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, "802.3ad 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. Return 1 if speed or duplex settings are
 376  * UNKNOWN; 0 otherwise.
 377  */
 378 static int bond_update_speed_duplex(struct slave *slave)
 379 {
 380         struct net_device *slave_dev = slave->dev;
 381         struct ethtool_link_ksettings ecmd;
 382         int res;
 383
 384         slave->speed = SPEED_UNKNOWN;
 385         slave->duplex = DUPLEX_UNKNOWN;
 386
 387         res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
 388         if (res < 0)
 389                 return 1;
 390         if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
 391                 return 1;
 392         switch (ecmd.base.duplex) {
 393         case DUPLEX_FULL:
 394         case DUPLEX_HALF:
 395                 break;
 396         default:
 397                 return 1;
 398         }
 399
 400         slave->speed = ecmd.base.speed;
 401         slave->duplex = ecmd.base.duplex;
 402
 403         return 0;
 404 }
 405
 406 const char *bond_slave_link_status(s8 link)
 407 {
 408         switch (link) {
 409         case BOND_LINK_UP:
 410                 return "up";
 411         case BOND_LINK_FAIL:
 412                 return "going down";
 413         case BOND_LINK_DOWN:
 414                 return "down";
 415         case BOND_LINK_BACK:
 416                 return "going back";
 417         default:
 418                 return "unknown";
 419         }
 420 }
 421
 422 /* if <dev> supports MII link status reporting, check its link status.
 423  *
 424  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 425  * depending upon the setting of the use_carrier parameter.
 426  *
 427  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 428  * can't tell and just pretend it is), or 0, meaning that the link is
 429  * down.
 430  *
 431  * If reporting is non-zero, instead of faking link up, return -1 if
 432  * both ETHTOOL and MII ioctls fail (meaning the device does not
 433  * support them).  If use_carrier is set, return whatever it says.
 434  * It'd be nice if there was a good way to tell if a driver supports
 435  * netif_carrier, but there really isn't.
 436  */
 437 static int bond_check_dev_link(struct bonding *bond,
 438                                struct net_device *slave_dev, int reporting)
 439 {
 440         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 441         int (*ioctl)(struct net_device *, struct ifreq *, int);
 442         struct ifreq ifr;
 443         struct mii_ioctl_data *mii;
 444
 445         if (!reporting && !netif_running(slave_dev))
 446                 return 0;
 447
 448         if (bond->params.use_carrier)
 449                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 450
 451         /* Try to get link status using Ethtool first. */
 452         if (slave_dev->ethtool_ops->get_link)
 453                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 454                         BMSR_LSTATUS : 0;
 455
 456         /* Ethtool can't be used, fallback to MII ioctls. */
 457         ioctl = slave_ops->ndo_do_ioctl;
 458         if (ioctl) {
 459                 /* TODO: set pointer to correct ioctl on a per team member
 460                  *       bases to make this more efficient. that is, once
 461                  *       we determine the correct ioctl, we will always
 462                  *       call it and not the others for that team
 463                  *       member.
 464                  */
 465
 466                 /* We cannot assume that SIOCGMIIPHY will also read a
 467                  * register; not all network drivers (e.g., e100)
 468                  * support that.
 469                  */
 470
 471                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 472                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 473                 mii = if_mii(&ifr);
 474                 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 475                         mii->reg_num = MII_BMSR;
 476                         if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
 477                                 return mii->val_out & BMSR_LSTATUS;
 478                 }
 479         }
 480
 481         /* If reporting, report that either there's no dev->do_ioctl,
 482          * or both SIOCGMIIREG and get_link failed (meaning that we
 483          * cannot report link status).  If not reporting, pretend
 484          * we're ok.
 485          */
 486         return reporting ? -1 : BMSR_LSTATUS;
 487 }
 488
 489 /*----------------------------- Multicast list ------------------------------*/
 490
 491 /* Push the promiscuity flag down to appropriate slaves */
 492 static int bond_set_promiscuity(struct bonding *bond, int inc)
 493 {
 494         struct list_head *iter;
 495         int err = 0;
 496
 497         if (bond_uses_primary(bond)) {
 498                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 499
 500                 if (curr_active)
 501                         err = dev_set_promiscuity(curr_active->dev, inc);
 502         } else {
 503                 struct slave *slave;
 504
 505                 bond_for_each_slave(bond, slave, iter) {
 506                         err = dev_set_promiscuity(slave->dev, inc);
 507                         if (err)
 508                                 return err;
 509                 }
 510         }
 511         return err;
 512 }
 513
 514 /* Push the allmulti flag down to all slaves */
 515 static int bond_set_allmulti(struct bonding *bond, int inc)
 516 {
 517         struct list_head *iter;
 518         int err = 0;
 519
 520         if (bond_uses_primary(bond)) {
 521                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 522
 523                 if (curr_active)
 524                         err = dev_set_allmulti(curr_active->dev, inc);
 525         } else {
 526                 struct slave *slave;
 527
 528                 bond_for_each_slave(bond, slave, iter) {
 529                         err = dev_set_allmulti(slave->dev, inc);
 530                         if (err)
 531                                 return err;
 532                 }
 533         }
 534         return err;
 535 }
 536
 537 /* Retrieve the list of registered multicast addresses for the bonding
 538  * device and retransmit an IGMP JOIN request to the current active
 539  * slave.
 540  */
 541 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 542 {
 543         struct bonding *bond = container_of(work, struct bonding,
 544                                             mcast_work.work);
 545
 546         if (!rtnl_trylock()) {
 547                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 548                 return;
 549         }
 550         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 551
 552         if (bond->igmp_retrans > 1) {
 553                 bond->igmp_retrans--;
 554                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 555         }
 556         rtnl_unlock();
 557 }
 558
 559 /* Flush bond's hardware addresses from slave */
 560 static void bond_hw_addr_flush(struct net_device *bond_dev,
 561                                struct net_device *slave_dev)
 562 {
 563         struct bonding *bond = netdev_priv(bond_dev);
 564
 565         dev_uc_unsync(slave_dev, bond_dev);
 566         dev_mc_unsync(slave_dev, bond_dev);
 567
 568         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 569                 /* del lacpdu mc addr from mc list */
 570                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 571
 572                 dev_mc_del(slave_dev, lacpdu_multicast);
 573         }
 574 }
 575
 576 /*--------------------------- Active slave change ---------------------------*/
 577
 578 /* Update the hardware address list and promisc/allmulti for the new and
 579  * old active slaves (if any).  Modes that are not using primary keep all
 580  * slaves up date at all times; only the modes that use primary need to call
 581  * this function to swap these settings during a failover.
 582  */
 583 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 584                               struct slave *old_active)
 585 {
 586         if (old_active) {
 587                 if (bond->dev->flags & IFF_PROMISC)
 588                         dev_set_promiscuity(old_active->dev, -1);
 589
 590                 if (bond->dev->flags & IFF_ALLMULTI)
 591                         dev_set_allmulti(old_active->dev, -1);
 592
 593                 bond_hw_addr_flush(bond->dev, old_active->dev);
 594         }
 595
 596         if (new_active) {
 597                 /* FIXME: Signal errors upstream. */
 598                 if (bond->dev->flags & IFF_PROMISC)
 599                         dev_set_promiscuity(new_active->dev, 1);
 600
 601                 if (bond->dev->flags & IFF_ALLMULTI)
 602                         dev_set_allmulti(new_active->dev, 1);
 603
 604                 netif_addr_lock_bh(bond->dev);
 605                 dev_uc_sync(new_active->dev, bond->dev);
 606                 dev_mc_sync(new_active->dev, bond->dev);
 607                 netif_addr_unlock_bh(bond->dev);
 608         }
 609 }
 610
 611 /**
 612  * bond_set_dev_addr - clone slave's address to bond
 613  * @bond_dev: bond net device
 614  * @slave_dev: slave net device
 615  *
 616  * Should be called with RTNL held.
 617  */
 618 static void bond_set_dev_addr(struct net_device *bond_dev,
 619                               struct net_device *slave_dev)
 620 {
 621         netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n",
 622                    bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len);
 623         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 624         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 625         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 626 }
 627
 628 static struct slave *bond_get_old_active(struct bonding *bond,
 629                                          struct slave *new_active)
 630 {
 631         struct slave *slave;
 632         struct list_head *iter;
 633
 634         bond_for_each_slave(bond, slave, iter) {
 635                 if (slave == new_active)
 636                         continue;
 637
 638                 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
 639                         return slave;
 640         }
 641
 642         return NULL;
 643 }
 644
 645 /* bond_do_fail_over_mac
 646  *
 647  * Perform special MAC address swapping for fail_over_mac settings
 648  *
 649  * Called with RTNL
 650  */
 651 static void bond_do_fail_over_mac(struct bonding *bond,
 652                                   struct slave *new_active,
 653                                   struct slave *old_active)
 654 {
 655         u8 tmp_mac[ETH_ALEN];
 656         struct sockaddr saddr;
 657         int rv;
 658
 659         switch (bond->params.fail_over_mac) {
 660         case BOND_FOM_ACTIVE:
 661                 if (new_active)
 662                         bond_set_dev_addr(bond->dev, new_active->dev);
 663                 break;
 664         case BOND_FOM_FOLLOW:
 665                 /* if new_active && old_active, swap them
 666                  * if just old_active, do nothing (going to no active slave)
 667                  * if just new_active, set new_active to bond's MAC
 668                  */
 669                 if (!new_active)
 670                         return;
 671
 672                 if (!old_active)
 673                         old_active = bond_get_old_active(bond, new_active);
 674
 675                 if (old_active) {
 676                         ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
 677                         ether_addr_copy(saddr.sa_data,
 678                                         old_active->dev->dev_addr);
 679                         saddr.sa_family = new_active->dev->type;
 680                 } else {
 681                         ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
 682                         saddr.sa_family = bond->dev->type;
 683                 }
 684
 685                 rv = dev_set_mac_address(new_active->dev, &saddr);
 686                 if (rv) {
 687                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 688                                    -rv, new_active->dev->name);
 689                         goto out;
 690                 }
 691
 692                 if (!old_active)
 693                         goto out;
 694
 695                 ether_addr_copy(saddr.sa_data, tmp_mac);
 696                 saddr.sa_family = old_active->dev->type;
 697
 698                 rv = dev_set_mac_address(old_active->dev, &saddr);
 699                 if (rv)
 700                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 701                                    -rv, new_active->dev->name);
 702 out:
 703                 break;
 704         default:
 705                 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
 706                            bond->params.fail_over_mac);
 707                 break;
 708         }
 709
 710 }
 711
 712 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
 713 {
 714         struct slave *prim = rtnl_dereference(bond->primary_slave);
 715         struct slave *curr = rtnl_dereference(bond->curr_active_slave);
 716
 717         if (!prim || prim->link != BOND_LINK_UP) {
 718                 if (!curr || curr->link != BOND_LINK_UP)
 719                         return NULL;
 720                 return curr;
 721         }
 722
 723         if (bond->force_primary) {
 724                 bond->force_primary = false;
 725                 return prim;
 726         }
 727
 728         if (!curr || curr->link != BOND_LINK_UP)
 729                 return prim;
 730
 731         /* At this point, prim and curr are both up */
 732         switch (bond->params.primary_reselect) {
 733         case BOND_PRI_RESELECT_ALWAYS:
 734                 return prim;
 735         case BOND_PRI_RESELECT_BETTER:
 736                 if (prim->speed < curr->speed)
 737                         return curr;
 738                 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
 739                         return curr;
 740                 return prim;
 741         case BOND_PRI_RESELECT_FAILURE:
 742                 return curr;
 743         default:
 744                 netdev_err(bond->dev, "impossible primary_reselect %d\n",
 745                            bond->params.primary_reselect);
 746                 return curr;
 747         }
 748 }
 749
 750 /**
 751  * bond_find_best_slave - select the best available slave to be the active one
 752  * @bond: our bonding struct
 753  */
 754 static struct slave *bond_find_best_slave(struct bonding *bond)
 755 {
 756         struct slave *slave, *bestslave = NULL;
 757         struct list_head *iter;
 758         int mintime = bond->params.updelay;
 759
 760         slave = bond_choose_primary_or_current(bond);
 761         if (slave)
 762                 return slave;
 763
 764         bond_for_each_slave(bond, slave, iter) {
 765                 if (slave->link == BOND_LINK_UP)
 766                         return slave;
 767                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
 768                     slave->delay < mintime) {
 769                         mintime = slave->delay;
 770                         bestslave = slave;
 771                 }
 772         }
 773
 774         return bestslave;
 775 }
 776
 777 static bool bond_should_notify_peers(struct bonding *bond)
 778 {
 779         struct slave *slave;
 780
 781         rcu_read_lock();
 782         slave = rcu_dereference(bond->curr_active_slave);
 783         rcu_read_unlock();
 784
 785         netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
 786                    slave ? slave->dev->name : "NULL");
 787
 788         if (!slave || !bond->send_peer_notif ||
 789             !netif_carrier_ok(bond->dev) ||
 790             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 791                 return false;
 792
 793         return true;
 794 }
 795
 796 /**
 797  * change_active_interface - change the active slave into the specified one
 798  * @bond: our bonding struct
 799  * @new: the new slave to make the active one
 800  *
 801  * Set the new slave to the bond's settings and unset them on the old
 802  * curr_active_slave.
