,
2130                                                               BOND_SLAVE_NOTIFY_NOW);
2131
2132                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2133                                     slave->dev->name);
2134
2135                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2136                                 bond_3ad_handle_link_change(slave,
2137                                                             BOND_LINK_DOWN);
2138
2139                         if (bond_is_lb(bond))
2140                                 bond_alb_handle_link_change(bond, slave,
2141                                                             BOND_LINK_DOWN);
2142
2143                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2144                                 bond_update_slave_arr(bond, NULL);
2145
2146                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2147                                 goto do_failover;
2148
2149                         continue;
2150
2151                 default:
2152                         netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2153                                    slave->new_link, slave->dev->name);
2154                         slave->new_link = BOND_LINK_NOCHANGE;
2155
2156                         continue;
2157                 }
2158
2159 do_failover:
2160                 block_netpoll_tx();
2161                 bond_select_active_slave(bond);
2162                 unblock_netpoll_tx();
2163         }
2164
2165         bond_set_carrier(bond);
2166 }
2167
2168 /* bond_mii_monitor
2169  *
2170  * Really a wrapper that splits the mii monitor into two phases: an
2171  * inspection, then (if inspection indicates something needs to be done)
2172  * an acquisition of appropriate locks followed by a commit phase to
2173  * implement whatever link state changes are indicated.
2174  */
2175 static void bond_mii_monitor(struct work_struct *work)
2176 {
2177         struct bonding *bond = container_of(work, struct bonding,
2178                                             mii_work.work);
2179         bool should_notify_peers = false;
2180         unsigned long delay;
2181
2182         delay = msecs_to_jiffies(bond->params.miimon);
2183
2184         if (!bond_has_slaves(bond))
2185                 goto re_arm;
2186
2187         rcu_read_lock();
2188
2189         should_notify_peers = bond_should_notify_peers(bond);
2190
2191         if (bond_miimon_inspect(bond)) {
2192                 rcu_read_unlock();
2193
2194                 /* Race avoidance with bond_close cancel of workqueue */
2195                 if (!rtnl_trylock()) {
2196                         delay = 1;
2197                         should_notify_peers = false;
2198                         goto re_arm;
2199                 }
2200
2201                 bond_miimon_commit(bond);
2202
2203                 rtnl_unlock();  /* might sleep, hold no other locks */
2204         } else
2205                 rcu_read_unlock();
2206
2207 re_arm:
2208         if (bond->params.miimon)
2209                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2210
2211         if (should_notify_peers) {
2212                 if (!rtnl_trylock())
2213                         return;
2214                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2215                 rtnl_unlock();
2216         }
2217 }
2218
2219 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2220 {
2221         struct net_device *upper;
2222         struct list_head *iter;
2223         bool ret = false;
2224
2225         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2226                 return true;
2227
2228         rcu_read_lock();
2229         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2230                 if (ip == bond_confirm_addr(upper, 0, ip)) {
2231                         ret = true;
2232                         break;
2233                 }
2234         }
2235         rcu_read_unlock();
2236
2237         return ret;
2238 }
2239
2240 /* We go to the (large) trouble of VLAN tagging ARP frames because
2241  * switches in VLAN mode (especially if ports are configured as
2242  * "native" to a VLAN) might not pass non-tagged frames.
2243  */
2244 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2245                           __be32 dest_ip, __be32 src_ip,
2246                           struct bond_vlan_tag *tags)
2247 {
2248         struct sk_buff *skb;
2249         struct bond_vlan_tag *outer_tag = tags;
2250
2251         netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2252                    arp_op, slave_dev->name, &dest_ip, &src_ip);
2253
2254         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2255                          NULL, slave_dev->dev_addr, NULL);
2256
2257         if (!skb) {
2258                 net_err_ratelimited("ARP packet allocation failed\n");
2259                 return;
2260         }
2261
2262         if (!tags || tags->vlan_proto == VLAN_N_VID)
2263                 goto xmit;
2264
2265         tags++;
2266
2267         /* Go through all the tags backwards and add them to the packet */
2268         while (tags->vlan_proto != VLAN_N_VID) {
2269                 if (!tags->vlan_id) {
2270                         tags++;
2271                         continue;
2272                 }
2273
2274                 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2275                            ntohs(outer_tag->vlan_proto), tags->vlan_id);
2276                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2277                                                 tags->vlan_id);
2278                 if (!skb) {
2279                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2280                         return;
2281                 }
2282
2283                 tags++;
2284         }
2285         /* Set the outer tag */
2286         if (outer_tag->vlan_id) {
2287                 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2288                            ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2289                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2290                                        outer_tag->vlan_id);
2291         }
2292
2293 xmit:
2294         arp_xmit(skb);
2295 }
2296
2297 /* Validate the device path between the @start_dev and the @end_dev.
2298  * The path is valid if the @end_dev is reachable through device
2299  * stacking.
2300  * When the path is validated, collect any vlan information in the
2301  * path.
2302  */
2303 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2304                                               struct net_device *end_dev,
2305                                               int level)
2306 {
2307         struct bond_vlan_tag *tags;
2308         struct net_device *upper;
2309         struct list_head  *iter;
2310
2311         if (start_dev == end_dev) {
2312                 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2313                 if (!tags)
2314                         return ERR_PTR(-ENOMEM);
2315                 tags[level].vlan_proto = VLAN_N_VID;
2316                 return tags;
2317         }
2318
2319         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2320                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2321                 if (IS_ERR_OR_NULL(tags)) {
2322                         if (IS_ERR(tags))
2323                                 return tags;
2324                         continue;
2325                 }
2326                 if (is_vlan_dev(upper)) {
2327                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2328                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2329                 }
2330
2331                 return tags;
2332         }
2333
2334         return NULL;
2335 }
2336
2337 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2338 {
2339         struct rtable *rt;
2340         struct bond_vlan_tag *tags;
2341         __be32 *targets = bond->params.arp_targets, addr;
2342         int i;
2343
2344         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2345                 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2346                 tags = NULL;
2347
2348                 /* Find out through which dev should the packet go */
2349                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2350                                      RTO_ONLINK, 0);
2351                 if (IS_ERR(rt)) {
2352                         /* there's no route to target - try to send arp
2353                          * probe to generate any traffic (arp_validate=0)
2354                          */
2355                         if (bond->params.arp_validate)
2356                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2357                                                      bond->dev->name,
2358                                                      &targets[i]);
2359                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2360                                       0, tags);
2361                         continue;
2362                 }
2363
2364                 /* bond device itself */
2365                 if (rt->dst.dev == bond->dev)
2366                         goto found;
2367
2368                 rcu_read_lock();
2369                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2370                 rcu_read_unlock();
2371
2372                 if (!IS_ERR_OR_NULL(tags))
2373                         goto found;
2374
2375                 /* Not our device - skip */
2376                 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2377                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2378
2379                 ip_rt_put(rt);
2380                 continue;
2381
2382 found:
2383                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2384                 ip_rt_put(rt);
2385                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2386                               addr, tags);
2387                 kfree(tags);
2388         }
2389 }
2390
2391 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2392 {
2393         int i;
2394
2395         if (!sip || !bond_has_this_ip(bond, tip)) {
2396                 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2397                            &sip, &tip);
2398                 return;
2399         }
2400
2401         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2402         if (i == -1) {
2403                 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2404                            &sip);
2405                 return;
2406         }
2407         slave->last_rx = jiffies;
2408         slave->target_last_arp_rx[i] = jiffies;
2409 }
2410
2411 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2412                  struct slave *slave)
2413 {
2414         struct arphdr *arp = (struct arphdr *)skb->data;
2415         struct slave *curr_active_slave, *curr_arp_slave;
2416         unsigned char *arp_ptr;
2417         __be32 sip, tip;
2418         int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2419
2420         if (!slave_do_arp_validate(bond, slave)) {
2421                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2422                     !slave_do_arp_validate_only(bond))
2423                         slave->last_rx = jiffies;
2424                 return RX_HANDLER_ANOTHER;
2425         } else if (!is_arp) {
2426                 return RX_HANDLER_ANOTHER;
2427         }
2428
2429         alen = arp_hdr_len(bond->dev);
2430
2431         netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2432                    skb->dev->name);
2433
2434         if (alen > skb_headlen(skb)) {
2435                 arp = kmalloc(alen, GFP_ATOMIC);
2436                 if (!arp)
2437                         goto out_unlock;
2438                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2439                         goto out_unlock;
2440         }
2441
2442         if (arp->ar_hln != bond->dev->addr_len ||
2443             skb->pkt_type == PACKET_OTHERHOST ||
2444             skb->pkt_type == PACKET_LOOPBACK ||
2445             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2446             arp->ar_pro != htons(ETH_P_IP) ||
2447             arp->ar_pln != 4)
2448                 goto out_unlock;
2449
2450         arp_ptr = (unsigned char *)(arp + 1);
2451         arp_ptr += bond->dev->addr_len;
2452         memcpy(&sip, arp_ptr, 4);
2453         arp_ptr += 4 + bond->dev->addr_len;
2454         memcpy(&tip, arp_ptr, 4);
2455
2456         netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2457                    slave->dev->name, bond_slave_state(slave),
2458                      bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2459                      &sip, &tip);
2460
2461         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2462         curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2463
2464         /* We 'trust' the received ARP enough to validate it if:
2465          *
2466          * (a) the slave receiving the ARP is active (which includes the
2467          * current ARP slave, if any), or
2468          *
2469          * (b) the receiving slave isn't active, but there is a currently
2470          * active slave and it received valid arp reply(s) after it became
2471          * the currently active slave, or
2472          *
2473          * (c) there is an ARP slave that sent an ARP during the prior ARP
2474          * interval, and we receive an ARP reply on any slave.  We accept
2475          * these because switch FDB update delays may deliver the ARP
2476          * reply to a slave other than the sender of the ARP request.
2477          *
2478          * Note: for (b), backup slaves are receiving the broadcast ARP
2479          * request, not a reply.  This request passes from the sending
2480          * slave through the L2 switch(es) to the receiving slave.  Since
2481          * this is checking the request, sip/tip are swapped for
2482          * validation.
2483          *
2484          * This is done to avoid endless looping when we can't reach the
2485          * arp_ip_target and fool ourselves with our own arp requests.