 803  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 804  *
 805  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 806  * because it is apparently the best available slave we have, even though its
 807  * updelay hasn't timed out yet.
 808  *
 809  * Caller must hold RTNL.
 810  */
 811 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 812 {
 813         struct slave *old_active;
 814
 815         ASSERT_RTNL();
 816
 817         old_active = rtnl_dereference(bond->curr_active_slave);
 818
 819         if (old_active == new_active)
 820                 return;
 821
 822         if (new_active) {
 823                 new_active->last_link_up = jiffies;
 824
 825                 if (new_active->link == BOND_LINK_BACK) {
 826                         if (bond_uses_primary(bond)) {
 827                                 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
 828                                             new_active->dev->name,
 829                                             (bond->params.updelay - new_active->delay) * bond->params.miimon);
 830                         }
 831
 832                         new_active->delay = 0;
 833                         bond_set_slave_link_state(new_active, BOND_LINK_UP,
 834                                                   BOND_SLAVE_NOTIFY_NOW);
 835
 836                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 837                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 838
 839                         if (bond_is_lb(bond))
 840                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 841                 } else {
 842                         if (bond_uses_primary(bond)) {
 843                                 netdev_info(bond->dev, "making interface %s the new active one\n",
 844                                             new_active->dev->name);
 845                         }
 846                 }
 847         }
 848
 849         if (bond_uses_primary(bond))
 850                 bond_hw_addr_swap(bond, new_active, old_active);
 851
 852         if (bond_is_lb(bond)) {
 853                 bond_alb_handle_active_change(bond, new_active);
 854                 if (old_active)
 855                         bond_set_slave_inactive_flags(old_active,
 856                                                       BOND_SLAVE_NOTIFY_NOW);
 857                 if (new_active)
 858                         bond_set_slave_active_flags(new_active,
 859                                                     BOND_SLAVE_NOTIFY_NOW);
 860         } else {
 861                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 862         }
 863
 864         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
 865                 if (old_active)
 866                         bond_set_slave_inactive_flags(old_active,
 867                                                       BOND_SLAVE_NOTIFY_NOW);
 868
 869                 if (new_active) {
 870                         bool should_notify_peers = false;
 871
 872                         bond_set_slave_active_flags(new_active,
 873                                                     BOND_SLAVE_NOTIFY_NOW);
 874
 875                         if (bond->params.fail_over_mac)
 876                                 bond_do_fail_over_mac(bond, new_active,
 877                                                       old_active);
 878
 879                         if (netif_running(bond->dev)) {
 880                                 bond->send_peer_notif =
 881                                         bond->params.num_peer_notif;
 882                                 should_notify_peers =
 883                                         bond_should_notify_peers(bond);
 884                         }
 885
 886                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 887                         if (should_notify_peers)
 888                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 889                                                          bond->dev);
 890                 }
 891         }
 892
 893         /* resend IGMP joins since active slave has changed or
 894          * all were sent on curr_active_slave.
 895          * resend only if bond is brought up with the affected
 896          * bonding modes and the retransmission is enabled
 897          */
 898         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 899             ((bond_uses_primary(bond) && new_active) ||
 900              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
 901                 bond->igmp_retrans = bond->params.resend_igmp;
 902                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 903         }
 904 }
 905
 906 /**
 907  * bond_select_active_slave - select a new active slave, if needed
 908  * @bond: our bonding struct
 909  *
 910  * This functions should be called when one of the following occurs:
 911  * - The old curr_active_slave has been released or lost its link.
 912  * - The primary_slave has got its link back.
 913  * - A slave has got its link back and there's no old curr_active_slave.
 914  *
 915  * Caller must hold RTNL.
 916  */
 917 void bond_select_active_slave(struct bonding *bond)
 918 {
 919         struct slave *best_slave;
 920         int rv;
 921
 922         ASSERT_RTNL();
 923
 924         best_slave = bond_find_best_slave(bond);
 925         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
 926                 bond_change_active_slave(bond, best_slave);
 927                 rv = bond_set_carrier(bond);
 928                 if (!rv)
 929                         return;
 930
 931                 if (netif_carrier_ok(bond->dev))
 932                         netdev_info(bond->dev, "first active interface up!\n");
 933                 else
 934                         netdev_info(bond->dev, "now running without any active interface!\n");
 935         }
 936 }
 937
 938 #ifdef CONFIG_NET_POLL_CONTROLLER
 939 static inline int slave_enable_netpoll(struct slave *slave)
 940 {
 941         struct netpoll *np;
 942         int err = 0;
 943
 944         np = kzalloc(sizeof(*np), GFP_KERNEL);
 945         err = -ENOMEM;
 946         if (!np)
 947                 goto out;
 948
 949         err = __netpoll_setup(np, slave->dev);
 950         if (err) {
 951                 kfree(np);
 952                 goto out;
 953         }
 954         slave->np = np;
 955 out:
 956         return err;
 957 }
 958 static inline void slave_disable_netpoll(struct slave *slave)
 959 {
 960         struct netpoll *np = slave->np;
 961
 962         if (!np)
 963                 return;
 964
 965         slave->np = NULL;
 966         __netpoll_free_async(np);
 967 }
 968
 969 static void bond_poll_controller(struct net_device *bond_dev)
 970 {
 971         struct bonding *bond = netdev_priv(bond_dev);
 972         struct slave *slave = NULL;
 973         struct list_head *iter;
 974         struct ad_info ad_info;
 975         struct netpoll_info *ni;
 976         const struct net_device_ops *ops;
 977
 978         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 979                 if (bond_3ad_get_active_agg_info(bond, &ad_info))
 980                         return;
 981
 982         bond_for_each_slave_rcu(bond, slave, iter) {
 983                 ops = slave->dev->netdev_ops;
 984                 if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller)
 985                         continue;
 986
 987                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 988                         struct aggregator *agg =
 989                             SLAVE_AD_INFO(slave)->port.aggregator;
 990
 991                         if (agg &&
 992                             agg->aggregator_identifier != ad_info.aggregator_id)
 993                                 continue;
 994                 }
 995
 996                 ni = rcu_dereference_bh(slave->dev->npinfo);
 997                 if (down_trylock(&ni->dev_lock))
 998                         continue;
 999                 ops->ndo_poll_controller(slave->dev);
1000                 up(&ni->dev_lock);
1001         }
1002 }
1003
1004 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1005 {
1006         struct bonding *bond = netdev_priv(bond_dev);
1007         struct list_head *iter;
1008         struct slave *slave;
1009
1010         bond_for_each_slave(bond, slave, iter)
1011                 if (bond_slave_is_up(slave))
1012                         slave_disable_netpoll(slave);
1013 }
1014
1015 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1016 {
1017         struct bonding *bond = netdev_priv(dev);
1018         struct list_head *iter;
1019         struct slave *slave;
1020         int err = 0;
1021
1022         bond_for_each_slave(bond, slave, iter) {
1023                 err = slave_enable_netpoll(slave);
1024                 if (err) {
1025                         bond_netpoll_cleanup(dev);
1026                         break;
1027                 }
1028         }
1029         return err;
1030 }
1031 #else
1032 static inline int slave_enable_netpoll(struct slave *slave)
1033 {
1034         return 0;
1035 }
1036 static inline void slave_disable_netpoll(struct slave *slave)
1037 {
1038 }
1039 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1040 {
1041 }
1042 #endif
1043
1044 /*---------------------------------- IOCTL ----------------------------------*/
1045
1046 static netdev_features_t bond_fix_features(struct net_device *dev,
1047                                            netdev_features_t features)
1048 {
1049         struct bonding *bond = netdev_priv(dev);
1050         struct list_head *iter;
1051         netdev_features_t mask;
1052         struct slave *slave;
1053
1054         mask = features;
1055
1056         features &= ~NETIF_F_ONE_FOR_ALL;
1057         features |= NETIF_F_ALL_FOR_ALL;
1058
1059         bond_for_each_slave(bond, slave, iter) {
1060                 features = netdev_increment_features(features,
1061                                                      slave->dev->features,
1062                                                      mask);
1063         }
1064         features = netdev_add_tso_features(features, mask);
1065
1066         return features;
1067 }
1068
1069 #define BOND_VLAN_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1070                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1071                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1072
1073 #define BOND_ENC_FEATURES       (NETIF_F_HW_CSUM | NETIF_F_SG | \
1074                                  NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1075
1076 static void bond_compute_features(struct bonding *bond)
1077 {
1078         unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1079                                         IFF_XMIT_DST_RELEASE_PERM;
1080         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1081         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1082         struct net_device *bond_dev = bond->dev;
1083         struct list_head *iter;
1084         struct slave *slave;
1085         unsigned short max_hard_header_len = ETH_HLEN;
1086         unsigned int gso_max_size = GSO_MAX_SIZE;
1087         u16 gso_max_segs = GSO_MAX_SEGS;
1088
1089         if (!bond_has_slaves(bond))
1090                 goto done;
1091         vlan_features &= NETIF_F_ALL_FOR_ALL;
1092
1093         bond_for_each_slave(bond, slave, iter) {
1094                 vlan_features = netdev_increment_features(vlan_features,
1095                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1096
1097                 enc_features = netdev_increment_features(enc_features,
1098                                                          slave->dev->hw_enc_features,
1099                                                          BOND_ENC_FEATURES);
1100                 dst_release_flag &= slave->dev->priv_flags;
1101                 if (slave->dev->hard_header_len > max_hard_header_len)
1102                         max_hard_header_len = slave->dev->hard_header_len;
1103
1104                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1105                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1106         }
1107
1108 done:
1109         bond_dev->vlan_features = vlan_features;
1110         bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1111                                     NETIF_F_HW_VLAN_CTAG_TX |
1112                                     NETIF_F_HW_VLAN_STAG_TX;
1113         bond_dev->hard_header_len = max_hard_header_len;
1114         bond_dev->gso_max_segs = gso_max_segs;
1115         netif_set_gso_max_size(bond_dev, gso_max_size);
1116
1117         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1118         if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1119             dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1120                 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1121
1122         netdev_change_features(bond_dev);
1123 }
1124
1125 static void bond_setup_by_slave(struct net_device *bond_dev,
1126                                 struct net_device *slave_dev)
1127 {
1128         bond_dev->header_ops        = slave_dev->header_ops;
1129
1130         bond_dev->type              = slave_dev->type;
1131         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1132         bond_dev->addr_len          = slave_dev->addr_len;
1133
1134         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1135                 slave_dev->addr_len);
1136 }
1137
1138 /* On bonding slaves other than the currently active slave, suppress
1139  * duplicates except for alb non-mcast/bcast.
1140  */
1141 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1142                                             struct slave *slave,
1143                                             struct bonding *bond)
1144 {
1145         if (bond_is_slave_inactive(slave)) {
1146                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1147                     skb->pkt_type != PACKET_BROADCAST &&
1148                     skb->pkt_type != PACKET_MULTICAST)
1149                         return false;
1150                 return true;
1151         }
1152         return false;
1153 }
1154
1155 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1156 {
1157         struct sk_buff *skb = *pskb;
1158         struct slave *slave;
1159         struct bonding *bond;
1160         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1161                           struct slave *);
1162         int ret = RX_HANDLER_ANOTHER;
1163
1164         skb = skb_share_check(skb, GFP_ATOMIC);
1165         if (unlikely(!skb))
1166                 return RX_HANDLER_CONSUMED;
1167
1168         *pskb = skb;
1169
1170         slave = bond_slave_get_rcu(skb->dev);
1171         bond = slave->bond;
1172
1173         recv_probe = ACCESS_ONCE(bond->recv_probe);
1174         if (recv_probe) {
1175                 ret = recv_probe(skb, bond, slave);
1176                 if (ret == RX_HANDLER_CONSUMED) {
1177                         consume_skb(skb);
1178                         return ret;
1179                 }
1180         }
1181
1182         if (bond_should_deliver_exact_match(skb, slave, bond))
1183                 return RX_HANDLER_EXACT;
1184
1185         skb->dev = bond->dev;
1186
1187         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1188             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1189             skb->pkt_type == PACKET_HOST) {
1190
1191                 if (unlikely(skb_cow_head(skb,
1192                                           skb->data - skb_mac_header(skb)))) {
1193                         kfree_skb(skb);
1194                         return RX_HANDLER_CONSUMED;
1195                 }
1196                 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1197         }
1198
1199         return ret;
1200 }
1201
1202 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1203 {
1204         switch (BOND_MODE(bond)) {
1205         case BOND_MODE_ROUNDROBIN:
1206                 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1207         case BOND_MODE_ACTIVEBACKUP:
1208                 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1209         case BOND_MODE_BROADCAST:
1210                 return NETDEV_LAG_TX_TYPE_BROADCAST;
1211         case BOND_MODE_XOR:
1212         case BOND_MODE_8023AD:
1213                 return NETDEV_LAG_TX_TYPE_HASH;
1214         default:
1215                 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1216         }
1217 }
1218
1219 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave)
1220 {
1221         struct netdev_lag_upper_info lag_upper_info;
1222         int err;
1223
1224         lag_upper_info.tx_type = bond_lag_tx_type(bond);
1225         err = netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1226                                            &lag_upper_info);
1227         if (err)
1228                 return err;
1229         rtmsg_ifinfo(RTM_NEWLINK, slave->dev, IFF_SLAVE, GFP_KERNEL);
1230         return 0;
1231 }
1232
1233 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1234 {
1235         netdev_upper_dev_unlink(slave->dev, bond->dev);
1236         slave->dev->flags &= ~IFF_SLAVE;
1237         rtmsg_ifinfo(RTM_NEWLINK, slave->dev, IFF_SLAVE, GFP_KERNEL);
1238 }
1239
1240 static struct slave *bond_alloc_slave(struct bonding *bond)
1241 {
1242         struct slave *slave = NULL;
1243
1244         slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1245         if (!slave)
1246                 return NULL;
1247
1248         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1249                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1250                                                GFP_KERNEL);
1251                 if (!SLAVE_AD_INFO(slave)) {
1252                         kfree(slave);
1253                         return NULL;
1254                 }
1255         }
1256         INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1257
1258         return slave;
1259 }
1260
1261 static void bond_free_slave(struct slave *slave)
1262 {
1263         struct bonding *bond = bond_get_bond_by_slave(slave);
1264
1265         cancel_delayed_work_sync(&slave->notify_work);
1266         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1267                 kfree(SLAVE_AD_INFO(slave));
1268
1269         kfree(slave);
1270 }
1271
1272 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1273 {
1274         info->bond_mode = BOND_MODE(bond);
1275         info->miimon = bond->params.miimon;
1276         info->num_slaves = bond->slave_cnt;
1277 }
1278
1279 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1280 {
1281         strcpy(info->slave_name, slave->dev->name);
1282         info->link = slave->link;
1283         info->state = bond_slave_state(slave);
1284         info->link_failure_count = slave->link_failure_count;
1285 }
1286
1287 static void bond_netdev_notify_work(struct work_struct *_work)
1288 {
1289         struct slave *slave = container_of(_work, struct slave,
1290                                            notify_work.work);
1291
1292         if (rtnl_trylock()) {
1293                 struct netdev_bonding_info binfo;
1294
1295                 bond_fill_ifslave(slave, &binfo.slave);
1296                 bond_fill_ifbond(slave->bond, &binfo.master);
1297                 netdev_bonding_info_change(slave->dev, &binfo);
1298                 rtnl_unlock();
1299         } else {
1300                 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1301         }
1302 }
1303
1304 void bond_queue_slave_event(struct slave *slave)
1305 {
1306         queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1307 }
1308
1309 void bond_lower_state_changed(struct slave *slave)
1310 {
1311         struct netdev_lag_lower_state_info info;
1312
1313         info.link_up = slave->link == BOND_LINK_UP ||
1314                        slave->link == BOND_LINK_FAIL;
1315         info.tx_enabled = bond_is_active_slave(slave);
1316         netdev_lower_state_changed(slave->dev, &info);
1317 }
1318
1319 /* enslave device <slave> to bond device <master> */
1320 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1321 {
1322         struct bonding *bond = netdev_priv(bond_dev);
1323         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1324         struct slave *new_slave = NULL, *prev_slave;
1325         struct sockaddr addr;
1326         int link_reporting;
1327         int res = 0, i;
1328
1329         if (!bond->params.use_carrier &&
1330             slave_dev->ethtool_ops->get_link == NULL &&
1331             slave_ops->ndo_do_ioctl == NULL) {
1332                 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1333                             slave_dev->name);
1334         }
1335
1336         /* already in-use? */
1337         if (netdev_is_rx_handler_busy(slave_dev)) {
1338                 netdev_err(bond_dev,
1339                            "Error: Device is in use and cannot be enslaved\n");
1340                 return -EBUSY;
1341         }
1342
1343         if (bond_dev == slave_dev) {
1344                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1345                 return -EPERM;
1346         }
1347
1348         /* vlan challenged mutual exclusion */
1349         /* no need to lock since we're protected by rtnl_lock */
1350         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1351                 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1352                            slave_dev->name);
1353                 if (vlan_uses_dev(bond_dev)) {
1354                         netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1355                                    slave_dev->name, bond_dev->name);
1356                         return -EPERM;
1357                 } else {
1358                         netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1359                                     slave_dev->name, slave_dev->name,
1360                                     bond_dev->name);
1361                 }
1362         } else {
1363                 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1364                            slave_dev->name);
1365         }
1366
1367         /* Old ifenslave binaries are no longer supported.  These can
1368          * be identified with moderate accuracy by the state of the slave:
1369          * the current ifenslave will set the interface down prior to
1370          * enslaving it; the old ifenslave will not.
1371          */
1372         if (slave_dev->flags & IFF_UP) {
1373                 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1374                            slave_dev->name);
1375                 return -EPERM;
1376         }
1377
1378         /* set bonding device ether type by slave - bonding netdevices are
1379          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1380          * there is a need to override some of the type dependent attribs/funcs.
1381          *
1382          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1383          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1384          */
1385         if (!bond_has_slaves(bond)) {
1386                 if (bond_dev->type != slave_dev->type) {
1387                         netdev_dbg(bond_dev, "change device type from %d to %d\n",
1388                                    bond_dev->type, slave_dev->type);
1389
1390                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1391                                                        bond_dev);
1392                         res = notifier_to_errno(res);
1393                         if (res) {
1394                                 netdev_err(bond_dev, "refused to change device type\n");
1395                                 return -EBUSY;
1396                         }
1397
1398                         /* Flush unicast and multicast addresses */
1399                         dev_uc_flush(bond_dev);
1400                         dev_mc_flush(bond_dev);
1401
1402                         if (slave_dev->type != ARPHRD_ETHER)
1403                                 bond_setup_by_slave(bond_dev, slave_dev);
1404                         else {
1405                                 ether_setup(bond_dev);
1406                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1407                         }
1408
1409                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1410                                                  bond_dev);
1411                 }
1412         } else if (bond_dev->type != slave_dev->type) {
1413                 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1414                            slave_dev->name, slave_dev->type, bond_dev->type);
1415                 return -EINVAL;
1416         }
1417
1418         if (slave_dev->type == ARPHRD_INFINIBAND &&
1419             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1420                 netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n",
1421                             slave_dev->type);
1422                 res = -EOPNOTSUPP;
1423                 goto err_undo_flags;
1424         }
1425
1426         if (!slave_ops->ndo_set_mac_address ||
1427             slave_dev->type == ARPHRD_INFINIBAND) {
1428                 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1429                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1430                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1431                         if (!bond_has_slaves(bond)) {
1432                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1433                                 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1434                         } else {
1435                                 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");
1436                                 res = -EOPNOTSUPP;
1437                                 goto err_undo_flags;
1438                         }
1439                 }
1440         }
1441
1442         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1443
1444         /* If this is the first slave, then we need to set the master's hardware
1445          * address to be the same as the slave's.
1446          */
1447         if (!bond_has_slaves(bond) &&
1448             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1449                 bond_set_dev_addr(bond->dev, slave_dev);
1450
1451         new_slave = bond_alloc_slave(bond);
1452         if (!new_slave) {
1453                 res = -ENOMEM;
1454                 goto err_undo_flags;
1455         }
1456
1457         new_slave->bond = bond;
1458         new_slave->dev = slave_dev;
1459         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1460          * is set via sysfs or module option if desired.
1461          */
1462         new_slave->queue_id = 0;
1463
1464         /* Save slave's original mtu and then set it to match the bond */
1465         new_slave->original_mtu = slave_dev->mtu;
1466         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1467         if (res) {
1468                 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1469                 goto err_free;
1470         }
1471
1472         /* Save slave's original ("permanent") mac address for modes
1473          * that need it, and for restoring it upon release, and then
1474          * set it to the master's address
1475          */
1476         ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1477
1478         if (!bond->params.fail_over_mac ||
1479             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1480                 /* Set slave to master's mac address.  The application already
1481                  * set the master's mac address to that of the first slave
1482                  */
1483                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1484                 addr.sa_family = slave_dev->type;
1485                 res = dev_set_mac_address(slave_dev, &addr);
1486                 if (res) {
1487                         netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1488                         goto err_restore_mtu;
1489                 }
1490         }
1491
1492         /* set slave flag before open to prevent IPv6 addrconf */
1493         slave_dev->flags |= IFF_SLAVE;
1494
1495         /* open the slave since the application closed it */
1496         res = dev_open(slave_dev);
1497         if (res) {
1498                 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1499                 goto err_restore_mac;
1500         }
1501
1502         slave_dev->priv_flags |= IFF_BONDING;
1503         /* initialize slave stats */
1504         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1505
1506         if (bond_is_lb(bond)) {
1507                 /* bond_alb_init_slave() must be called before all other stages since
1508                  * it might fail and we do not want to have to undo everything
1509                  */
1510                 res = bond_alb_init_slave(bond, new_slave);
1511                 if (res)
1512                         goto err_close;
1513         }
1514
1515         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1516         if (res) {
1517                 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1518                            slave_dev->name);
1519                 goto err_close;
1520         }
1521
1522         prev_slave = bond_last_slave(bond);
1523
1524         new_slave->delay = 0;
1525         new_slave->link_failure_count = 0;
1526
1527         if (bond_update_speed_duplex(new_slave) &&
1528             bond_needs_speed_duplex(bond))
1529                 new_slave->link = BOND_LINK_DOWN;
1530
1531         new_slave->last_rx = jiffies -
1532                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1533         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1534                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1535
1536         if (bond->params.miimon && !bond->params.use_carrier) {
1537                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1538
1539                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1540                         /* miimon is set but a bonded network driver
1541                          * does not support ETHTOOL/MII and
1542                          * arp_interval is not set.  Note: if
1543                          * use_carrier is enabled, we will never go
1544                          * here (because netif_carrier is always
1545                          * supported); thus, we don't need to change
1546                          * the messages for netif_carrier.
1547                          */
1548                         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",
1549                                     slave_dev->name);
1550                 } else if (link_reporting == -1) {
1551                         /* unable get link status using mii/ethtool */
1552                         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",
1553                                     slave_dev->name);
1554                 }
1555         }
1556
1557         /* check for initial state */
1558         new_slave->link = BOND_LINK_NOCHANGE;
1559         if (bond->params.miimon) {
1560                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1561                         if (bond->params.updelay) {
1562                                 bond_set_slave_link_state(new_slave,
1563                                                           BOND_LINK_BACK,
1564                                                           BOND_SLAVE_NOTIFY_NOW);
1565                                 new_slave->delay = bond->params.updelay;
1566                         } else {
1567                                 bond_set_slave_link_state(new_slave,
1568                                                           BOND_LINK_UP,
1569                                                           BOND_SLAVE_NOTIFY_NOW);
1570                         }
1571                 } else {
1572                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1573                                                   BOND_SLAVE_NOTIFY_NOW);
1574                 }
1575         } else if (bond->params.arp_interval) {
1576                 bond_set_slave_link_state(new_slave,
1577                                           (netif_carrier_ok(slave_dev) ?