2486          */
2487         if (bond_is_active_slave(slave))
2488                 bond_validate_arp(bond, slave, sip, tip);
2489         else if (curr_active_slave &&
2490                  time_after(slave_last_rx(bond, curr_active_slave),
2491                             curr_active_slave->last_link_up))
2492                 bond_validate_arp(bond, slave, tip, sip);
2493         else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2494                  bond_time_in_interval(bond,
2495                                        dev_trans_start(curr_arp_slave->dev), 1))
2496                 bond_validate_arp(bond, slave, sip, tip);
2497
2498 out_unlock:
2499         if (arp != (struct arphdr *)skb->data)
2500                 kfree(arp);
2501         return RX_HANDLER_ANOTHER;
2502 }
2503
2504 /* function to verify if we're in the arp_interval timeslice, returns true if
2505  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2506  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2507  */
2508 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2509                                   int mod)
2510 {
2511         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2512
2513         return time_in_range(jiffies,
2514                              last_act - delta_in_ticks,
2515                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2516 }
2517
2518 /* This function is called regularly to monitor each slave's link
2519  * ensuring that traffic is being sent and received when arp monitoring
2520  * is used in load-balancing mode. if the adapter has been dormant, then an
2521  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2522  * arp monitoring in active backup mode.
2523  */
2524 static void bond_loadbalance_arp_mon(struct work_struct *work)
2525 {
2526         struct bonding *bond = container_of(work, struct bonding,
2527                                             arp_work.work);
2528         struct slave *slave, *oldcurrent;
2529         struct list_head *iter;
2530         int do_failover = 0, slave_state_changed = 0;
2531
2532         if (!bond_has_slaves(bond))
2533                 goto re_arm;
2534
2535         rcu_read_lock();
2536
2537         oldcurrent = rcu_dereference(bond->curr_active_slave);
2538         /* see if any of the previous devices are up now (i.e. they have
2539          * xmt and rcv traffic). the curr_active_slave does not come into
2540          * the picture unless it is null. also, slave->last_link_up is not
2541          * needed here because we send an arp on each slave and give a slave
2542          * as long as it needs to get the tx/rx within the delta.
2543          * TODO: what about up/down delay in arp mode? it wasn't here before
2544          *       so it can wait
2545          */
2546         bond_for_each_slave_rcu(bond, slave, iter) {
2547                 unsigned long trans_start = dev_trans_start(slave->dev);
2548
2549                 slave->new_link = BOND_LINK_NOCHANGE;
2550
2551                 if (slave->link != BOND_LINK_UP) {
2552                         if (bond_time_in_interval(bond, trans_start, 1) &&
2553                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2554
2555                                 slave->new_link = BOND_LINK_UP;
2556                                 slave_state_changed = 1;
2557
2558                                 /* primary_slave has no meaning in round-robin
2559                                  * mode. the window of a slave being up and
2560                                  * curr_active_slave being null after enslaving
2561                                  * is closed.
2562                                  */
2563                                 if (!oldcurrent) {
2564                                         netdev_info(bond->dev, "link status definitely up for interface %s\n",
2565                                                     slave->dev->name);
2566                                         do_failover = 1;
2567                                 } else {
2568                                         netdev_info(bond->dev, "interface %s is now up\n",
2569                                                     slave->dev->name);
2570                                 }
2571                         }
2572                 } else {
2573                         /* slave->link == BOND_LINK_UP */
2574
2575                         /* not all switches will respond to an arp request
2576                          * when the source ip is 0, so don't take the link down
2577                          * if we don't know our ip yet
2578                          */
2579                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2580                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2581
2582                                 slave->new_link = BOND_LINK_DOWN;
2583                                 slave_state_changed = 1;
2584
2585                                 if (slave->link_failure_count < UINT_MAX)
2586                                         slave->link_failure_count++;
2587
2588                                 netdev_info(bond->dev, "interface %s is now down\n",
2589                                             slave->dev->name);
2590
2591                                 if (slave == oldcurrent)
2592                                         do_failover = 1;
2593                         }
2594                 }
2595
2596                 /* note: if switch is in round-robin mode, all links
2597                  * must tx arp to ensure all links rx an arp - otherwise
2598                  * links may oscillate or not come up at all; if switch is
2599                  * in something like xor mode, there is nothing we can
2600                  * do - all replies will be rx'ed on same link causing slaves
2601                  * to be unstable during low/no traffic periods
2602                  */
2603                 if (bond_slave_is_up(slave))
2604                         bond_arp_send_all(bond, slave);
2605         }
2606
2607         rcu_read_unlock();
2608
2609         if (do_failover || slave_state_changed) {
2610                 if (!rtnl_trylock())
2611                         goto re_arm;
2612
2613                 bond_for_each_slave(bond, slave, iter) {
2614                         if (slave->new_link != BOND_LINK_NOCHANGE)
2615                                 slave->link = slave->new_link;
2616                 }
2617
2618                 if (slave_state_changed) {
2619                         bond_slave_state_change(bond);
2620                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2621                                 bond_update_slave_arr(bond, NULL);
2622                 }
2623                 if (do_failover) {
2624                         block_netpoll_tx();
2625                         bond_select_active_slave(bond);
2626                         unblock_netpoll_tx();
2627                 }
2628                 rtnl_unlock();
2629         }
2630
2631 re_arm:
2632         if (bond->params.arp_interval)
2633                 queue_delayed_work(bond->wq, &bond->arp_work,
2634                                    msecs_to_jiffies(bond->params.arp_interval));
2635 }
2636
2637 /* Called to inspect slaves for active-backup mode ARP monitor link state
2638  * changes.  Sets new_link in slaves to specify what action should take
2639  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2640  * to link states must be committed.
2641  *
2642  * Called with rcu_read_lock held.
2643  */
2644 static int bond_ab_arp_inspect(struct bonding *bond)
2645 {
2646         unsigned long trans_start, last_rx;
2647         struct list_head *iter;
2648         struct slave *slave;
2649         int commit = 0;
2650
2651         bond_for_each_slave_rcu(bond, slave, iter) {
2652                 slave->new_link = BOND_LINK_NOCHANGE;
2653                 last_rx = slave_last_rx(bond, slave);
2654
2655                 if (slave->link != BOND_LINK_UP) {
2656                         if (bond_time_in_interval(bond, last_rx, 1)) {
2657                                 slave->new_link = BOND_LINK_UP;
2658                                 commit++;
2659                         }
2660                         continue;
2661                 }
2662
2663                 /* Give slaves 2*delta after being enslaved or made
2664                  * active.  This avoids bouncing, as the last receive
2665                  * times need a full ARP monitor cycle to be updated.
2666                  */
2667                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2668                         continue;
2669
2670                 /* Backup slave is down if:
2671                  * - No current_arp_slave AND
2672                  * - more than 3*delta since last receive AND
2673                  * - the bond has an IP address
2674                  *
2675                  * Note: a non-null current_arp_slave indicates
2676                  * the curr_active_slave went down and we are
2677                  * searching for a new one; under this condition
2678                  * we only take the curr_active_slave down - this
2679                  * gives each slave a chance to tx/rx traffic
2680                  * before being taken out
2681                  */
2682                 if (!bond_is_active_slave(slave) &&
2683                     !rcu_access_pointer(bond->current_arp_slave) &&
2684                     !bond_time_in_interval(bond, last_rx, 3)) {
2685                         slave->new_link = BOND_LINK_DOWN;
2686                         commit++;
2687                 }
2688
2689                 /* Active slave is down if:
2690                  * - more than 2*delta since transmitting OR
2691                  * - (more than 2*delta since receive AND
2692                  *    the bond has an IP address)
2693                  */
2694                 trans_start = dev_trans_start(slave->dev);
2695                 if (bond_is_active_slave(slave) &&
2696                     (!bond_time_in_interval(bond, trans_start, 2) ||
2697                      !bond_time_in_interval(bond, last_rx, 2))) {
2698                         slave->new_link = BOND_LINK_DOWN;
2699                         commit++;
2700                 }
2701         }
2702
2703         return commit;
2704 }
2705
2706 /* Called to commit link state changes noted by inspection step of
2707  * active-backup mode ARP monitor.
2708  *
2709  * Called with RTNL hold.
2710  */
2711 static void bond_ab_arp_commit(struct bonding *bond)
2712 {
2713         unsigned long trans_start;
2714         struct list_head *iter;
2715         struct slave *slave;
2716
2717         bond_for_each_slave(bond, slave, iter) {
2718                 switch (slave->new_link) {
2719                 case BOND_LINK_NOCHANGE:
2720                         continue;
2721
2722                 case BOND_LINK_UP:
2723                         trans_start = dev_trans_start(slave->dev);
2724                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
2725                             (!rtnl_dereference(bond->curr_active_slave) &&
2726                              bond_time_in_interval(bond, trans_start, 1))) {
2727                                 struct slave *current_arp_slave;
2728
2729                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2730                                 bond_set_slave_link_state(slave, BOND_LINK_UP);
2731                                 if (current_arp_slave) {
2732                                         bond_set_slave_inactive_flags(
2733                                                 current_arp_slave,
2734                                                 BOND_SLAVE_NOTIFY_NOW);
2735                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2736                                 }
2737
2738                                 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2739                                             slave->dev->name);
2740
2741                                 if (!rtnl_dereference(bond->curr_active_slave) ||
2742                                     slave == rtnl_dereference(bond->primary_slave))
2743                                         goto do_failover;
2744
2745                         }
2746
2747                         continue;
2748
2749                 case BOND_LINK_DOWN:
2750                         if (slave->link_failure_count < UINT_MAX)
2751                                 slave->link_failure_count++;
2752
2753                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2754                         bond_set_slave_inactive_flags(slave,
2755                                                       BOND_SLAVE_NOTIFY_NOW);
2756
2757                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2758                                     slave->dev->name);
2759
2760                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
2761                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2762                                 goto do_failover;
2763                         }
2764
2765                         continue;
2766
2767                 default:
2768                         netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2769                                    slave->new_link, slave->dev->name);
2770                         continue;
2771                 }
2772
2773 do_failover:
2774                 block_netpoll_tx();
2775                 bond_select_active_slave(bond);
2776                 unblock_netpoll_tx();
2777         }
2778
2779         bond_set_carrier(bond);
2780 }
2781
2782 /* Send ARP probes for active-backup mode ARP monitor.
2783  *
2784  * Called with rcu_read_lock held.