1578                                           BOND_LINK_UP : BOND_LINK_DOWN),
1579                                           BOND_SLAVE_NOTIFY_NOW);
1580         } else {
1581                 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1582                                           BOND_SLAVE_NOTIFY_NOW);
1583         }
1584
1585         if (new_slave->link != BOND_LINK_DOWN)
1586                 new_slave->last_link_up = jiffies;
1587         netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1588                    new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1589                    (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1590
1591         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1592                 /* if there is a primary slave, remember it */
1593                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1594                         rcu_assign_pointer(bond->primary_slave, new_slave);
1595                         bond->force_primary = true;
1596                 }
1597         }
1598
1599         switch (BOND_MODE(bond)) {
1600         case BOND_MODE_ACTIVEBACKUP:
1601                 bond_set_slave_inactive_flags(new_slave,
1602                                               BOND_SLAVE_NOTIFY_NOW);
1603                 break;
1604         case BOND_MODE_8023AD:
1605                 /* in 802.3ad mode, the internal mechanism
1606                  * will activate the slaves in the selected
1607                  * aggregator
1608                  */
1609                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1610                 /* if this is the first slave */
1611                 if (!prev_slave) {
1612                         SLAVE_AD_INFO(new_slave)->id = 1;
1613                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1614                          * can be called only after the mac address of the bond is set
1615                          */
1616                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1617                 } else {
1618                         SLAVE_AD_INFO(new_slave)->id =
1619                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1620                 }
1621
1622                 bond_3ad_bind_slave(new_slave);
1623                 break;
1624         case BOND_MODE_TLB:
1625         case BOND_MODE_ALB:
1626                 bond_set_active_slave(new_slave);
1627                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1628                 break;
1629         default:
1630                 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1631
1632                 /* always active in trunk mode */
1633                 bond_set_active_slave(new_slave);
1634
1635                 /* In trunking mode there is little meaning to curr_active_slave
1636                  * anyway (it holds no special properties of the bond device),
1637                  * so we can change it without calling change_active_interface()
1638                  */
1639                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1640                     new_slave->link == BOND_LINK_UP)
1641                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1642
1643                 break;
1644         } /* switch(bond_mode) */
1645
1646 #ifdef CONFIG_NET_POLL_CONTROLLER
1647         if (bond->dev->npinfo) {
1648                 if (slave_enable_netpoll(new_slave)) {
1649                         netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1650                         res = -EBUSY;
1651                         goto err_detach;
1652                 }
1653         }
1654 #endif
1655
1656         if (!(bond_dev->features & NETIF_F_LRO))
1657                 dev_disable_lro(slave_dev);
1658
1659         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1660                                          new_slave);
1661         if (res) {
1662                 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1663                 goto err_detach;
1664         }
1665
1666         res = bond_master_upper_dev_link(bond, new_slave);
1667         if (res) {
1668                 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1669                 goto err_unregister;
1670         }
1671
1672         res = bond_sysfs_slave_add(new_slave);
1673         if (res) {
1674                 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1675                 goto err_upper_unlink;
1676         }
1677
1678         bond->nest_level = dev_get_nest_level(bond_dev) + 1;
1679
1680         /* If the mode uses primary, then the following is handled by
1681          * bond_change_active_slave().
1682          */
1683         if (!bond_uses_primary(bond)) {
1684                 /* set promiscuity level to new slave */
1685                 if (bond_dev->flags & IFF_PROMISC) {
1686                         res = dev_set_promiscuity(slave_dev, 1);
1687                         if (res)
1688                                 goto err_sysfs_del;
1689                 }
1690
1691                 /* set allmulti level to new slave */
1692                 if (bond_dev->flags & IFF_ALLMULTI) {
1693                         res = dev_set_allmulti(slave_dev, 1);
1694                         if (res) {
1695                                 if (bond_dev->flags & IFF_PROMISC)
1696                                         dev_set_promiscuity(slave_dev, -1);
1697                                 goto err_sysfs_del;
1698                         }
1699                 }
1700
1701                 netif_addr_lock_bh(bond_dev);
1702                 dev_mc_sync_multiple(slave_dev, bond_dev);
1703                 dev_uc_sync_multiple(slave_dev, bond_dev);
1704                 netif_addr_unlock_bh(bond_dev);
1705
1706                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1707                         /* add lacpdu mc addr to mc list */
1708                         u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1709
1710                         dev_mc_add(slave_dev, lacpdu_multicast);
1711                 }
1712         }
1713
1714         bond->slave_cnt++;
1715         bond_compute_features(bond);
1716         bond_set_carrier(bond);
1717
1718         if (bond_uses_primary(bond)) {
1719                 block_netpoll_tx();
1720                 bond_select_active_slave(bond);
1721                 unblock_netpoll_tx();
1722         }
1723
1724         if (bond_mode_uses_xmit_hash(bond))
1725                 bond_update_slave_arr(bond, NULL);
1726
1727
1728         netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1729                     slave_dev->name,
1730                     bond_is_active_slave(new_slave) ? "an active" : "a backup",
1731                     new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1732
1733         /* enslave is successful */
1734         bond_queue_slave_event(new_slave);
1735         return 0;
1736
1737 /* Undo stages on error */
1738 err_sysfs_del:
1739         bond_sysfs_slave_del(new_slave);
1740
1741 err_upper_unlink:
1742         bond_upper_dev_unlink(bond, new_slave);
1743
1744 err_unregister:
1745         netdev_rx_handler_unregister(slave_dev);
1746
1747 err_detach:
1748         vlan_vids_del_by_dev(slave_dev, bond_dev);
1749         if (rcu_access_pointer(bond->primary_slave) == new_slave)
1750                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1751         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1752                 block_netpoll_tx();
1753                 bond_change_active_slave(bond, NULL);
1754                 bond_select_active_slave(bond);
1755                 unblock_netpoll_tx();
1756         }
1757         /* either primary_slave or curr_active_slave might've changed */
1758         synchronize_rcu();
1759         slave_disable_netpoll(new_slave);
1760
1761 err_close:
1762         slave_dev->priv_flags &= ~IFF_BONDING;
1763         dev_close(slave_dev);
1764
1765 err_restore_mac:
1766         slave_dev->flags &= ~IFF_SLAVE;
1767         if (!bond->params.fail_over_mac ||
1768             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1769                 /* XXX TODO - fom follow mode needs to change master's
1770                  * MAC if this slave's MAC is in use by the bond, or at
1771                  * least print a warning.
1772                  */
1773                 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1774                 addr.sa_family = slave_dev->type;
1775                 dev_set_mac_address(slave_dev, &addr);
1776         }
1777
1778 err_restore_mtu:
1779         dev_set_mtu(slave_dev, new_slave->original_mtu);
1780
1781 err_free:
1782         bond_free_slave(new_slave);
1783
1784 err_undo_flags:
1785         /* Enslave of first slave has failed and we need to fix master's mac */
1786         if (!bond_has_slaves(bond)) {
1787                 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1788                                             slave_dev->dev_addr))
1789                         eth_hw_addr_random(bond_dev);
1790                 if (bond_dev->type != ARPHRD_ETHER) {
1791                         dev_close(bond_dev);
1792                         ether_setup(bond_dev);
1793                         bond_dev->flags |= IFF_MASTER;
1794                         bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1795                 }
1796         }
1797
1798         return res;
1799 }
1800
1801 /* Try to release the slave device <slave> from the bond device <master>
1802  * It is legal to access curr_active_slave without a lock because all the function
1803  * is RTNL-locked. If "all" is true it means that the function is being called
1804  * while destroying a bond interface and all slaves are being released.
1805  *
1806  * The rules for slave state should be:
1807  *   for Active/Backup:
1808  *     Active stays on all backups go down
1809  *   for Bonded connections:
1810  *     The first up interface should be left on and all others downed.
1811  */
1812 static int __bond_release_one(struct net_device *bond_dev,
1813                               struct net_device *slave_dev,
1814                               bool all)
1815 {
1816         struct bonding *bond = netdev_priv(bond_dev);
1817         struct slave *slave, *oldcurrent;
1818         struct sockaddr addr;
1819         int old_flags = bond_dev->flags;
1820         netdev_features_t old_features = bond_dev->features;
1821
1822         /* slave is not a slave or master is not master of this slave */
1823         if (!(slave_dev->flags & IFF_SLAVE) ||
1824             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1825                 netdev_dbg(bond_dev, "cannot release %s\n",
1826                            slave_dev->name);
1827                 return -EINVAL;
1828         }
1829
1830         block_netpoll_tx();
1831
1832         slave = bond_get_slave_by_dev(bond, slave_dev);
1833         if (!slave) {
1834                 /* not a slave of this bond */
1835                 netdev_info(bond_dev, "%s not enslaved\n",
1836                             slave_dev->name);
1837                 unblock_netpoll_tx();
1838                 return -EINVAL;
1839         }
1840
1841         bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1842
1843         bond_sysfs_slave_del(slave);
1844
1845         /* recompute stats just before removing the slave */
1846         bond_get_stats(bond->dev, &bond->bond_stats);
1847
1848         bond_upper_dev_unlink(bond, slave);
1849         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1850          * for this slave anymore.
1851          */
1852         netdev_rx_handler_unregister(slave_dev);
1853
1854         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1855                 bond_3ad_unbind_slave(slave);
1856
1857         if (bond_mode_uses_xmit_hash(bond))
1858                 bond_update_slave_arr(bond, slave);
1859
1860         netdev_info(bond_dev, "Releasing %s interface %s\n",
1861                     bond_is_active_slave(slave) ? "active" : "backup",
1862                     slave_dev->name);
1863
1864         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1865
1866         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1867
1868         if (!all && (!bond->params.fail_over_mac ||
1869                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1870                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1871                     bond_has_slaves(bond))
1872                         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",
1873                                     slave_dev->name, slave->perm_hwaddr,
1874                                     bond_dev->name, slave_dev->name);
1875         }
1876
1877         if (rtnl_dereference(bond->primary_slave) == slave)
1878                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1879
1880         if (oldcurrent == slave)
1881                 bond_change_active_slave(bond, NULL);
1882
1883         if (bond_is_lb(bond)) {
1884                 /* Must be called only after the slave has been
1885                  * detached from the list and the curr_active_slave
1886                  * has been cleared (if our_slave == old_current),
1887                  * but before a new active slave is selected.
1888                  */
1889                 bond_alb_deinit_slave(bond, slave);
1890         }
1891
1892         if (all) {
1893                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1894         } else if (oldcurrent == slave) {
1895                 /* Note that we hold RTNL over this sequence, so there
1896                  * is no concern that another slave add/remove event
1897                  * will interfere.
1898                  */
1899                 bond_select_active_slave(bond);
1900         }
1901
1902         if (!bond_has_slaves(bond)) {
1903                 bond_set_carrier(bond);
1904                 eth_hw_addr_random(bond_dev);
1905                 bond->nest_level = SINGLE_DEPTH_NESTING;
1906         } else {
1907                 bond->nest_level = dev_get_nest_level(bond_dev) + 1;
1908         }
1909
1910         unblock_netpoll_tx();
1911         synchronize_rcu();
1912         bond->slave_cnt--;
1913
1914         if (!bond_has_slaves(bond)) {
1915                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1916                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1917         }
1918
1919         bond_compute_features(bond);
1920         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1921             (old_features & NETIF_F_VLAN_CHALLENGED))
1922                 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1923                             slave_dev->name, bond_dev->name);
1924
1925         vlan_vids_del_by_dev(slave_dev, bond_dev);
1926
1927         /* If the mode uses primary, then this case was handled above by
1928          * bond_change_active_slave(..., NULL)
1929          */
1930         if (!bond_uses_primary(bond)) {
1931                 /* unset promiscuity level from slave
1932                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1933                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1934                  * value of that flag before that change, as that was the value
1935                  * when this slave was attached, so we cache at the start of the
1936                  * function and use it here. Same goes for ALLMULTI below
1937                  */
1938                 if (old_flags & IFF_PROMISC)
1939                         dev_set_promiscuity(slave_dev, -1);
1940
1941                 /* unset allmulti level from slave */
1942                 if (old_flags & IFF_ALLMULTI)
1943                         dev_set_allmulti(slave_dev, -1);
1944
1945                 bond_hw_addr_flush(bond_dev, slave_dev);
1946         }
1947
1948         slave_disable_netpoll(slave);
1949
1950         /* close slave before restoring its mac address */
1951         dev_close(slave_dev);
1952
1953         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1954             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1955                 /* restore original ("permanent") mac address */
1956                 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1957                 addr.sa_family = slave_dev->type;
1958                 dev_set_mac_address(slave_dev, &addr);
1959         }
1960
1961         dev_set_mtu(slave_dev, slave->original_mtu);
1962
1963         slave_dev->priv_flags &= ~IFF_BONDING;
1964
1965         bond_free_slave(slave);
1966
1967         return 0;
1968 }
1969
1970 /* A wrapper used because of ndo_del_link */
1971 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1972 {
1973         return __bond_release_one(bond_dev, slave_dev, false);
1974 }
1975
1976 /* First release a slave and then destroy the bond if no more slaves are left.