2785  */
2786 static bool bond_ab_arp_probe(struct bonding *bond)
2787 {
2788         struct slave *slave, *before = NULL, *new_slave = NULL,
2789                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2790                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2791         struct list_head *iter;
2792         bool found = false;
2793         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2794
2795         if (curr_arp_slave && curr_active_slave)
2796                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2797                             curr_arp_slave->dev->name,
2798                             curr_active_slave->dev->name);
2799
2800         if (curr_active_slave) {
2801                 bond_arp_send_all(bond, curr_active_slave);
2802                 return should_notify_rtnl;
2803         }
2804
2805         /* if we don't have a curr_active_slave, search for the next available
2806          * backup slave from the current_arp_slave and make it the candidate
2807          * for becoming the curr_active_slave
2808          */
2809
2810         if (!curr_arp_slave) {
2811                 curr_arp_slave = bond_first_slave_rcu(bond);
2812                 if (!curr_arp_slave)
2813                         return should_notify_rtnl;
2814         }
2815
2816         bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2817
2818         bond_for_each_slave_rcu(bond, slave, iter) {
2819                 if (!found && !before && bond_slave_is_up(slave))
2820                         before = slave;
2821
2822                 if (found && !new_slave && bond_slave_is_up(slave))
2823                         new_slave = slave;
2824                 /* if the link state is up at this point, we
2825                  * mark it down - this can happen if we have
2826                  * simultaneous link failures and
2827                  * reselect_active_interface doesn't make this
2828                  * one the current slave so it is still marked
2829                  * up when it is actually down
2830                  */
2831                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2832                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2833                         if (slave->link_failure_count < UINT_MAX)
2834                                 slave->link_failure_count++;
2835
2836                         bond_set_slave_inactive_flags(slave,
2837                                                       BOND_SLAVE_NOTIFY_LATER);
2838
2839                         netdev_info(bond->dev, "backup interface %s is now down\n",
2840                                     slave->dev->name);
2841                 }
2842                 if (slave == curr_arp_slave)
2843                         found = true;
2844         }
2845
2846         if (!new_slave && before)
2847                 new_slave = before;
2848
2849         if (!new_slave)
2850                 goto check_state;
2851
2852         bond_set_slave_link_state(new_slave, BOND_LINK_BACK);
2853         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2854         bond_arp_send_all(bond, new_slave);
2855         new_slave->last_link_up = jiffies;
2856         rcu_assign_pointer(bond->current_arp_slave, new_slave);
2857
2858 check_state:
2859         bond_for_each_slave_rcu(bond, slave, iter) {
2860                 if (slave->should_notify) {
2861                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2862                         break;
2863                 }
2864         }
2865         return should_notify_rtnl;
2866 }
2867
2868 static void bond_activebackup_arp_mon(struct work_struct *work)
2869 {
2870         struct bonding *bond = container_of(work, struct bonding,
2871                                             arp_work.work);
2872         bool should_notify_peers = false;
2873         bool should_notify_rtnl = false;
2874         int delta_in_ticks;
2875
2876         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2877
2878         if (!bond_has_slaves(bond))
2879                 goto re_arm;
2880
2881         rcu_read_lock();
2882
2883         should_notify_peers = bond_should_notify_peers(bond);
2884
2885         if (bond_ab_arp_inspect(bond)) {
2886                 rcu_read_unlock();
2887
2888                 /* Race avoidance with bond_close flush of workqueue */
2889                 if (!rtnl_trylock()) {
2890                         delta_in_ticks = 1;
2891                         should_notify_peers = false;
2892                         goto re_arm;
2893                 }
2894
2895                 bond_ab_arp_commit(bond);
2896
2897                 rtnl_unlock();
2898                 rcu_read_lock();
2899         }
2900
2901         should_notify_rtnl = bond_ab_arp_probe(bond);
2902         rcu_read_unlock();
2903
2904 re_arm:
2905         if (bond->params.arp_interval)
2906                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2907
2908         if (should_notify_peers || should_notify_rtnl) {
2909                 if (!rtnl_trylock())
2910                         return;
2911
2912                 if (should_notify_peers)
2913                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2914                                                  bond->dev);
2915                 if (should_notify_rtnl)
2916                         bond_slave_state_notify(bond);
2917
2918                 rtnl_unlock();
2919         }
2920 }
2921
2922 /*-------------------------- netdev event handling --------------------------*/
2923
2924 /* Change device name */
2925 static int bond_event_changename(struct bonding *bond)
2926 {
2927         bond_remove_proc_entry(bond);
2928         bond_create_proc_entry(bond);
2929
2930         bond_debug_reregister(bond);
2931
2932         return NOTIFY_DONE;
2933 }
2934
2935 static int bond_master_netdev_event(unsigned long event,
2936                                     struct net_device *bond_dev)
2937 {
2938         struct bonding *event_bond = netdev_priv(bond_dev);
2939
2940         switch (event) {
2941         case NETDEV_CHANGENAME:
2942                 return bond_event_changename(event_bond);
2943         case NETDEV_UNREGISTER:
2944                 bond_remove_proc_entry(event_bond);
2945                 break;
2946         case NETDEV_REGISTER:
2947                 bond_create_proc_entry(event_bond);
2948                 break;
2949         case NETDEV_NOTIFY_PEERS:
2950                 if (event_bond->send_peer_notif)
2951                         event_bond->send_peer_notif--;
2952                 break;
2953         default:
2954                 break;
2955         }
2956
2957         return NOTIFY_DONE;
2958 }
2959
2960 static int bond_slave_netdev_event(unsigned long event,
2961                                    struct net_device *slave_dev)
2962 {
2963         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
2964         struct bonding *bond;
2965         struct net_device *bond_dev;
2966
2967         /* A netdev event can be generated while enslaving a device
2968          * before netdev_rx_handler_register is called in which case
2969          * slave will be NULL
2970          */
2971         if (!slave)
2972                 return NOTIFY_DONE;
2973         bond_dev = slave->bond->dev;
2974         bond = slave->bond;
2975         primary = rtnl_dereference(bond->primary_slave);
2976
2977         switch (event) {
2978         case NETDEV_UNREGISTER:
2979                 if (bond_dev->type != ARPHRD_ETHER)
2980                         bond_release_and_destroy(bond_dev, slave_dev);
2981                 else
2982                         bond_release(bond_dev, slave_dev);
2983                 break;
2984         case NETDEV_UP:
2985         case NETDEV_CHANGE:
2986                 bond_update_speed_duplex(slave);
2987                 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2988                         bond_3ad_adapter_speed_duplex_changed(slave);
2989                 /* Fallthrough */
2990         case NETDEV_DOWN:
2991                 /* Refresh slave-array if applicable!
2992                  * If the setup does not use miimon or arpmon (mode-specific!),
2993                  * then these events will not cause the slave-array to be
2994                  * refreshed. This will cause xmit to use a slave that is not
2995                  * usable. Avoid such situation by refeshing the array at these
2996                  * events. If these (miimon/arpmon) parameters are configured
2997                  * then array gets refreshed twice and that should be fine!
2998                  */
2999                 if (bond_mode_uses_xmit_hash(bond))
3000                         bond_update_slave_arr(bond, NULL);
3001                 break;
3002         case NETDEV_CHANGEMTU:
3003                 /* TODO: Should slaves be allowed to
3004                  * independently alter their MTU?  For
3005                  * an active-backup bond, slaves need
3006                  * not be the same type of device, so
3007                  * MTUs may vary.  For other modes,
3008                  * slaves arguably should have the
3009                  * same MTUs. To do this, we'd need to
3010                  * take over the slave's change_mtu
3011                  * function for the duration of their
3012                  * servitude.
3013                  */
3014                 break;
3015         case NETDEV_CHANGENAME:
3016                 /* we don't care if we don't have primary set */
3017                 if (!bond_uses_primary(bond) ||
3018                     !bond->params.primary[0])
3019                         break;
3020
3021                 if (slave == primary) {
3022                         /* slave's name changed - he's no longer primary */
3023                         RCU_INIT_POINTER(bond->primary_slave, NULL);
3024                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3025                         /* we have a new primary slave */
3026                         rcu_assign_pointer(bond->primary_slave, slave);
3027                 } else { /* we didn't change primary - exit */
3028                         break;
3029                 }
3030
3031                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3032                             primary ? slave_dev->name : "none");
3033
3034                 block_netpoll_tx();
3035                 bond_select_active_slave(bond);
3036                 unblock_netpoll_tx();
3037                 break;
3038         case NETDEV_FEAT_CHANGE:
3039                 bond_compute_features(bond);
3040                 break;
3041         case NETDEV_RESEND_IGMP:
3042                 /* Propagate to master device */
3043                 call_netdevice_notifiers(event, slave->bond->dev);
3044                 break;
3045         default:
3046                 break;
3047         }
3048
3049         return NOTIFY_DONE;
3050 }
3051
3052 /* bond_netdev_event: handle netdev notifier chain events.
3053  *
3054  * This function receives events for the netdev chain.  The caller (an
3055  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3056  * locks for us to safely manipulate the slave devices (RTNL lock,
3057  * dev_probe_lock).
3058  */
3059 static int bond_netdev_event(struct notifier_block *this,
3060                              unsigned long event, void *ptr)
3061 {
3062         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3063
3064         netdev_dbg(event_dev, "event: %lx\n", event);
3065
3066         if (!(event_dev->priv_flags & IFF_BONDING))
3067                 return NOTIFY_DONE;
3068
3069         if (event_dev->flags & IFF_MASTER) {
3070                 netdev_dbg(event_dev, "IFF_MASTER\n");
3071                 return bond_master_netdev_event(event, event_dev);
3072         }
3073
3074         if (event_dev->flags & IFF_SLAVE) {
3075                 netdev_dbg(event_dev, "IFF_SLAVE\n");
3076                 return bond_slave_netdev_event(event, event_dev);
3077         }
3078
3079         return NOTIFY_DONE;
3080 }
3081
3082 static struct notifier_block bond_netdev_notifier = {
3083         .notifier_call = bond_netdev_event,
3084 };
3085
3086 /*---------------------------- Hashing Policies -----------------------------*/
3087
3088 /* L2 hash helper */
3089 static inline u32 bond_eth_hash(struct sk_buff *skb)
3090 {
3091         struct ethhdr *ep, hdr_tmp;
3092
3093         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3094         if (ep)
3095                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3096         return 0;
3097 }
3098
3099 /* Extract the appropriate headers based on bond's xmit policy */
3100 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3101                               struct flow_keys *fk)
3102 {
3103         const struct ipv6hdr *iph6;
3104         const struct iphdr *iph;
3105         int noff, proto = -1;
3106
3107         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3108                 return skb_flow_dissect_flow_keys(skb, fk, 0);
3109
3110         fk->ports.ports = 0;
3111         noff = skb_network_offset(skb);
3112         if (skb->protocol == htons(ETH_P_IP)) {
3113                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3114                         return false;
3115                 iph = ip_hdr(skb);
3116                 iph_to_flow_copy_v4addrs(fk, iph);
3117                 noff += iph->ihl << 2;
3118                 if (!ip_is_fragment(iph))
3119                         proto = iph->protocol;
3120         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3121                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3122                         return false;
3123                 iph6 = ipv6_hdr(skb);
3124                 iph_to_flow_copy_v6addrs(fk, iph6);
3125                 noff += sizeof(*iph6);
3126                 proto = iph6->nexthdr;
3127         } else {
3128                 return false;
3129         }
3130         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3131                 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3132
3133         return true;
3134 }
3135
3136 /**
3137  * bond_xmit_hash - generate a hash value based on the xmit policy
3138  * @bond: bonding device
3139  * @skb: buffer to use for headers
3140  *
3141  * This function will extract the necessary headers from the skb buffer and use
3142  * them to generate a hash based on the xmit_policy set in the bonding device
3143  */
3144 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3145 {
3146         struct flow_keys flow;
3147         u32 hash;
3148
3149         if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3150             skb->l4_hash)
3151                 return skb->hash;
3152
3153         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3154             !bond_flow_dissect(bond, skb, &flow))
3155                 return bond_eth_hash(skb);
3156
3157         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3158             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3159                 hash = bond_eth_hash(skb);
3160         else
3161                 hash = (__force u32)flow.ports.ports;
3162         hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3163                 (__force u32)flow_get_u32_src(&flow);
3164         hash ^= (hash >> 16);
3165         hash ^= (hash >> 8);
3166
3167         return hash >> 1;
3168 }
3169
3170 /*-------------------------- Device entry points ----------------------------*/
3171
3172 static void bond_work_init_all(struct bonding *bond)
3173 {
3174         INIT_DELAYED_WORK(&bond->mcast_work,
3175                           bond_resend_igmp_join_requests_delayed);
3176         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3177         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3178         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3179                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3180         else
3181                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3182         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3183         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3184 }
3185
3186 static void bond_work_cancel_all(struct bonding *bond)
3187 {
3188         cancel_delayed_work_sync(&bond->mii_work);
3189         cancel_delayed_work_sync(&bond->arp_work);
3190         cancel_delayed_work_sync(&bond->alb_work);
3191         cancel_delayed_work_sync(&bond->ad_work);
3192         cancel_delayed_work_sync(&bond->mcast_work);
3193         cancel_delayed_work_sync(&bond->slave_arr_work);
3194 }
3195
3196 static int bond_open(struct net_device *bond_dev)
3197 {
3198         struct bonding *bond = netdev_priv(bond_dev);
3199         struct list_head *iter;
3200         struct slave *slave;
3201
3202         /* reset slave->backup and slave->inactive */
3203         if (bond_has_slaves(bond)) {
3204                 bond_for_each_slave(bond, slave, iter) {
3205                         if (bond_uses_primary(bond) &&
3206                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3207                                 bond_set_slave_inactive_flags(slave,
3208                                                               BOND_SLAVE_NOTIFY_NOW);
3209                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3210                                 bond_set_slave_active_flags(slave,
3211                                                             BOND_SLAVE_NOTIFY_NOW);
3212                         }
3213                 }
3214         }
3215
3216         bond_work_init_all(bond);
3217
3218         if (bond_is_lb(bond)) {
3219                 /* bond_alb_initialize must be called before the timer
3220                  * is started.
3221                  */
3222                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3223                         return -ENOMEM;
3224                 if (bond->params.tlb_dynamic_lb)
3225                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3226         }
3227
3228         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3229                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3230
3231         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3232                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3233                 bond->recv_probe = bond_arp_rcv;
3234         }
3235
3236         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3237                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3238                 /* register to receive LACPDUs */
3239                 bond->recv_probe = bond_3ad_lacpdu_recv;
3240                 bond_3ad_initiate_agg_selection(bond, 1);
3241         }
3242
3243         if (bond_mode_uses_xmit_hash(bond))
3244                 bond_update_slave_arr(bond, NULL);
3245
3246         return 0;
3247 }
3248
3249 static int bond_close(struct net_device *bond_dev)
3250 {
3251         struct bonding *bond = netdev_priv(bond_dev);
3252
3253         bond_work_cancel_all(bond);
3254         bond->send_peer_notif = 0;
3255         if (bond_is_lb(bond))
3256                 bond_alb_deinitialize(bond);
3257         bond->recv_probe = NULL;
3258
3259         return 0;
3260 }
3261
3262 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3263  * that some drivers can provide 32bit values only.
3264  */
3265 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3266                             const struct rtnl_link_stats64 *_new,
3267                             const struct rtnl_link_stats64 *_old)
3268 {
3269         const u64 *new = (const u64 *)_new;
3270         const u64 *old = (const u64 *)_old;
3271         u64 *res = (u64 *)_res;
3272         int i;
3273
3274         for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3275                 u64 nv = new[i];
3276                 u64 ov = old[i];
3277                 s64 delta = nv - ov;
3278
3279                 /* detects if this particular field is 32bit only */
3280                 if (((nv | ov) >> 32) == 0)
3281                         delta = (s64)(s32)((u32)nv - (u32)ov);
3282
3283                 /* filter anomalies, some drivers reset their stats
3284                  * at down/up events.
3285                  */
3286                 if (delta > 0)
3287                         res[i] += delta;
3288         }
3289 }
3290
3291 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3292                                                 struct rtnl_link_stats64 *stats)
3293 {
3294         struct bonding *bond = netdev_priv(bond_dev);
3295         struct rtnl_link_stats64 temp;
3296         struct list_head *iter;
3297         struct slave *slave;
3298
3299         spin_lock(&bond->stats_lock);
3300         memcpy(stats, &bond->bond_stats, sizeof(*stats));
3301
3302         rcu_read_lock();
3303         bond_for_each_slave_rcu(bond, slave, iter) {
3304                 const struct rtnl_link_stats64 *new =
3305                         dev_get_stats(slave->dev, &temp);
3306
3307                 bond_fold_stats(stats, new, &slave->slave_stats);
3308
3309                 /* save off the slave stats for the next run */
3310                 memcpy(&slave->slave_stats, new, sizeof(*new));
3311         }
3312         rcu_read_unlock();
3313
3314         memcpy(&bond->bond_stats, stats, sizeof(*stats));
3315         spin_unlock(&bond->stats_lock);
3316
3317         return stats;
3318 }
3319
3320 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3321 {
3322         struct bonding *bond = netdev_priv(bond_dev);
3323         struct net_device *slave_dev = NULL;
3324         struct ifbond k_binfo;
3325         struct ifbond __user *u_binfo = NULL;
3326         struct ifslave k_sinfo;
3327         struct ifslave __user *u_sinfo = NULL;
3328         struct mii_ioctl_data *mii = NULL;
3329         struct bond_opt_value newval;
3330         struct net *net;
3331         int res = 0;
3332
3333         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3334
3335         switch (cmd) {
3336         case SIOCGMIIPHY:
3337                 mii = if_mii(ifr);
3338                 if (!mii)
3339                         return -EINVAL;
3340
3341                 mii->phy_id = 0;
3342                 /* Fall Through */
3343         case SIOCGMIIREG:
3344                 /* We do this again just in case we were called by SIOCGMIIREG
3345                  * instead of SIOCGMIIPHY.
3346                  */
3347                 mii = if_mii(ifr);
3348                 if (!mii)
3349                         return -EINVAL;
3350
3351                 if (mii->reg_num == 1) {
3352                         mii->val_out = 0;
3353                         if (netif_carrier_ok(bond->dev))
3354                                 mii->val_out = BMSR_LSTATUS;
3355                 }
3356
3357                 return 0;
3358         case BOND_INFO_QUERY_OLD:
3359         case SIOCBONDINFOQUERY:
3360                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3361
3362                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3363                         return -EFAULT;
3364
3365                 res = bond_info_query(bond_dev, &k_binfo);
3366                 if (res == 0 &&
3367                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3368                         return -EFAULT;
3369
3370                 return res;
3371         case BOND_SLAVE_INFO_QUERY_OLD:
3372         case SIOCBONDSLAVEINFOQUERY:
3373                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3374
3375                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3376                         return -EFAULT;
3377
3378                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3379                 if (res == 0 &&
3380                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3381                         return -EFAULT;
3382
3383                 return res;
3384         default:
3385                 break;
3386         }
3387
3388         net = dev_net(bond_dev);
3389
3390         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3391                 return -EPERM;
3392
3393         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3394
3395         netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3396
3397         if (!slave_dev)
3398                 return -ENODEV;
3399
3400         netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3401         switch (cmd) {
3402         case BOND_ENSLAVE_OLD:
3403         case SIOCBONDENSLAVE:
3404                 res = bond_enslave(bond_dev, slave_dev);
3405                 break;
3406         case BOND_RELEASE_OLD:
3407         case SIOCBONDRELEASE:
3408                 res = bond_release(bond_dev, slave_dev);
3409                 break;
3410         case BOND_SETHWADDR_OLD:
3411         case SIOCBONDSETHWADDR:
3412                 bond_set_dev_addr(bond_dev, slave_dev);
3413                 res = 0;
3414                 break;
3415         case BOND_CHANGE_ACTIVE_OLD:
3416         case SIOCBONDCHANGEACTIVE:
3417                 bond_opt_initstr(&newval, slave_dev->name);
3418                 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3419                 break;
3420         default:
3421                 res = -EOPNOTSUPP;
3422         }
3423
3424         return res;
3425 }
3426
3427 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3428 {
3429         struct bonding *bond = netdev_priv(bond_dev);
3430
3431         if (change & IFF_PROMISC)
3432                 bond_set_promiscuity(bond,
3433                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3434
3435         if (change & IFF_ALLMULTI)
3436                 bond_set_allmulti(bond,
3437                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3438 }
3439
3440 static void bond_set_rx_mode(struct net_device *bond_dev)
3441 {
3442         struct bonding *bond = netdev_priv(bond_dev);
3443         struct list_head *iter;
3444         struct slave *slave;
3445
3446         rcu_read_lock();
3447         if (bond_uses_primary(bond)) {
3448                 slave = rcu_dereference(bond->curr_active_slave);
3449                 if (slave) {
3450                         dev_uc_sync(slave->dev, bond_dev);
3451                         dev_mc_sync(slave->dev, bond_dev);
3452                 }
3453         } else {
3454                 bond_for_each_slave_rcu(bond, slave, iter) {
3455                         dev_uc_sync_multiple(slave->dev, bond_dev);
3456                         dev_mc_sync_multiple(slave->dev, bond_dev);
3457                 }
3458         }
3459         rcu_read_unlock();
3460 }
3461
3462 static int bond_neigh_init(struct neighbour *n)
3463 {
3464         struct bonding *bond = netdev_priv(n->dev);
3465         const struct net_device_ops *slave_ops;
3466         struct neigh_parms parms;
3467         struct slave *slave;
3468         int ret;
3469
3470         slave = bond_first_slave(bond);
3471         if (!slave)
3472                 return 0;
3473         slave_ops = slave->dev->netdev_ops;
3474         if (!slave_ops->ndo_neigh_setup)
3475                 return 0;
3476
3477         parms.neigh_setup = NULL;
3478         parms.neigh_cleanup = NULL;
3479         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3480         if (ret)
3481                 return ret;
3482
3483         /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3484          * after the last slave has been detached.  Assumes that all slaves
3485          * utilize the same neigh_cleanup (true at this writing as only user
3486          * is ipoib).