1977  * Must be under rtnl_lock when this function is called.
1978  */
1979 static int  bond_release_and_destroy(struct net_device *bond_dev,
1980                                      struct net_device *slave_dev)
1981 {
1982         struct bonding *bond = netdev_priv(bond_dev);
1983         int ret;
1984
1985         ret = bond_release(bond_dev, slave_dev);
1986         if (ret == 0 && !bond_has_slaves(bond)) {
1987                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1988                 netdev_info(bond_dev, "Destroying bond %s\n",
1989                             bond_dev->name);
1990                 bond_remove_proc_entry(bond);
1991                 unregister_netdevice(bond_dev);
1992         }
1993         return ret;
1994 }
1995
1996 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1997 {
1998         struct bonding *bond = netdev_priv(bond_dev);
1999         bond_fill_ifbond(bond, info);
2000         return 0;
2001 }
2002
2003 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2004 {
2005         struct bonding *bond = netdev_priv(bond_dev);
2006         struct list_head *iter;
2007         int i = 0, res = -ENODEV;
2008         struct slave *slave;
2009
2010         bond_for_each_slave(bond, slave, iter) {
2011                 if (i++ == (int)info->slave_id) {
2012                         res = 0;
2013                         bond_fill_ifslave(slave, info);
2014                         break;
2015                 }
2016         }
2017
2018         return res;
2019 }
2020
2021 /*-------------------------------- Monitoring -------------------------------*/
2022
2023 /* called with rcu_read_lock() */
2024 static int bond_miimon_inspect(struct bonding *bond)
2025 {
2026         int link_state, commit = 0;
2027         struct list_head *iter;
2028         struct slave *slave;
2029         bool ignore_updelay;
2030
2031         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2032
2033         bond_for_each_slave_rcu(bond, slave, iter) {
2034                 slave->new_link = BOND_LINK_NOCHANGE;
2035
2036                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2037
2038                 switch (slave->link) {
2039                 case BOND_LINK_UP:
2040                         if (link_state)
2041                                 continue;
2042
2043                         bond_set_slave_link_state(slave, BOND_LINK_FAIL,
2044                                                   BOND_SLAVE_NOTIFY_LATER);
2045                         slave->delay = bond->params.downdelay;
2046                         if (slave->delay) {
2047                                 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2048                                             (BOND_MODE(bond) ==
2049                                              BOND_MODE_ACTIVEBACKUP) ?
2050                                              (bond_is_active_slave(slave) ?
2051                                               "active " : "backup ") : "",
2052                                             slave->dev->name,
2053                                             bond->params.downdelay * bond->params.miimon);
2054                         }
2055                         /*FALLTHRU*/
2056                 case BOND_LINK_FAIL:
2057                         if (link_state) {
2058                                 /* recovered before downdelay expired */
2059                                 bond_set_slave_link_state(slave, BOND_LINK_UP,
2060                                                           BOND_SLAVE_NOTIFY_LATER);
2061                                 slave->last_link_up = jiffies;
2062                                 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2063                                             (bond->params.downdelay - slave->delay) *
2064                                             bond->params.miimon,
2065                                             slave->dev->name);
2066                                 commit++;
2067                                 continue;
2068                         }
2069
2070                         if (slave->delay <= 0) {
2071                                 slave->new_link = BOND_LINK_DOWN;
2072                                 commit++;
2073                                 continue;
2074                         }
2075
2076                         slave->delay--;
2077                         break;
2078
2079                 case BOND_LINK_DOWN:
2080                         if (!link_state)
2081                                 continue;
2082
2083                         bond_set_slave_link_state(slave, BOND_LINK_BACK,
2084                                                   BOND_SLAVE_NOTIFY_LATER);
2085                         slave->delay = bond->params.updelay;
2086
2087                         if (slave->delay) {
2088                                 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2089                                             slave->dev->name,
2090                                             ignore_updelay ? 0 :
2091                                             bond->params.updelay *
2092                                             bond->params.miimon);
2093                         }
2094                         /*FALLTHRU*/
2095                 case BOND_LINK_BACK:
2096                         if (!link_state) {
2097                                 bond_set_slave_link_state(slave,
2098                                                           BOND_LINK_DOWN,
2099                                                           BOND_SLAVE_NOTIFY_LATER);
2100                                 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2101                                             (bond->params.updelay - slave->delay) *
2102                                             bond->params.miimon,
2103                                             slave->dev->name);
2104                                 commit++;
2105                                 continue;
2106                         }
2107
2108                         if (ignore_updelay)
2109                                 slave->delay = 0;
2110
2111                         if (slave->delay <= 0) {
2112                                 slave->new_link = BOND_LINK_UP;
2113                                 commit++;
2114                                 ignore_updelay = false;
2115                                 continue;
2116                         }
2117
2118                         slave->delay--;
2119                         break;
2120                 }
2121         }
2122
2123         return commit;
2124 }
2125
2126 static void bond_miimon_commit(struct bonding *bond)
2127 {
2128         struct list_head *iter;
2129         struct slave *slave, *primary;
2130
2131         bond_for_each_slave(bond, slave, iter) {
2132                 switch (slave->new_link) {
2133                 case BOND_LINK_NOCHANGE:
2134                         continue;
2135
2136                 case BOND_LINK_UP:
2137                         if (bond_update_speed_duplex(slave) &&
2138                             bond_needs_speed_duplex(bond)) {
2139                                 slave->link = BOND_LINK_DOWN;
2140                                 if (net_ratelimit())
2141                                         netdev_warn(bond->dev,
2142                                                     "failed to get link speed/duplex for %s\n",
2143                                                     slave->dev->name);
2144                                 continue;
2145                         }
2146                         bond_set_slave_link_state(slave, BOND_LINK_UP,
2147                                                   BOND_SLAVE_NOTIFY_NOW);
2148                         slave->last_link_up = jiffies;
2149
2150                         primary = rtnl_dereference(bond->primary_slave);
2151                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2152                                 /* prevent it from being the active one */
2153                                 bond_set_backup_slave(slave);
2154                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2155                                 /* make it immediately active */
2156                                 bond_set_active_slave(slave);
2157                         } else if (slave != primary) {
2158                                 /* prevent it from being the active one */
2159                                 bond_set_backup_slave(slave);
2160                         }
2161
2162                         netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2163                                     slave->dev->name,
2164                                     slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2165                                     slave->duplex ? "full" : "half");
2166
2167                         /* notify ad that the link status has changed */
2168                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2169                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2170
2171                         if (bond_is_lb(bond))
2172                                 bond_alb_handle_link_change(bond, slave,
2173                                                             BOND_LINK_UP);
2174
2175                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2176                                 bond_update_slave_arr(bond, NULL);
2177
2178                         if (!bond->curr_active_slave || slave == primary)
2179                                 goto do_failover;
2180
2181                         continue;
2182
2183                 case BOND_LINK_DOWN:
2184                         if (slave->link_failure_count < UINT_MAX)
2185                                 slave->link_failure_count++;
2186
2187                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2188                                                   BOND_SLAVE_NOTIFY_NOW);
2189
2190                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2191                             BOND_MODE(bond) == BOND_MODE_8023AD)
2192                                 bond_set_slave_inactive_flags(slave,
2193                                                               BOND_SLAVE_NOTIFY_NOW);
2194
2195                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2196                                     slave->dev->name);
2197
2198                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2199                                 bond_3ad_handle_link_change(slave,
2200                                                             BOND_LINK_DOWN);
2201
2202                         if (bond_is_lb(bond))
2203                                 bond_alb_handle_link_change(bond, slave,
2204                                                             BOND_LINK_DOWN);
2205
2206                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2207                                 bond_update_slave_arr(bond, NULL);
2208
2209                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2210                                 goto do_failover;
2211
2212                         continue;
2213
2214                 default:
2215                         netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2216                                    slave->new_link, slave->dev->name);
2217                         slave->new_link = BOND_LINK_NOCHANGE;
2218
2219                         continue;
2220                 }
2221
2222 do_failover:
2223                 block_netpoll_tx();
2224                 bond_select_active_slave(bond);
2225                 unblock_netpoll_tx();
2226         }
2227
2228         bond_set_carrier(bond);
2229 }
2230
2231 /* bond_mii_monitor
2232  *
2233  * Really a wrapper that splits the mii monitor into two phases: an
2234  * inspection, then (if inspection indicates something needs to be done)
2235  * an acquisition of appropriate locks followed by a commit phase to
2236  * implement whatever link state changes are indicated.
2237  */
2238 static void bond_mii_monitor(struct work_struct *work)
2239 {
2240         struct bonding *bond = container_of(work, struct bonding,
2241                                             mii_work.work);
2242         bool should_notify_peers = false;
2243         unsigned long delay;
2244
2245         delay = msecs_to_jiffies(bond->params.miimon);
2246
2247         if (!bond_has_slaves(bond))
2248                 goto re_arm;
2249
2250         rcu_read_lock();
2251
2252         should_notify_peers = bond_should_notify_peers(bond);
2253
2254         if (bond_miimon_inspect(bond)) {
2255                 rcu_read_unlock();
2256
2257                 /* Race avoidance with bond_close cancel of workqueue */
2258                 if (!rtnl_trylock()) {
2259                         delay = 1;
2260                         should_notify_peers = false;
2261                         goto re_arm;
2262                 }
2263
2264                 bond_miimon_commit(bond);
2265
2266                 rtnl_unlock();  /* might sleep, hold no other locks */
2267         } else
2268                 rcu_read_unlock();
2269
2270 re_arm:
2271         if (bond->params.miimon)
2272                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2273
2274         if (should_notify_peers) {
2275                 if (!rtnl_trylock())
2276                         return;
2277                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2278                 rtnl_unlock();
2279         }
2280 }
2281
2282 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2283 {
2284         struct net_device *upper;
2285         struct list_head *iter;
2286         bool ret = false;
2287
2288         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2289                 return true;
2290
2291         rcu_read_lock();
2292         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2293                 if (ip == bond_confirm_addr(upper, 0, ip)) {
2294                         ret = true;
2295                         break;
2296                 }
2297         }
2298         rcu_read_unlock();
2299
2300         return ret;
2301 }
2302
2303 /* We go to the (large) trouble of VLAN tagging ARP frames because
2304  * switches in VLAN mode (especially if ports are configured as
2305  * "native" to a VLAN) might not pass non-tagged frames.
2306  */
2307 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2308                           __be32 dest_ip, __be32 src_ip,
2309                           struct bond_vlan_tag *tags)
2310 {
2311         struct sk_buff *skb;
2312         struct bond_vlan_tag *outer_tag = tags;
2313
2314         netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2315                    arp_op, slave_dev->name, &dest_ip, &src_ip);
2316
2317         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2318                          NULL, slave_dev->dev_addr, NULL);
2319
2320         if (!skb) {
2321                 net_err_ratelimited("ARP packet allocation failed\n");
2322                 return;
2323         }
2324
2325         if (!tags || tags->vlan_proto == VLAN_N_VID)
2326                 goto xmit;
2327
2328         tags++;
2329
2330         /* Go through all the tags backwards and add them to the packet */
2331         while (tags->vlan_proto != VLAN_N_VID) {
2332                 if (!tags->vlan_id) {
2333                         tags++;
2334                         continue;
2335                 }
2336
2337                 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2338                            ntohs(outer_tag->vlan_proto), tags->vlan_id);
2339                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2340                                                 tags->vlan_id);
2341                 if (!skb) {
2342                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2343                         return;
2344                 }
2345
2346                 tags++;
2347         }
2348         /* Set the outer tag */
2349         if (outer_tag->vlan_id) {
2350                 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2351                            ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2352                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2353                                        outer_tag->vlan_id);
2354         }
2355
2356 xmit:
2357         arp_xmit(skb);
2358 }
2359
2360 /* Validate the device path between the @start_dev and the @end_dev.
2361  * The path is valid if the @end_dev is reachable through device
2362  * stacking.
2363  * When the path is validated, collect any vlan information in the
2364  * path.