3487          */
3488         n->parms->neigh_cleanup = parms.neigh_cleanup;
3489
3490         if (!parms.neigh_setup)
3491                 return 0;
3492
3493         return parms.neigh_setup(n);
3494 }
3495
3496 /* The bonding ndo_neigh_setup is called at init time beofre any
3497  * slave exists. So we must declare proxy setup function which will
3498  * be used at run time to resolve the actual slave neigh param setup.
3499  *
3500  * It's also called by master devices (such as vlans) to setup their
3501  * underlying devices. In that case - do nothing, we're already set up from
3502  * our init.
3503  */
3504 static int bond_neigh_setup(struct net_device *dev,
3505                             struct neigh_parms *parms)
3506 {
3507         /* modify only our neigh_parms */
3508         if (parms->dev == dev)
3509                 parms->neigh_setup = bond_neigh_init;
3510
3511         return 0;
3512 }
3513
3514 /* Change the MTU of all of a master's slaves to match the master */
3515 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3516 {
3517         struct bonding *bond = netdev_priv(bond_dev);
3518         struct slave *slave, *rollback_slave;
3519         struct list_head *iter;
3520         int res = 0;
3521
3522         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3523
3524         bond_for_each_slave(bond, slave, iter) {
3525                 netdev_dbg(bond_dev, "s %p c_m %p\n",
3526                            slave, slave->dev->netdev_ops->ndo_change_mtu);
3527
3528                 res = dev_set_mtu(slave->dev, new_mtu);
3529
3530                 if (res) {
3531                         /* If we failed to set the slave's mtu to the new value
3532                          * we must abort the operation even in ACTIVE_BACKUP
3533                          * mode, because if we allow the backup slaves to have
3534                          * different mtu values than the active slave we'll
3535                          * need to change their mtu when doing a failover. That
3536                          * means changing their mtu from timer context, which
3537                          * is probably not a good idea.
3538                          */
3539                         netdev_dbg(bond_dev, "err %d %s\n", res,
3540                                    slave->dev->name);
3541                         goto unwind;
3542                 }
3543         }
3544
3545         bond_dev->mtu = new_mtu;
3546
3547         return 0;
3548
3549 unwind:
3550         /* unwind from head to the slave that failed */
3551         bond_for_each_slave(bond, rollback_slave, iter) {
3552                 int tmp_res;
3553
3554                 if (rollback_slave == slave)
3555                         break;
3556
3557                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3558                 if (tmp_res) {
3559                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3560                                    tmp_res, rollback_slave->dev->name);
3561                 }
3562         }
3563
3564         return res;
3565 }
3566
3567 /* Change HW address
3568  *
3569  * Note that many devices must be down to change the HW address, and
3570  * downing the master releases all slaves.  We can make bonds full of
3571  * bonding devices to test this, however.
3572  */
3573 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3574 {
3575         struct bonding *bond = netdev_priv(bond_dev);
3576         struct slave *slave, *rollback_slave;
3577         struct sockaddr *sa = addr, tmp_sa;
3578         struct list_head *iter;
3579         int res = 0;
3580
3581         if (BOND_MODE(bond) == BOND_MODE_ALB)
3582                 return bond_alb_set_mac_address(bond_dev, addr);
3583
3584
3585         netdev_dbg(bond_dev, "bond=%p\n", bond);
3586
3587         /* If fail_over_mac is enabled, do nothing and return success.
3588          * Returning an error causes ifenslave to fail.
3589          */
3590         if (bond->params.fail_over_mac &&
3591             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3592                 return 0;
3593
3594         if (!is_valid_ether_addr(sa->sa_data))
3595                 return -EADDRNOTAVAIL;
3596
3597         bond_for_each_slave(bond, slave, iter) {
3598                 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3599                 res = dev_set_mac_address(slave->dev, addr);
3600                 if (res) {
3601                         /* TODO: consider downing the slave
3602                          * and retry ?
3603                          * User should expect communications
3604                          * breakage anyway until ARP finish
3605                          * updating, so...
3606                          */
3607                         netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3608                         goto unwind;
3609                 }
3610         }
3611
3612         /* success */
3613         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3614         return 0;
3615
3616 unwind:
3617         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3618         tmp_sa.sa_family = bond_dev->type;
3619
3620         /* unwind from head to the slave that failed */
3621         bond_for_each_slave(bond, rollback_slave, iter) {
3622                 int tmp_res;
3623
3624                 if (rollback_slave == slave)
3625                         break;
3626
3627                 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3628                 if (tmp_res) {
3629                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3630                                    tmp_res, rollback_slave->dev->name);
3631                 }
3632         }
3633
3634         return res;
3635 }
3636
3637 /**
3638  * bond_xmit_slave_id - transmit skb through slave with slave_id
3639  * @bond: bonding device that is transmitting
3640  * @skb: buffer to transmit
3641  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3642  *
3643  * This function tries to transmit through slave with slave_id but in case
3644  * it fails, it tries to find the first available slave for transmission.
3645  * The skb is consumed in all cases, thus the function is void.
3646  */
3647 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3648 {
3649         struct list_head *iter;
3650         struct slave *slave;
3651         int i = slave_id;
3652
3653         /* Here we start from the slave with slave_id */
3654         bond_for_each_slave_rcu(bond, slave, iter) {
3655                 if (--i < 0) {
3656                         if (bond_slave_can_tx(slave)) {
3657                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3658                                 return;
3659                         }
3660                 }
3661         }
3662
3663         /* Here we start from the first slave up to slave_id */
3664         i = slave_id;
3665         bond_for_each_slave_rcu(bond, slave, iter) {
3666                 if (--i < 0)
3667                         break;
3668                 if (bond_slave_can_tx(slave)) {
3669                         bond_dev_queue_xmit(bond, skb, slave->dev);
3670                         return;
3671                 }
3672         }
3673         /* no slave that can tx has been found */
3674         bond_tx_drop(bond->dev, skb);
3675 }
3676
3677 /**
3678  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3679  * @bond: bonding device to use
3680  *
3681  * Based on the value of the bonding device's packets_per_slave parameter
3682  * this function generates a slave id, which is usually used as the next
3683  * slave to transmit through.
3684  */
3685 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3686 {
3687         u32 slave_id;
3688         struct reciprocal_value reciprocal_packets_per_slave;
3689         int packets_per_slave = bond->params.packets_per_slave;
3690
3691         switch (packets_per_slave) {
3692         case 0:
3693                 slave_id = prandom_u32();
3694                 break;
3695         case 1:
3696                 slave_id = bond->rr_tx_counter;
3697                 break;
3698         default:
3699                 reciprocal_packets_per_slave =
3700                         bond->params.reciprocal_packets_per_slave;
3701                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3702                                              reciprocal_packets_per_slave);
3703                 break;
3704         }
3705         bond->rr_tx_counter++;
3706
3707         return slave_id;
3708 }
3709
3710 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3711 {
3712         struct bonding *bond = netdev_priv(bond_dev);
3713         struct iphdr *iph = ip_hdr(skb);
3714         struct slave *slave;
3715         u32 slave_id;
3716
3717         /* Start with the curr_active_slave that joined the bond as the
3718          * default for sending IGMP traffic.  For failover purposes one
3719          * needs to maintain some consistency for the interface that will
3720          * send the join/membership reports.  The curr_active_slave found
3721          * will send all of this type of traffic.
3722          */
3723         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3724                 slave = rcu_dereference(bond->curr_active_slave);
3725                 if (slave)
3726                         bond_dev_queue_xmit(bond, skb, slave->dev);
3727                 else
3728                         bond_xmit_slave_id(bond, skb, 0);
3729         } else {
3730                 int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3731
3732                 if (likely(slave_cnt)) {
3733                         slave_id = bond_rr_gen_slave_id(bond);
3734                         bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3735                 } else {
3736                         bond_tx_drop(bond_dev, skb);
3737                 }
3738         }
3739
3740         return NETDEV_TX_OK;
3741 }
3742
3743 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3744  * the bond has a usable interface.
3745  */
3746 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3747 {
3748         struct bonding *bond = netdev_priv(bond_dev);
3749         struct slave *slave;
3750
3751         slave = rcu_dereference(bond->curr_active_slave);
3752         if (slave)
3753                 bond_dev_queue_xmit(bond, skb, slave->dev);
3754         else
3755                 bond_tx_drop(bond_dev, skb);
3756
3757         return NETDEV_TX_OK;
3758 }
3759
3760 /* Use this to update slave_array when (a) it's not appropriate to update
3761  * slave_array right away (note that update_slave_array() may sleep)
3762  * and / or (b) RTNL is not held.
3763  */
3764 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3765 {
3766         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3767 }
3768
3769 /* Slave array work handler. Holds only RTNL */
3770 static void bond_slave_arr_handler(struct work_struct *work)
3771 {
3772         struct bonding *bond = container_of(work, struct bonding,
3773                                             slave_arr_work.work);
3774         int ret;
3775
3776         if (!rtnl_trylock())
3777                 goto err;
3778
3779         ret = bond_update_slave_arr(bond, NULL);
3780         rtnl_unlock();
3781         if (ret) {
3782                 pr_warn_ratelimited("Failed to update slave array from WT\n");
3783                 goto err;
3784         }
3785         return;
3786
3787 err:
3788         bond_slave_arr_work_rearm(bond, 1);
3789 }
3790
3791 /* Build the usable slaves array in control path for modes that use xmit-hash
3792  * to determine the slave interface -
3793  * (a) BOND_MODE_8023AD
3794  * (b) BOND_MODE_XOR
3795  * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3796  *
3797  * The caller is expected to hold RTNL only and NO other lock!