2365  */
2366 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2367                                               struct net_device *end_dev,
2368                                               int level)
2369 {
2370         struct bond_vlan_tag *tags;
2371         struct net_device *upper;
2372         struct list_head  *iter;
2373
2374         if (start_dev == end_dev) {
2375                 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2376                 if (!tags)
2377                         return ERR_PTR(-ENOMEM);
2378                 tags[level].vlan_proto = VLAN_N_VID;
2379                 return tags;
2380         }
2381
2382         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2383                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2384                 if (IS_ERR_OR_NULL(tags)) {
2385                         if (IS_ERR(tags))
2386                                 return tags;
2387                         continue;
2388                 }
2389                 if (is_vlan_dev(upper)) {
2390                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2391                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2392                 }
2393
2394                 return tags;
2395         }
2396
2397         return NULL;
2398 }
2399
2400 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2401 {
2402         struct rtable *rt;
2403         struct bond_vlan_tag *tags;
2404         __be32 *targets = bond->params.arp_targets, addr;
2405         int i;
2406
2407         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2408                 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2409                 tags = NULL;
2410
2411                 /* Find out through which dev should the packet go */
2412                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2413                                      RTO_ONLINK, 0);
2414                 if (IS_ERR(rt)) {
2415                         /* there's no route to target - try to send arp
2416                          * probe to generate any traffic (arp_validate=0)
2417                          */
2418                         if (bond->params.arp_validate)
2419                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2420                                                      bond->dev->name,
2421                                                      &targets[i]);
2422                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2423                                       0, tags);
2424                         continue;
2425                 }
2426
2427                 /* bond device itself */
2428                 if (rt->dst.dev == bond->dev)
2429                         goto found;
2430
2431                 rcu_read_lock();
2432                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2433                 rcu_read_unlock();
2434
2435                 if (!IS_ERR_OR_NULL(tags))
2436                         goto found;
2437
2438                 /* Not our device - skip */
2439                 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2440                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2441
2442                 ip_rt_put(rt);
2443                 continue;
2444
2445 found:
2446                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2447                 ip_rt_put(rt);
2448                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2449                               addr, tags);
2450                 kfree(tags);
2451         }
2452 }
2453
2454 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2455 {
2456         int i;
2457
2458         if (!sip || !bond_has_this_ip(bond, tip)) {
2459                 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2460                            &sip, &tip);
2461                 return;
2462         }
2463
2464         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2465         if (i == -1) {
2466                 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2467                            &sip);
2468                 return;
2469         }
2470         slave->last_rx = jiffies;
2471         slave->target_last_arp_rx[i] = jiffies;
2472 }
2473
2474 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2475                  struct slave *slave)
2476 {
2477         struct arphdr *arp = (struct arphdr *)skb->data;
2478         struct slave *curr_active_slave, *curr_arp_slave;
2479         unsigned char *arp_ptr;
2480         __be32 sip, tip;
2481         int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2482
2483         if (!slave_do_arp_validate(bond, slave)) {
2484                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2485                     !slave_do_arp_validate_only(bond))
2486                         slave->last_rx = jiffies;
2487                 return RX_HANDLER_ANOTHER;
2488         } else if (!is_arp) {
2489                 return RX_HANDLER_ANOTHER;
2490         }
2491
2492         alen = arp_hdr_len(bond->dev);
2493
2494         netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2495                    skb->dev->name);
2496
2497         if (alen > skb_headlen(skb)) {
2498                 arp = kmalloc(alen, GFP_ATOMIC);
2499                 if (!arp)
2500                         goto out_unlock;
2501                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2502                         goto out_unlock;
2503         }
2504
2505         if (arp->ar_hln != bond->dev->addr_len ||
2506             skb->pkt_type == PACKET_OTHERHOST ||
2507             skb->pkt_type == PACKET_LOOPBACK ||
2508             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2509             arp->ar_pro != htons(ETH_P_IP) ||
2510             arp->ar_pln != 4)
2511                 goto out_unlock;
2512
2513         arp_ptr = (unsigned char *)(arp + 1);
2514         arp_ptr += bond->dev->addr_len;
2515         memcpy(&sip, arp_ptr, 4);
2516         arp_ptr += 4 + bond->dev->addr_len;
2517         memcpy(&tip, arp_ptr, 4);
2518
2519         netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2520                    slave->dev->name, bond_slave_state(slave),
2521                      bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2522                      &sip, &tip);
2523
2524         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2525         curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2526
2527         /* We 'trust' the received ARP enough to validate it if:
2528          *
2529          * (a) the slave receiving the ARP is active (which includes the
2530          * current ARP slave, if any), or
2531          *
2532          * (b) the receiving slave isn't active, but there is a currently
2533          * active slave and it received valid arp reply(s) after it became
2534          * the currently active slave, or
2535          *
2536          * (c) there is an ARP slave that sent an ARP during the prior ARP
2537          * interval, and we receive an ARP reply on any slave.  We accept
2538          * these because switch FDB update delays may deliver the ARP
2539          * reply to a slave other than the sender of the ARP request.
2540          *
2541          * Note: for (b), backup slaves are receiving the broadcast ARP
2542          * request, not a reply.  This request passes from the sending
2543          * slave through the L2 switch(es) to the receiving slave.  Since
2544          * this is checking the request, sip/tip are swapped for
2545          * validation.
2546          *
2547          * This is done to avoid endless looping when we can't reach the
2548          * arp_ip_target and fool ourselves with our own arp requests.
2549          */
2550         if (bond_is_active_slave(slave))
2551                 bond_validate_arp(bond, slave, sip, tip);
2552         else if (curr_active_slave &&
2553                  time_after(slave_last_rx(bond, curr_active_slave),
2554                             curr_active_slave->last_link_up))
2555                 bond_validate_arp(bond, slave, tip, sip);
2556         else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2557                  bond_time_in_interval(bond,
2558                                        dev_trans_start(curr_arp_slave->dev), 1))
2559                 bond_validate_arp(bond, slave, sip, tip);
2560
2561 out_unlock:
2562         if (arp != (struct arphdr *)skb->data)
2563                 kfree(arp);
2564         return RX_HANDLER_ANOTHER;
2565 }
2566
2567 /* function to verify if we're in the arp_interval timeslice, returns true if
2568  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2569  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2570  */
2571 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2572                                   int mod)
2573 {
2574         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2575
2576         return time_in_range(jiffies,
2577                              last_act - delta_in_ticks,
2578                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2579 }
2580
2581 /* This function is called regularly to monitor each slave's link
2582  * ensuring that traffic is being sent and received when arp monitoring
2583  * is used in load-balancing mode. if the adapter has been dormant, then an
2584  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2585  * arp monitoring in active backup mode.
2586  */
2587 static void bond_loadbalance_arp_mon(struct work_struct *work)
2588 {
2589         struct bonding *bond = container_of(work, struct bonding,
2590                                             arp_work.work);
2591         struct slave *slave, *oldcurrent;
2592         struct list_head *iter;
2593         int do_failover = 0, slave_state_changed = 0;
2594
2595         if (!bond_has_slaves(bond))
2596                 goto re_arm;
2597
2598         rcu_read_lock();
2599
2600         oldcurrent = rcu_dereference(bond->curr_active_slave);
2601         /* see if any of the previous devices are up now (i.e. they have
2602          * xmt and rcv traffic). the curr_active_slave does not come into
2603          * the picture unless it is null. also, slave->last_link_up is not
2604          * needed here because we send an arp on each slave and give a slave
2605          * as long as it needs to get the tx/rx within the delta.
2606          * TODO: what about up/down delay in arp mode? it wasn't here before
2607          *       so it can wait
2608          */
2609         bond_for_each_slave_rcu(bond, slave, iter) {
2610                 unsigned long trans_start = dev_trans_start(slave->dev);
2611
2612                 slave->new_link = BOND_LINK_NOCHANGE;
2613
2614                 if (slave->link != BOND_LINK_UP) {
2615                         if (bond_time_in_interval(bond, trans_start, 1) &&
2616                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2617
2618                                 slave->new_link = BOND_LINK_UP;
2619                                 slave_state_changed = 1;
2620
2621                                 /* primary_slave has no meaning in round-robin
2622                                  * mode. the window of a slave being up and
2623                                  * curr_active_slave being null after enslaving
2624                                  * is closed.
2625                                  */
2626                                 if (!oldcurrent) {
2627                                         netdev_info(bond->dev, "link status definitely up for interface %s\n",
2628                                                     slave->dev->name);
2629                                         do_failover = 1;
2630                                 } else {
2631                                         netdev_info(bond->dev, "interface %s is now up\n",
2632                                                     slave->dev->name);
2633                                 }
2634                         }
2635                 } else {
2636                         /* slave->link == BOND_LINK_UP */
2637
2638                         /* not all switches will respond to an arp request
2639                          * when the source ip is 0, so don't take the link down
2640                          * if we don't know our ip yet
2641                          */
2642                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2643                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2644
2645                                 slave->new_link = BOND_LINK_DOWN;
2646                                 slave_state_changed = 1;
2647
2648                                 if (slave->link_failure_count < UINT_MAX)
2649                                         slave->link_failure_count++;
2650
2651                                 netdev_info(bond->dev, "interface %s is now down\n",
2652                                             slave->dev->name);
2653
2654                                 if (slave == oldcurrent)
2655                                         do_failover = 1;
2656                         }
2657                 }
2658
2659                 /* note: if switch is in round-robin mode, all links
2660                  * must tx arp to ensure all links rx an arp - otherwise
2661                  * links may oscillate or not come up at all; if switch is
2662                  * in something like xor mode, there is nothing we can
2663                  * do - all replies will be rx'ed on same link causing slaves
2664                  * to be unstable during low/no traffic periods
2665                  */
2666                 if (bond_slave_is_up(slave))
2667                         bond_arp_send_all(bond, slave);
2668         }
2669
2670         rcu_read_unlock();
2671
2672         if (do_failover || slave_state_changed) {
2673                 if (!rtnl_trylock())
2674                         goto re_arm;
2675
2676                 bond_for_each_slave(bond, slave, iter) {
2677                         if (slave->new_link != BOND_LINK_NOCHANGE)
2678                                 slave->link = slave->new_link;
2679                 }
2680
2681                 if (slave_state_changed) {
2682                         bond_slave_state_change(bond);
2683                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2684                                 bond_update_slave_arr(bond, NULL);
2685                 }
2686                 if (do_failover) {
2687                         block_netpoll_tx();
2688                         bond_select_active_slave(bond);
2689                         unblock_netpoll_tx();
2690                 }
2691                 rtnl_unlock();
2692         }
2693
2694 re_arm:
2695         if (bond->params.arp_interval)
2696                 queue_delayed_work(bond->wq, &bond->arp_work,
2697                                    msecs_to_jiffies(bond->params.arp_interval));
2698 }
2699
2700 /* Called to inspect slaves for active-backup mode ARP monitor link state
2701  * changes.  Sets new_link in slaves to specify what action should take
2702  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2703  * to link states must be committed.
2704  *
2705  * Called with rcu_read_lock held.
2706  */
2707 static int bond_ab_arp_inspect(struct bonding *bond)
2708 {
2709         unsigned long trans_start, last_rx;
2710         struct list_head *iter;
2711         struct slave *slave;
2712         int commit = 0;
2713
2714         bond_for_each_slave_rcu(bond, slave, iter) {
2715                 slave->new_link = BOND_LINK_NOCHANGE;
2716                 last_rx = slave_last_rx(bond, slave);
2717
2718                 if (slave->link != BOND_LINK_UP) {
2719                         if (bond_time_in_interval(bond, last_rx, 1)) {
2720                                 slave->new_link = BOND_LINK_UP;
2721                                 commit++;
2722                         }
2723                         continue;
2724                 }
2725
2726                 /* Give slaves 2*delta after being enslaved or made
2727                  * active.  This avoids bouncing, as the last receive
2728                  * times need a full ARP monitor cycle to be updated.
2729                  */
2730                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2731                         continue;
2732
2733                 /* Backup slave is down if:
2734                  * - No current_arp_slave AND
2735                  * - more than 3*delta since last receive AND
2736                  * - the bond has an IP address
2737                  *
2738                  * Note: a non-null current_arp_slave indicates
2739                  * the curr_active_slave went down and we are
2740                  * searching for a new one; under this condition
2741                  * we only take the curr_active_slave down - this
2742                  * gives each slave a chance to tx/rx traffic
2743                  * before being taken out
2744                  */
2745                 if (!bond_is_active_slave(slave) &&
2746                     !rcu_access_pointer(bond->current_arp_slave) &&
2747                     !bond_time_in_interval(bond, last_rx, 3)) {
2748                         slave->new_link = BOND_LINK_DOWN;
2749                         commit++;
2750                 }
2751
2752                 /* Active slave is down if:
2753                  * - more than 2*delta since transmitting OR
2754                  * - (more than 2*delta since receive AND
2755                  *    the bond has an IP address)
2756                  */
2757                 trans_start = dev_trans_start(slave->dev);
2758                 if (bond_is_active_slave(slave) &&
2759                     (!bond_time_in_interval(bond, trans_start, 2) ||
2760                      !bond_time_in_interval(bond, last_rx, 2))) {
2761                         slave->new_link = BOND_LINK_DOWN;
2762                         commit++;
2763                 }
2764         }
2765
2766         return commit;
2767 }
2768
2769 /* Called to commit link state changes noted by inspection step of
2770  * active-backup mode ARP monitor.
2771  *
2772  * Called with RTNL hold.
2773  */
2774 static void bond_ab_arp_commit(struct bonding *bond)
2775 {
2776         unsigned long trans_start;
2777         struct list_head *iter;
2778         struct slave *slave;
2779
2780         bond_for_each_slave(bond, slave, iter) {
2781                 switch (slave->new_link) {
2782                 case BOND_LINK_NOCHANGE:
2783                         continue;
2784
2785                 case BOND_LINK_UP:
2786                         trans_start = dev_trans_start(slave->dev);
2787                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
2788                             (!rtnl_dereference(bond->curr_active_slave) &&
2789                              bond_time_in_interval(bond, trans_start, 1))) {
2790                                 struct slave *current_arp_slave;
2791
2792                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2793                                 bond_set_slave_link_state(slave, BOND_LINK_UP,
2794                                                           BOND_SLAVE_NOTIFY_NOW);
2795                                 if (current_arp_slave) {
2796                                         bond_set_slave_inactive_flags(
2797                                                 current_arp_slave,
2798                                                 BOND_SLAVE_NOTIFY_NOW);
2799                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2800                                 }
2801
2802                                 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2803                                             slave->dev->name);
2804
2805                                 if (!rtnl_dereference(bond->curr_active_slave) ||
2806                                     slave == rtnl_dereference(bond->primary_slave))
2807                                         goto do_failover;
2808
2809                         }
2810
2811                         continue;
2812
2813                 case BOND_LINK_DOWN:
2814                         if (slave->link_failure_count < UINT_MAX)
2815                                 slave->link_failure_count++;
2816
2817                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2818                                                   BOND_SLAVE_NOTIFY_NOW);
2819                         bond_set_slave_inactive_flags(slave,
2820                                                       BOND_SLAVE_NOTIFY_NOW);
2821
2822                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2823                                     slave->dev->name);
2824
2825                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
2826                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2827                                 goto do_failover;
2828                         }
2829
2830                         continue;
2831
2832                 default:
2833                         netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2834                                    slave->new_link, slave->dev->name);
2835                         continue;
2836                 }
2837
2838 do_failover:
2839                 block_netpoll_tx();
2840                 bond_select_active_slave(bond);
2841                 unblock_netpoll_tx();
2842         }
2843
2844         bond_set_carrier(bond);
2845 }
2846
2847 /* Send ARP probes for active-backup mode ARP monitor.
2848  *
2849  * Called with rcu_read_lock held.
2850  */
2851 static bool bond_ab_arp_probe(struct bonding *bond)
2852 {
2853         struct slave *slave, *before = NULL, *new_slave = NULL,
2854                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2855                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2856         struct list_head *iter;
2857         bool found = false;
2858         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2859
2860         if (curr_arp_slave && curr_active_slave)
2861                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2862                             curr_arp_slave->dev->name,
2863                             curr_active_slave->dev->name);
2864
2865         if (curr_active_slave) {
2866                 bond_arp_send_all(bond, curr_active_slave);
2867                 return should_notify_rtnl;
2868         }
2869
2870         /* if we don't have a curr_active_slave, search for the next available
2871          * backup slave from the current_arp_slave and make it the candidate
2872          * for becoming the curr_active_slave
2873          */
2874
2875         if (!curr_arp_slave) {
2876                 curr_arp_slave = bond_first_slave_rcu(bond);
2877                 if (!curr_arp_slave)
2878                         return should_notify_rtnl;
2879         }
2880
2881         bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2882
2883         bond_for_each_slave_rcu(bond, slave, iter) {
2884                 if (!found && !before && bond_slave_is_up(slave))
2885                         before = slave;
2886
2887                 if (found && !new_slave && bond_slave_is_up(slave))
2888                         new_slave = slave;
2889                 /* if the link state is up at this point, we
2890                  * mark it down - this can happen if we have
2891                  * simultaneous link failures and
2892                  * reselect_active_interface doesn't make this
2893                  * one the current slave so it is still marked
2894                  * up when it is actually down
2895                  */
2896                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2897                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2898                                                   BOND_SLAVE_NOTIFY_LATER);
2899                         if (slave->link_failure_count < UINT_MAX)
2900                                 slave->link_failure_count++;
2901
2902                         bond_set_slave_inactive_flags(slave,
2903                                                       BOND_SLAVE_NOTIFY_LATER);
2904
2905                         netdev_info(bond->dev, "backup interface %s is now down\n",
2906                                     slave->dev->name);
2907                 }
2908                 if (slave == curr_arp_slave)
2909                         found = true;
2910         }
2911
2912         if (!new_slave && before)
2913                 new_slave = before;
2914
2915         if (!new_slave)
2916                 goto check_state;
2917
2918         bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2919                                   BOND_SLAVE_NOTIFY_LATER);
2920         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2921         bond_arp_send_all(bond, new_slave);
2922         new_slave->last_link_up = jiffies;
2923         rcu_assign_pointer(bond->current_arp_slave, new_slave);
2924
2925 check_state:
2926         bond_for_each_slave_rcu(bond, slave, iter) {
2927                 if (slave->should_notify || slave->should_notify_link) {
2928                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2929                         break;
2930                 }
2931         }
2932         return should_notify_rtnl;
2933 }
2934
2935 static void bond_activebackup_arp_mon(struct work_struct *work)
2936 {
2937         struct bonding *bond = container_of(work, struct bonding,
2938                                             arp_work.work);
2939         bool should_notify_peers = false;
2940         bool should_notify_rtnl = false;
2941         int delta_in_ticks;
2942
2943         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2944
2945         if (!bond_has_slaves(bond))
2946                 goto re_arm;
2947
2948         rcu_read_lock();
2949
2950         should_notify_peers = bond_should_notify_peers(bond);
2951
2952         if (bond_ab_arp_inspect(bond)) {
2953                 rcu_read_unlock();
2954
2955                 /* Race avoidance with bond_close flush of workqueue */
2956                 if (!rtnl_trylock()) {
2957                         delta_in_ticks = 1;
2958                         should_notify_peers = false;
2959                         goto re_arm;
2960                 }
2961
2962                 bond_ab_arp_commit(bond);
2963
2964                 rtnl_unlock();
2965                 rcu_read_lock();
2966         }
2967
2968         should_notify_rtnl = bond_ab_arp_probe(bond);
2969         rcu_read_unlock();
2970
2971 re_arm:
2972         if (bond->params.arp_interval)
2973                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2974
2975         if (should_notify_peers || should_notify_rtnl) {
2976                 if (!rtnl_trylock())
2977                         return;
2978
2979                 if (should_notify_peers)
2980                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2981                                                  bond->dev);
2982                 if (should_notify_rtnl) {
2983                         bond_slave_state_notify(bond);
2984                         bond_slave_link_notify(bond);
2985                 }
2986
2987                 rtnl_unlock();
2988         }
2989 }
2990
2991 /*-------------------------- netdev event handling --------------------------*/
2992
2993 /* Change device name */
2994 static int bond_event_changename(struct bonding *bond)
2995 {
2996         bond_remove_proc_entry(bond);
2997         bond_create_proc_entry(bond);
2998
2999         bond_debug_reregister(bond);
3000
3001         return NOTIFY_DONE;
3002 }
3003
3004 static int bond_master_netdev_event(unsigned long event,
3005                                     struct net_device *bond_dev)
3006 {
3007         struct bonding *event_bond = netdev_priv(bond_dev);
3008
3009         switch (event) {
3010         case NETDEV_CHANGENAME:
3011                 return bond_event_changename(event_bond);
3012         case NETDEV_UNREGISTER:
3013                 bond_remove_proc_entry(event_bond);
3014                 break;
3015         case NETDEV_REGISTER:
3016                 bond_create_proc_entry(event_bond);
3017                 break;
3018         case NETDEV_NOTIFY_PEERS:
3019                 if (event_bond->send_peer_notif)
3020                         event_bond->send_peer_notif--;
3021                 break;
3022         default:
3023                 break;
3024         }
3025
3026         return NOTIFY_DONE;
3027 }
3028
3029 static int bond_slave_netdev_event(unsigned long event,
3030                                    struct net_device *slave_dev)
3031 {
3032         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3033         struct bonding *bond;
3034         struct net_device *bond_dev;
3035
3036         /* A netdev event can be generated while enslaving a device
3037          * before netdev_rx_handler_register is called in which case
3038          * slave will be NULL
3039          */
3040         if (!slave)
3041                 return NOTIFY_DONE;
3042         bond_dev = slave->bond->dev;
3043         bond = slave->bond;
3044         primary = rtnl_dereference(bond->primary_slave);
3045
3046         switch (event) {
3047         case NETDEV_UNREGISTER:
3048                 if (bond_dev->type != ARPHRD_ETHER)
3049                         bond_release_and_destroy(bond_dev, slave_dev);
3050                 else
3051                         bond_release(bond_dev, slave_dev);
3052                 break;
3053         case NETDEV_UP:
3054         case NETDEV_CHANGE:
3055                 bond_update_speed_duplex(slave);
3056                 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3057                         bond_3ad_adapter_speed_duplex_changed(slave);
3058                 /* Fallthrough */
3059         case NETDEV_DOWN:
3060                 /* Refresh slave-array if applicable!
3061                  * If the setup does not use miimon or arpmon (mode-specific!),
3062                  * then these events will not cause the slave-array to be
3063                  * refreshed. This will cause xmit to use a slave that is not
3064                  * usable. Avoid such situation by refeshing the array at these
3065                  * events. If these (miimon/arpmon) parameters are configured
3066                  * then array gets refreshed twice and that should be fine!
3067                  */
3068                 if (bond_mode_uses_xmit_hash(bond))
3069                         bond_update_slave_arr(bond, NULL);
3070                 break;
3071         case NETDEV_CHANGEMTU:
3072                 /* TODO: Should slaves be allowed to
3073                  * independently alter their MTU?  For
3074                  * an active-backup bond, slaves need
3075                  * not be the same type of device, so
3076                  * MTUs may vary.  For other modes,
3077                  * slaves arguably should have the
3078                  * same MTUs. To do this, we'd need to
3079                  * take over the slave's change_mtu
3080                  * function for the duration of their
3081                  * servitude.
3082                  */
3083                 break;
3084         case NETDEV_CHANGENAME:
3085                 /* we don't care if we don't have primary set */
3086                 if (!bond_uses_primary(bond) ||
3087                     !bond->params.primary[0])
3088                         break;
3089
3090                 if (slave == primary) {
3091                         /* slave's name changed - he's no longer primary */
3092                         RCU_INIT_POINTER(bond->primary_slave, NULL);
3093                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3094                         /* we have a new primary slave */
3095                         rcu_assign_pointer(bond->primary_slave, slave);
3096                 } else { /* we didn't change primary - exit */
3097                         break;
3098                 }
3099
3100                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3101                             primary ? slave_dev->name : "none");
3102
3103                 block_netpoll_tx();
3104                 bond_select_active_slave(bond);
3105                 unblock_netpoll_tx();
3106                 break;
3107         case NETDEV_FEAT_CHANGE:
3108                 bond_compute_features(bond);
3109                 break;
3110         case NETDEV_RESEND_IGMP:
3111                 /* Propagate to master device */
3112                 call_netdevice_notifiers(event, slave->bond->dev);
3113                 break;
3114         default:
3115                 break;
3116         }
3117
3118         return NOTIFY_DONE;
3119 }
3120
3121 /* bond_netdev_event: handle netdev notifier chain events.
3122  *
3123  * This function receives events for the netdev chain.  The caller (an
3124  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3125  * locks for us to safely manipulate the slave devices (RTNL lock,
3126  * dev_probe_lock).