3798  */
3799 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3800 {
3801         struct slave *slave;
3802         struct list_head *iter;
3803         struct bond_up_slave *new_arr, *old_arr;
3804         int agg_id = 0;
3805         int ret = 0;
3806
3807 #ifdef CONFIG_LOCKDEP
3808         WARN_ON(lockdep_is_held(&bond->mode_lock));
3809 #endif
3810
3811         new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3812                           GFP_KERNEL);
3813         if (!new_arr) {
3814                 ret = -ENOMEM;
3815                 pr_err("Failed to build slave-array.\n");
3816                 goto out;
3817         }
3818         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3819                 struct ad_info ad_info;
3820
3821                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3822                         pr_debug("bond_3ad_get_active_agg_info failed\n");
3823                         kfree_rcu(new_arr, rcu);
3824                         /* No active aggragator means it's not safe to use
3825                          * the previous array.
3826                          */
3827                         old_arr = rtnl_dereference(bond->slave_arr);
3828                         if (old_arr) {
3829                                 RCU_INIT_POINTER(bond->slave_arr, NULL);
3830                                 kfree_rcu(old_arr, rcu);
3831                         }
3832                         goto out;
3833                 }
3834                 agg_id = ad_info.aggregator_id;
3835         }
3836         bond_for_each_slave(bond, slave, iter) {
3837                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3838                         struct aggregator *agg;
3839
3840                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
3841                         if (!agg || agg->aggregator_identifier != agg_id)
3842                                 continue;
3843                 }
3844                 if (!bond_slave_can_tx(slave))
3845                         continue;
3846                 if (skipslave == slave)
3847                         continue;
3848                 new_arr->arr[new_arr->count++] = slave;
3849         }
3850
3851         old_arr = rtnl_dereference(bond->slave_arr);
3852         rcu_assign_pointer(bond->slave_arr, new_arr);
3853         if (old_arr)
3854                 kfree_rcu(old_arr, rcu);
3855 out:
3856         if (ret != 0 && skipslave) {
3857                 int idx;
3858
3859                 /* Rare situation where caller has asked to skip a specific
3860                  * slave but allocation failed (most likely!). BTW this is
3861                  * only possible when the call is initiated from
3862                  * __bond_release_one(). In this situation; overwrite the
3863                  * skipslave entry in the array with the last entry from the
3864                  * array to avoid a situation where the xmit path may choose
3865                  * this to-be-skipped slave to send a packet out.
3866                  */
3867                 old_arr = rtnl_dereference(bond->slave_arr);
3868                 for (idx = 0; idx < old_arr->count; idx++) {
3869                         if (skipslave == old_arr->arr[idx]) {
3870                                 old_arr->arr[idx] =
3871                                     old_arr->arr[old_arr->count-1];
3872                                 old_arr->count--;
3873                                 break;
3874                         }
3875                 }
3876         }
3877         return ret;
3878 }
3879
3880 /* Use this Xmit function for 3AD as well as XOR modes. The current
3881  * usable slave array is formed in the control path. The xmit function
3882  * just calculates hash and sends the packet out.
3883  */
3884 static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3885 {
3886         struct bonding *bond = netdev_priv(dev);
3887         struct slave *slave;
3888         struct bond_up_slave *slaves;
3889         unsigned int count;
3890
3891         slaves = rcu_dereference(bond->slave_arr);
3892         count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3893         if (likely(count)) {
3894                 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3895                 bond_dev_queue_xmit(bond, skb, slave->dev);
3896         } else {
3897                 bond_tx_drop(dev, skb);
3898         }
3899
3900         return NETDEV_TX_OK;
3901 }
3902
3903 /* in broadcast mode, we send everything to all usable interfaces. */
3904 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3905 {
3906         struct bonding *bond = netdev_priv(bond_dev);
3907         struct slave *slave = NULL;
3908         struct list_head *iter;
3909
3910         bond_for_each_slave_rcu(bond, slave, iter) {
3911                 if (bond_is_last_slave(bond, slave))
3912                         break;
3913                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3914                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3915
3916                         if (!skb2) {
3917                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3918                                                     bond_dev->name, __func__);
3919                                 continue;
3920                         }
3921                         bond_dev_queue_xmit(bond, skb2, slave->dev);
3922                 }
3923         }
3924         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3925                 bond_dev_queue_xmit(bond, skb, slave->dev);
3926         else
3927                 bond_tx_drop(bond_dev, skb);
3928
3929         return NETDEV_TX_OK;
3930 }
3931
3932 /*------------------------- Device initialization ---------------------------*/
3933
3934 /* Lookup the slave that corresponds to a qid */
3935 static inline int bond_slave_override(struct bonding *bond,
3936                                       struct sk_buff *skb)
3937 {
3938         struct slave *slave = NULL;
3939         struct list_head *iter;
3940
3941         if (!skb->queue_mapping)
3942                 return 1;
3943
3944         /* Find out if any slaves have the same mapping as this skb. */
3945         bond_for_each_slave_rcu(bond, slave, iter) {
3946                 if (slave->queue_id == skb->queue_mapping) {
3947                         if (bond_slave_is_up(slave) &&
3948                             slave->link == BOND_LINK_UP) {
3949                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3950                                 return 0;
3951                         }
3952                         /* If the slave isn't UP, use default transmit policy. */
3953                         break;
3954                 }
3955         }
3956
3957         return 1;
3958 }
3959
3960
3961 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3962                              void *accel_priv, select_queue_fallback_t fallback)
3963 {
3964         /* This helper function exists to help dev_pick_tx get the correct
3965          * destination queue.  Using a helper function skips a call to
3966          * skb_tx_hash and will put the skbs in the queue we expect on their
3967          * way down to the bonding driver.
3968          */
3969         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3970
3971         /* Save the original txq to restore before passing to the driver */
3972         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3973
3974         if (unlikely(txq >= dev->real_num_tx_queues)) {
3975                 do {
3976                         txq -= dev->real_num_tx_queues;
3977                 } while (txq >= dev->real_num_tx_queues);
3978         }
3979         return txq;
3980 }
3981
3982 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3983 {
3984         struct bonding *bond = netdev_priv(dev);
3985
3986         if (bond_should_override_tx_queue(bond) &&
3987             !bond_slave_override(bond, skb))
3988                 return NETDEV_TX_OK;
3989
3990         switch (BOND_MODE(bond)) {
3991         case BOND_MODE_ROUNDROBIN:
3992                 return bond_xmit_roundrobin(skb, dev);
3993         case BOND_MODE_ACTIVEBACKUP:
3994                 return bond_xmit_activebackup(skb, dev);
3995         case BOND_MODE_8023AD:
3996         case BOND_MODE_XOR:
3997                 return bond_3ad_xor_xmit(skb, dev);
3998         case BOND_MODE_BROADCAST:
3999                 return bond_xmit_broadcast(skb, dev);
4000         case BOND_MODE_ALB:
4001                 return bond_alb_xmit(skb, dev);
4002         case BOND_MODE_TLB:
4003                 return bond_tlb_xmit(skb, dev);
4004         default:
4005                 /* Should never happen, mode already checked */
4006                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4007                 WARN_ON_ONCE(1);
4008                 bond_tx_drop(dev, skb);
4009                 return NETDEV_TX_OK;
4010         }
4011 }
4012
4013 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4014 {
4015         struct bonding *bond = netdev_priv(dev);
4016         netdev_tx_t ret = NETDEV_TX_OK;
4017
4018         /* If we risk deadlock from transmitting this in the
4019          * netpoll path, tell netpoll to queue the frame for later tx
4020          */
4021         if (unlikely(is_netpoll_tx_blocked(dev)))
4022                 return NETDEV_TX_BUSY;
4023
4024         rcu_read_lock();
4025         if (bond_has_slaves(bond))
4026                 ret = __bond_start_xmit(skb, dev);
4027         else
4028                 bond_tx_drop(dev, skb);
4029         rcu_read_unlock();
4030
4031         return ret;
4032 }
4033
4034 static int bond_ethtool_get_settings(struct net_device *bond_dev,
4035                                      struct ethtool_cmd *ecmd)
4036 {
4037         struct bonding *bond = netdev_priv(bond_dev);
4038         unsigned long speed = 0;
4039         struct list_head *iter;
4040         struct slave *slave;
4041
4042         ecmd->duplex = DUPLEX_UNKNOWN;
4043         ecmd->port = PORT_OTHER;
4044
4045         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4046          * do not need to check mode.  Though link speed might not represent
4047          * the true receive or transmit bandwidth (not all modes are symmetric)
4048          * this is an accurate maximum.
4049          */
4050         bond_for_each_slave(bond, slave, iter) {
4051                 if (bond_slave_can_tx(slave)) {
4052                         if (slave->speed != SPEED_UNKNOWN)
4053                                 speed += slave->speed;
4054                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
4055                             slave->duplex != DUPLEX_UNKNOWN)
4056                                 ecmd->duplex = slave->duplex;
4057                 }
4058         }
4059         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
4060
4061         return 0;
4062 }
4063
4064 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4065                                      struct ethtool_drvinfo *drvinfo)
4066 {
4067         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4068         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4069         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4070                  BOND_ABI_VERSION);
4071 }
4072
4073 static const struct ethtool_ops bond_ethtool_ops = {
4074         .get_drvinfo            = bond_ethtool_get_drvinfo,
4075         .get_settings           = bond_ethtool_get_settings,
4076         .get_link               = ethtool_op_get_link,
4077 };
4078
4079 static const struct net_device_ops bond_netdev_ops = {
4080         .ndo_init               = bond_init,
4081         .ndo_uninit             = bond_uninit,
4082         .ndo_open               = bond_open,
4083         .ndo_stop               = bond_close,
4084         .ndo_start_xmit         = bond_start_xmit,
4085         .ndo_select_queue       = bond_select_queue,
4086         .ndo_get_stats64        = bond_get_stats,
4087         .ndo_do_ioctl           = bond_do_ioctl,
4088         .ndo_change_rx_flags    = bond_change_rx_flags,
4089         .ndo_set_rx_mode        = bond_set_rx_mode,
4090         .ndo_change_mtu         = bond_change_mtu,
4091         .ndo_set_mac_address    = bond_set_mac_address,
4092         .ndo_neigh_setup        = bond_neigh_setup,
4093         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4094         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4095 #ifdef CONFIG_NET_POLL_CONTROLLER
4096         .ndo_netpoll_setup      = bond_netpoll_setup,
4097         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4098         .ndo_poll_controller    = bond_poll_controller,
4099 #endif
4100         .ndo_add_slave          = bond_enslave,
4101         .ndo_del_slave          = bond_release,
4102         .ndo_fix_features       = bond_fix_features,
4103         .ndo_bridge_setlink     = switchdev_port_bridge_setlink,
4104         .ndo_bridge_getlink     = switchdev_port_bridge_getlink,
4105         .ndo_bridge_dellink     = switchdev_port_bridge_dellink,
4106         .ndo_fdb_add            = switchdev_port_fdb_add,
4107         .ndo_fdb_del            = switchdev_port_fdb_del,
4108         .ndo_fdb_dump           = switchdev_port_fdb_dump,
4109         .ndo_features_check     = passthru_features_check,
4110 };
4111
4112 static const struct device_type bond_type = {
4113         .name = "bond",
4114 };
4115
4116 static void bond_destructor(struct net_device *bond_dev)
4117 {
4118         struct bonding *bond = netdev_priv(bond_dev);
4119         if (bond->wq)
4120                 destroy_workqueue(bond->wq);
4121         free_netdev(bond_dev);
4122 }
4123
4124 void bond_setup(struct net_device *bond_dev)
4125 {
4126         struct bonding *bond = netdev_priv(bond_dev);
4127
4128         spin_lock_init(&bond->mode_lock);
4129         spin_lock_init(&bond->stats_lock);
4130         bond->params = bonding_defaults;
4131
4132         /* Initialize pointers */
4133         bond->dev = bond_dev;
4134
4135         /* Initialize the device entry points */
4136         ether_setup(bond_dev);
4137         bond_dev->netdev_ops = &bond_netdev_ops;
4138         bond_dev->ethtool_ops = &bond_ethtool_ops;
4139
4140         bond_dev->destructor = bond_destructor;
4141
4142         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4143
4144         /* Initialize the device options */
4145         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4146         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4147         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4148
4149         /* don't acquire bond device's netif_tx_lock when transmitting */
4150         bond_dev->features |= NETIF_F_LLTX;
4151
4152         /* By default, we declare the bond to be fully
4153          * VLAN hardware accelerated capable. Special
4154          * care is taken in the various xmit functions
4155          * when there are slaves that are not hw accel
4156          * capable
4157          */
4158
4159         /* Don't allow bond devices to change network namespaces. */
4160         bond_dev->features |= NETIF_F_NETNS_LOCAL;
4161
4162         bond_dev->hw_features = BOND_VLAN_FEATURES |
4163                                 NETIF_F_HW_VLAN_CTAG_TX |
4164                                 NETIF_F_HW_VLAN_CTAG_RX |
4165                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4166
4167         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4168         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4169         bond_dev->features |= bond_dev->hw_features;
4170 }
4171
4172 /* Destroy a bonding device.
4173  * Must be under rtnl_lock when this function is called.
4174  */
4175 static void bond_uninit(struct net_device *bond_dev)
4176 {
4177         struct bonding *bond = netdev_priv(bond_dev);
4178         struct list_head *iter;
4179         struct slave *slave;
4180         struct bond_up_slave *arr;
4181
4182         bond_netpoll_cleanup(bond_dev);
4183
4184         /* Release the bonded slaves */
4185         bond_for_each_slave(bond, slave, iter)
4186                 __bond_release_one(bond_dev, slave->dev, true);
4187         netdev_info(bond_dev, "Released all slaves\n");
4188
4189         arr = rtnl_dereference(bond->slave_arr);
4190         if (arr) {
4191                 RCU_INIT_POINTER(bond->slave_arr, NULL);
4192                 kfree_rcu(arr, rcu);
4193         }
4194
4195         list_del(&bond->bond_list);
4196
4197         bond_debug_unregister(bond);
4198 }
4199
4200 /*------------------------- Module initialization ---------------------------*/
4201
4202 static int bond_check_params(struct bond_params *params)
4203 {
4204         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4205         struct bond_opt_value newval;
4206         const struct bond_opt_value *valptr;
4207         int arp_all_targets_value;
4208         u16 ad_actor_sys_prio = 0;
4209         u16 ad_user_port_key = 0;
4210
4211         /* Convert string parameters. */
4212         if (mode) {
4213                 bond_opt_initstr(&newval, mode);
4214                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4215                 if (!valptr) {
4216                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4217                         return -EINVAL;
4218                 }
4219                 bond_mode = valptr->value;
4220         }
4221
4222         if (xmit_hash_policy) {
4223                 if ((bond_mode != BOND_MODE_XOR) &&
4224                     (bond_mode != BOND_MODE_8023AD) &&
4225                     (bond_mode != BOND_MODE_TLB)) {
4226                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4227                                 bond_mode_name(bond_mode));
4228                 } else {
4229                         bond_opt_initstr(&newval, xmit_hash_policy);
4230                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4231                                                 &newval);
4232                         if (!valptr) {
4233                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4234                                        xmit_hash_policy);
4235                                 return -EINVAL;
4236                         }
4237                         xmit_hashtype = valptr->value;
4238                 }
4239         }
4240
4241         if (lacp_rate) {
4242                 if (bond_mode != BOND_MODE_8023AD) {
4243                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4244                                 bond_mode_name(bond_mode));
4245                 } else {
4246                         bond_opt_initstr(&newval, lacp_rate);
4247                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4248                                                 &newval);
4249                         if (!valptr) {
4250                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4251                                        lacp_rate);
4252                                 return -EINVAL;
4253                         }
4254                         lacp_fast = valptr->value;
4255                 }
4256         }
4257
4258         if (ad_select) {
4259                 bond_opt_initstr(&newval, ad_select);
4260                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4261                                         &newval);
4262                 if (!valptr) {
4263                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4264                         return -EINVAL;
4265                 }
4266                 params->ad_select = valptr->value;
4267                 if (bond_mode != BOND_MODE_8023AD)
4268                         pr_warn("ad_select param only affects 802.3ad mode\n");
4269         } else {
4270                 params->ad_select = BOND_AD_STABLE;
4271         }
4272
4273         if (max_bonds < 0) {
4274                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4275                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4276                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4277         }
4278
4279         if (miimon < 0) {
4280                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4281                         miimon, INT_MAX);
4282                 miimon = 0;
4283         }
4284
4285         if (updelay < 0) {
4286                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4287                         updelay, INT_MAX);
4288                 updelay = 0;
4289         }
4290
4291         if (downdelay < 0) {
4292                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4293                         downdelay, INT_MAX);
4294                 downdelay = 0;
4295         }
4296
4297         if ((use_carrier != 0) && (use_carrier != 1)) {
4298                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4299                         use_carrier);
4300                 use_carrier = 1;
4301         }
4302
4303         if (num_peer_notif < 0 || num_peer_notif > 255) {
4304                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4305                         num_peer_notif);
4306                 num_peer_notif = 1;
4307         }
4308
4309         /* reset values for 802.3ad/TLB/ALB */
4310         if (!bond_mode_uses_arp(bond_mode)) {
4311                 if (!miimon) {
4312                         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");
4313                         pr_warn("Forcing miimon to 100msec\n");
4314                         miimon = BOND_DEFAULT_MIIMON;
4315                 }
4316         }
4317
4318         if (tx_queues < 1 || tx_queues > 255) {
4319                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4320                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4321                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4322         }
4323
4324         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4325                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4326                         all_slaves_active);
4327                 all_slaves_active = 0;
4328         }
4329
4330         if (resend_igmp < 0 || resend_igmp > 255) {
4331                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4332                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4333                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4334         }
4335
4336         bond_opt_initval(&newval, packets_per_slave);
4337         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4338                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4339                         packets_per_slave, USHRT_MAX);
4340                 packets_per_slave = 1;
4341         }
4342
4343         if (bond_mode == BOND_MODE_ALB) {
4344                 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",
4345                           updelay);
4346         }
4347
4348         if (!miimon) {
4349                 if (updelay || downdelay) {
4350                         /* just warn the user the up/down delay will have
4351                          * no effect since miimon is zero...
4352                          */
4353                         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",
4354                                 updelay, downdelay);
4355                 }
4356         } else {
4357                 /* don't allow arp monitoring */
4358                 if (arp_interval) {
4359                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4360                                 miimon, arp_interval);
4361                         arp_interval = 0;
4362                 }
4363
4364                 if ((updelay % miimon) != 0) {
4365                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4366                                 updelay, miimon, (updelay / miimon) * miimon);
4367                 }
4368
4369                 updelay /= miimon;
4370
4371                 if ((downdelay % miimon) != 0) {
4372                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4373                                 downdelay, miimon,
4374                                 (downdelay / miimon) * miimon);
4375                 }
4376
4377                 downdelay /= miimon;
4378         }
4379
4380         if (arp_interval < 0) {
4381                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4382                         arp_interval, INT_MAX);
4383                 arp_interval = 0;
4384         }
4385
4386         for (arp_ip_count = 0, i = 0;
4387              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4388                 __be32 ip;
4389
4390                 /* not a complete check, but good enough to catch mistakes */
4391                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4392                     !bond_is_ip_target_ok(ip)) {
4393                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4394                                 arp_ip_target[i]);
4395                         arp_interval = 0;
4396                 } else {
4397                         if (bond_get_targets_ip(arp_target, ip) == -1)
4398                                 arp_target[arp_ip_count++] = ip;
4399                         else
4400                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4401                                         &ip);
4402                 }
4403         }
4404
4405         if (arp_interval && !arp_ip_count) {
4406                 /* don't allow arping if no arp_ip_target given... */
4407                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4408                         arp_interval);
4409                 arp_interval = 0;
4410         }
4411
4412         if (arp_validate) {
4413                 if (!arp_interval) {
4414                         pr_err("arp_validate requires arp_interval\n");
4415                         return -EINVAL;
4416                 }
4417
4418                 bond_opt_initstr(&newval, arp_validate);
4419                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4420                                         &newval);
4421                 if (!valptr) {
4422                         pr_err("Error: invalid arp_validate \"%s\"\n",
4423                                arp_validate);
4424                         return -EINVAL;
4425                 }
4426                 arp_validate_value = valptr->value;
4427         } else {
4428                 arp_validate_value = 0;
4429         }
4430
4431         arp_all_targets_value = 0;
4432         if (arp_all_targets) {
4433                 bond_opt_initstr(&newval, arp_all_targets);
4434                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4435                                         &newval);
4436                 if (!valptr) {
4437                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4438                                arp_all_targets);
4439                         arp_all_targets_value = 0;
4440                 } else {
4441                         arp_all_targets_value = valptr->value;
4442                 }
4443         }
4444
4445         if (miimon) {
4446                 pr_info("MII link monitoring set to %d ms\n", miimon);
4447         } else if (arp_interval) {
4448                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4449                                           arp_validate_value);
4450                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4451                         arp_interval, valptr->string, arp_ip_count);
4452
4453                 for (i = 0; i < arp_ip_count; i++)
4454                         pr_cont(" %s", arp_ip_target[i]);
4455
4456                 pr_cont("\n");
4457
4458         } else if (max_bonds) {
4459                 /* miimon and arp_interval not set, we need one so things
4460                  * work as expected, see bonding.txt for details
4461                  */
4462                 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");
4463         }
4464
4465         if (primary && !bond_mode_uses_primary(bond_mode)) {
4466                 /* currently, using a primary only makes sense
4467                  * in active backup, TLB or ALB modes
4468                  */
4469                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4470                         primary, bond_mode_name(bond_mode));
4471                 primary = NULL;
4472         }
4473
4474         if (primary && primary_reselect) {
4475                 bond_opt_initstr(&newval, primary_reselect);
4476                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4477                                         &newval);
4478                 if (!valptr) {
4479                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4480                                primary_reselect);
4481                         return -EINVAL;
4482                 }
4483                 primary_reselect_value = valptr->value;
4484         } else {
4485                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4486         }
4487
4488         if (fail_over_mac) {
4489                 bond_opt_initstr(&newval, fail_over_mac);
4490                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4491                                         &newval);
4492                 if (!valptr) {
4493                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4494                                fail_over_mac);
4495                         return -EINVAL;
4496                 }
4497                 fail_over_mac_value = valptr->value;
4498                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4499                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4500         } else {
4501                 fail_over_mac_value = BOND_FOM_NONE;
4502         }
4503
4504         bond_opt_initstr(&newval, "default");
4505         valptr = bond_opt_parse(
4506                         bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4507                                      &newval);
4508         if (!valptr) {
4509                 pr_err("Error: No ad_actor_sys_prio default value");
4510                 return -EINVAL;
4511         }
4512         ad_actor_sys_prio = valptr->value;
4513
4514         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4515                                 &newval);
4516         if (!valptr) {
4517                 pr_err("Error: No ad_user_port_key default value");
4518                 return -EINVAL;
4519         }
4520         ad_user_port_key = valptr->value;
4521
4522         if (lp_interval == 0) {
4523                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4524                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4525                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4526         }
4527
4528         /* fill params struct with the proper values */
4529         params->mode = bond_mode;
4530         params->xmit_policy = xmit_hashtype;
4531         params->miimon = miimon;
4532         params->num_peer_notif = num_peer_notif;
4533         params->arp_interval = arp_interval;
4534         params->arp_validate = arp_validate_value;
4535         params->arp_all_targets = arp_all_targets_value;
4536         params->updelay = updelay;
4537         params->downdelay = downdelay;
4538         params->use_carrier = use_carrier;
4539         params->lacp_fast = lacp_fast;
4540         params->primary[0] = 0;
4541         params->primary_reselect = primary_reselect_value;
4542         params->fail_over_mac = fail_over_mac_value;
4543         params->tx_queues = tx_queues;
4544         params->all_slaves_active = all_slaves_active;
4545         params->resend_igmp = resend_igmp;
4546         params->min_links = min_links;
4547         params->lp_interval = lp_interval;
4548         params->packets_per_slave = packets_per_slave;
4549         params->tlb_dynamic_lb = 1; /* Default value */
4550         params->ad_actor_sys_prio = ad_actor_sys_prio;
4551         eth_zero_addr(params->ad_actor_system);
4552         params->ad_user_port_key = ad_user_port_key;
4553         if (packets_per_slave > 0) {
4554                 params->reciprocal_packets_per_slave =
4555                         reciprocal_value(packets_per_slave);
4556         } else {
4557                 /* reciprocal_packets_per_slave is unused if
4558                  * packets_per_slave is 0 or 1, just initialize it
4559                  */
4560                 params->reciprocal_packets_per_slave =
4561                         (struct reciprocal_value) { 0 };
4562         }
4563
4564         if (primary) {
4565                 strncpy(params->primary, primary, IFNAMSIZ);
4566                 params->primary[IFNAMSIZ - 1] = 0;
4567         }
4568
4569         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4570
4571         return 0;
4572 }
4573
4574 static struct lock_class_key bonding_netdev_xmit_lock_key;
4575 static struct lock_class_key bonding_netdev_addr_lock_key;
4576 static struct lock_class_key bonding_tx_busylock_key;
4577
4578 static void bond_set_lockdep_class_one(struct net_device *dev,
4579                                        struct netdev_queue *txq,
4580                                        void *_unused)
4581 {
4582         lockdep_set_class(&txq->_xmit_lock,
4583                           &bonding_netdev_xmit_lock_key);
4584 }
4585
4586 static void bond_set_lockdep_class(struct net_device *dev)
4587 {
4588         lockdep_set_class(&dev->addr_list_lock,
4589                           &bonding_netdev_addr_lock_key);
4590         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4591         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4592 }
4593
4594 /* Called from registration process */
4595 static int bond_init(struct net_device *bond_dev)
4596 {
4597         struct bonding *bond = netdev_priv(bond_dev);
4598         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4599
4600         netdev_dbg(bond_dev, "Begin bond_init\n");
4601
4602         bond->wq = create_singlethread_workqueue(bond_dev->name);
4603         if (!bond->wq)
4604                 return -ENOMEM;
4605
4606         bond_set_lockdep_class(bond_dev);
4607
4608         list_add_tail(&bond->bond_list, &bn->dev_list);
4609
4610         bond_prepare_sysfs_group(bond);
4611
4612         bond_debug_register(bond);
4613
4614         /* Ensure valid dev_addr */
4615         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4616             bond_dev->addr_assign_type == NET_ADDR_PERM)
4617                 eth_hw_addr_random(bond_dev);
4618
4619         return 0;
4620 }
4621
4622 unsigned int bond_get_num_tx_queues(void)
4623 {
4624         return tx_queues;
4625 }
4626
4627 /* Create a new bond based on the specified name and bonding parameters.
4628  * If name is NULL, obtain a suitable "bond%d" name for us.
4629  * Caller must NOT hold rtnl_lock; we need to release it here before we
4630  * set up our sysfs entries.
4631  */
4632 int bond_create(struct net *net, const char *name)
4633 {
4634         struct net_device *bond_dev;
4635         struct bonding *bond;
4636         struct alb_bond_info *bond_info;
4637         int res;
4638
4639         rtnl_lock();
4640
4641         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4642                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
4643                                    bond_setup, tx_queues);
4644         if (!bond_dev) {
4645                 pr_err("%s: eek! can't alloc netdev!\n", name);
4646                 rtnl_unlock();
4647                 return -ENOMEM;
4648         }
4649
4650         /*
4651          * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4652          * It is set to 0 by default which is wrong.
4653          */
4654         bond = netdev_priv(bond_dev);
4655         bond_info = &(BOND_ALB_INFO(bond));
4656         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4657
4658         dev_net_set(bond_dev, net);
4659         bond_dev->rtnl_link_ops = &bond_link_ops;
4660
4661         res = register_netdevice(bond_dev);
4662
4663         netif_carrier_off(bond_dev);
4664
4665         rtnl_unlock();
4666         if (res < 0)
4667                 bond_destructor(bond_dev);
4668         return res;
4669 }
4670
4671 static int __net_init bond_net_init(struct net *net)
4672 {
4673         struct bond_net *bn = net_generic(net, bond_net_id);
4674
4675         bn->net = net;
4676         INIT_LIST_HEAD(&bn->dev_list);
4677
4678         bond_create_proc_dir(bn);
4679         bond_create_sysfs(bn);
4680
4681         return 0;
4682 }
4683
4684 static void __net_exit bond_net_exit(struct net *net)
4685 {
4686         struct bond_net *bn = net_generic(net, bond_net_id);
4687         struct bonding *bond, *tmp_bond;
4688         LIST_HEAD(list);
4689
4690         bond_destroy_sysfs(bn);
4691
4692         /* Kill off any bonds created after unregistering bond rtnl ops */
4693         rtnl_lock();
4694         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4695                 unregister_netdevice_queue(bond->dev, &list);
4696         unregister_netdevice_many(&list);
4697         rtnl_unlock();
4698
4699         bond_destroy_proc_dir(bn);
4700 }
4701
4702 static struct pernet_operations bond_net_ops = {
4703         .init = bond_net_init,
4704         .exit = bond_net_exit,
4705         .id   = &bond_net_id,
4706         .size = sizeof(struct bond_net),
4707 };
4708
4709 static int __init bonding_init(void)
4710 {
4711         int i;
4712         int res;
4713
4714         pr_info("%s", bond_version);
4715
4716         res = bond_check_params(&bonding_defaults);
4717         if (res)
4718                 goto out;
4719
4720         res = register_pernet_subsys(&bond_net_ops);
4721         if (res)
4722                 goto out;
4723
4724         res = bond_netlink_init();
4725         if (res)
4726                 goto err_link;
4727
4728         bond_create_debugfs();
4729
4730         for (i = 0; i < max_bonds; i++) {
4731                 res = bond_create(&init_net, NULL);
4732                 if (res)
4733                         goto err;
4734         }
4735
4736         register_netdevice_notifier(&bond_netdev_notifier);
4737 out:
4738         return res;
4739 err:
4740         bond_destroy_debugfs();
4741         bond_netlink_fini();
4742 err_link:
4743         unregister_pernet_subsys(&bond_net_ops);
4744         goto out;
4745
4746 }
4747
4748 static void __exit bonding_exit(void)
4749 {
4750         unregister_netdevice_notifier(&bond_netdev_notifier);
4751
4752         bond_destroy_debugfs();
4753
4754         bond_netlink_fini();
4755         unregister_pernet_subsys(&bond_net_ops);
4756
4757 #ifdef CONFIG_NET_POLL_CONTROLLER
4758         /* Make sure we don't have an imbalance on our netpoll blocking */
4759         WARN_ON(atomic_read(&netpoll_block_tx));
4760 #endif
4761 }
4762
4763 module_init(bonding_init);
4764 module_exit(bonding_exit);
4765 MODULE_LICENSE("GPL");
4766 MODULE_VERSION(DRV_VERSION);
4767 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4768 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");