3127  */
3128 static int bond_netdev_event(struct notifier_block *this,
3129                              unsigned long event, void *ptr)
3130 {
3131         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3132
3133         netdev_dbg(event_dev, "event: %lx\n", event);
3134
3135         if (!(event_dev->priv_flags & IFF_BONDING))
3136                 return NOTIFY_DONE;
3137
3138         if (event_dev->flags & IFF_MASTER) {
3139                 int ret;
3140
3141                 netdev_dbg(event_dev, "IFF_MASTER\n");
3142                 ret = bond_master_netdev_event(event, event_dev);
3143                 if (ret != NOTIFY_DONE)
3144                         return ret;
3145         }
3146
3147         if (event_dev->flags & IFF_SLAVE) {
3148                 netdev_dbg(event_dev, "IFF_SLAVE\n");
3149                 return bond_slave_netdev_event(event, event_dev);
3150         }
3151
3152         return NOTIFY_DONE;
3153 }
3154
3155 static struct notifier_block bond_netdev_notifier = {
3156         .notifier_call = bond_netdev_event,
3157 };
3158
3159 /*---------------------------- Hashing Policies -----------------------------*/
3160
3161 /* L2 hash helper */
3162 static inline u32 bond_eth_hash(struct sk_buff *skb)
3163 {
3164         struct ethhdr *ep, hdr_tmp;
3165
3166         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3167         if (ep)
3168                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3169         return 0;
3170 }
3171
3172 /* Extract the appropriate headers based on bond's xmit policy */
3173 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3174                               struct flow_keys *fk)
3175 {
3176         const struct ipv6hdr *iph6;
3177         const struct iphdr *iph;
3178         int noff, proto = -1;
3179
3180         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3181                 return skb_flow_dissect_flow_keys(skb, fk, 0);
3182
3183         fk->ports.ports = 0;
3184         noff = skb_network_offset(skb);
3185         if (skb->protocol == htons(ETH_P_IP)) {
3186                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3187                         return false;
3188                 iph = ip_hdr(skb);
3189                 iph_to_flow_copy_v4addrs(fk, iph);
3190                 noff += iph->ihl << 2;
3191                 if (!ip_is_fragment(iph))
3192                         proto = iph->protocol;
3193         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3194                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3195                         return false;
3196                 iph6 = ipv6_hdr(skb);
3197                 iph_to_flow_copy_v6addrs(fk, iph6);
3198                 noff += sizeof(*iph6);
3199                 proto = iph6->nexthdr;
3200         } else {
3201                 return false;
3202         }
3203         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3204                 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3205
3206         return true;
3207 }
3208
3209 /**
3210  * bond_xmit_hash - generate a hash value based on the xmit policy
3211  * @bond: bonding device
3212  * @skb: buffer to use for headers
3213  *
3214  * This function will extract the necessary headers from the skb buffer and use
3215  * them to generate a hash based on the xmit_policy set in the bonding device
3216  */
3217 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3218 {
3219         struct flow_keys flow;
3220         u32 hash;
3221
3222         if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3223             skb->l4_hash)
3224                 return skb->hash;
3225
3226         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3227             !bond_flow_dissect(bond, skb, &flow))
3228                 return bond_eth_hash(skb);
3229
3230         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3231             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3232                 hash = bond_eth_hash(skb);
3233         else
3234                 hash = (__force u32)flow.ports.ports;
3235         hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3236                 (__force u32)flow_get_u32_src(&flow);
3237         hash ^= (hash >> 16);
3238         hash ^= (hash >> 8);
3239
3240         return hash >> 1;
3241 }
3242
3243 /*-------------------------- Device entry points ----------------------------*/
3244
3245 static void bond_work_init_all(struct bonding *bond)
3246 {
3247         INIT_DELAYED_WORK(&bond->mcast_work,
3248                           bond_resend_igmp_join_requests_delayed);
3249         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3250         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3251         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3252                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3253         else
3254                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3255         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3256         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3257 }
3258
3259 static void bond_work_cancel_all(struct bonding *bond)
3260 {
3261         cancel_delayed_work_sync(&bond->mii_work);
3262         cancel_delayed_work_sync(&bond->arp_work);
3263         cancel_delayed_work_sync(&bond->alb_work);
3264         cancel_delayed_work_sync(&bond->ad_work);
3265         cancel_delayed_work_sync(&bond->mcast_work);
3266         cancel_delayed_work_sync(&bond->slave_arr_work);
3267 }
3268
3269 static int bond_open(struct net_device *bond_dev)
3270 {
3271         struct bonding *bond = netdev_priv(bond_dev);
3272         struct list_head *iter;
3273         struct slave *slave;
3274
3275         /* reset slave->backup and slave->inactive */
3276         if (bond_has_slaves(bond)) {
3277                 bond_for_each_slave(bond, slave, iter) {
3278                         if (bond_uses_primary(bond) &&
3279                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3280                                 bond_set_slave_inactive_flags(slave,
3281                                                               BOND_SLAVE_NOTIFY_NOW);
3282                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3283                                 bond_set_slave_active_flags(slave,
3284                                                             BOND_SLAVE_NOTIFY_NOW);
3285                         }
3286                 }
3287         }
3288
3289         bond_work_init_all(bond);
3290
3291         if (bond_is_lb(bond)) {
3292                 /* bond_alb_initialize must be called before the timer
3293                  * is started.
3294                  */
3295                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3296                         return -ENOMEM;
3297                 if (bond->params.tlb_dynamic_lb)
3298                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3299         }
3300
3301         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3302                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3303
3304         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3305                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3306                 bond->recv_probe = bond_arp_rcv;
3307         }
3308
3309         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3310                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3311                 /* register to receive LACPDUs */
3312                 bond->recv_probe = bond_3ad_lacpdu_recv;
3313                 bond_3ad_initiate_agg_selection(bond, 1);
3314         }
3315
3316         if (bond_mode_uses_xmit_hash(bond))
3317                 bond_update_slave_arr(bond, NULL);
3318
3319         return 0;
3320 }
3321
3322 static int bond_close(struct net_device *bond_dev)
3323 {
3324         struct bonding *bond = netdev_priv(bond_dev);
3325
3326         bond_work_cancel_all(bond);
3327         bond->send_peer_notif = 0;
3328         if (bond_is_lb(bond))
3329                 bond_alb_deinitialize(bond);
3330         bond->recv_probe = NULL;
3331
3332         return 0;
3333 }
3334
3335 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3336  * that some drivers can provide 32bit values only.
3337  */
3338 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3339                             const struct rtnl_link_stats64 *_new,
3340                             const struct rtnl_link_stats64 *_old)
3341 {
3342         const u64 *new = (const u64 *)_new;
3343         const u64 *old = (const u64 *)_old;
3344         u64 *res = (u64 *)_res;
3345         int i;
3346
3347         for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3348                 u64 nv = new[i];
3349                 u64 ov = old[i];
3350                 s64 delta = nv - ov;
3351
3352                 /* detects if this particular field is 32bit only */
3353                 if (((nv | ov) >> 32) == 0)
3354                         delta = (s64)(s32)((u32)nv - (u32)ov);
3355
3356                 /* filter anomalies, some drivers reset their stats
3357                  * at down/up events.
3358                  */
3359                 if (delta > 0)
3360                         res[i] += delta;
3361         }
3362 }
3363
3364 static int bond_get_nest_level(struct net_device *bond_dev)
3365 {
3366         struct bonding *bond = netdev_priv(bond_dev);
3367
3368         return bond->nest_level;
3369 }
3370
3371 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3372                                                 struct rtnl_link_stats64 *stats)
3373 {
3374         struct bonding *bond = netdev_priv(bond_dev);
3375         struct rtnl_link_stats64 temp;
3376         struct list_head *iter;
3377         struct slave *slave;
3378
3379         spin_lock_nested(&bond->stats_lock, bond_get_nest_level(bond_dev));
3380         memcpy(stats, &bond->bond_stats, sizeof(*stats));
3381
3382         rcu_read_lock();
3383         bond_for_each_slave_rcu(bond, slave, iter) {
3384                 const struct rtnl_link_stats64 *new =
3385                         dev_get_stats(slave->dev, &temp);
3386
3387                 bond_fold_stats(stats, new, &slave->slave_stats);
3388
3389                 /* save off the slave stats for the next run */
3390                 memcpy(&slave->slave_stats, new, sizeof(*new));
3391         }
3392         rcu_read_unlock();
3393
3394         memcpy(&bond->bond_stats, stats, sizeof(*stats));
3395         spin_unlock(&bond->stats_lock);
3396
3397         return stats;
3398 }
3399
3400 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3401 {
3402         struct bonding *bond = netdev_priv(bond_dev);
3403         struct net_device *slave_dev = NULL;
3404         struct ifbond k_binfo;
3405         struct ifbond __user *u_binfo = NULL;
3406         struct ifslave k_sinfo;
3407         struct ifslave __user *u_sinfo = NULL;
3408         struct mii_ioctl_data *mii = NULL;
3409         struct bond_opt_value newval;
3410         struct net *net;
3411         int res = 0;
3412
3413         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3414
3415         switch (cmd) {
3416         case SIOCGMIIPHY:
3417                 mii = if_mii(ifr);
3418                 if (!mii)
3419                         return -EINVAL;
3420
3421                 mii->phy_id = 0;
3422                 /* Fall Through */
3423         case SIOCGMIIREG:
3424                 /* We do this again just in case we were called by SIOCGMIIREG
3425                  * instead of SIOCGMIIPHY.
3426                  */
3427                 mii = if_mii(ifr);
3428                 if (!mii)
3429                         return -EINVAL;
3430
3431                 if (mii->reg_num == 1) {
3432                         mii->val_out = 0;
3433                         if (netif_carrier_ok(bond->dev))
3434                                 mii->val_out = BMSR_LSTATUS;
3435                 }
3436
3437                 return 0;
3438         case BOND_INFO_QUERY_OLD:
3439         case SIOCBONDINFOQUERY:
3440                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3441
3442                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3443                         return -EFAULT;
3444
3445                 res = bond_info_query(bond_dev, &k_binfo);
3446                 if (res == 0 &&
3447                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3448                         return -EFAULT;
3449
3450                 return res;
3451         case BOND_SLAVE_INFO_QUERY_OLD:
3452         case SIOCBONDSLAVEINFOQUERY:
3453                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3454
3455                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3456                         return -EFAULT;
3457
3458                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3459                 if (res == 0 &&
3460                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3461                         return -EFAULT;
3462
3463                 return res;
3464         default:
3465                 break;
3466         }
3467
3468         net = dev_net(bond_dev);
3469
3470         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3471                 return -EPERM;
3472
3473         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3474
3475         netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3476
3477         if (!slave_dev)
3478                 return -ENODEV;
3479
3480         netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3481         switch (cmd) {
3482         case BOND_ENSLAVE_OLD:
3483         case SIOCBONDENSLAVE:
3484                 res = bond_enslave(bond_dev, slave_dev);
3485                 break;
3486         case BOND_RELEASE_OLD:
3487         case SIOCBONDRELEASE:
3488                 res = bond_release(bond_dev, slave_dev);
3489                 break;
3490         case BOND_SETHWADDR_OLD:
3491         case SIOCBONDSETHWADDR:
3492                 bond_set_dev_addr(bond_dev, slave_dev);
3493                 res = 0;
3494                 break;
3495         case BOND_CHANGE_ACTIVE_OLD:
3496         case SIOCBONDCHANGEACTIVE:
3497                 bond_opt_initstr(&newval, slave_dev->name);
3498                 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3499                 break;
3500         default:
3501                 res = -EOPNOTSUPP;
3502         }
3503
3504         return res;
3505 }
3506
3507 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3508 {
3509         struct bonding *bond = netdev_priv(bond_dev);
3510
3511         if (change & IFF_PROMISC)
3512                 bond_set_promiscuity(bond,
3513                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3514
3515         if (change & IFF_ALLMULTI)
3516                 bond_set_allmulti(bond,
3517                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3518 }
3519
3520 static void bond_set_rx_mode(struct net_device *bond_dev)
3521 {
3522         struct bonding *bond = netdev_priv(bond_dev);
3523         struct list_head *iter;
3524         struct slave *slave;
3525
3526         rcu_read_lock();
3527         if (bond_uses_primary(bond)) {
3528                 slave = rcu_dereference(bond->curr_active_slave);
3529                 if (slave) {
3530                         dev_uc_sync(slave->dev, bond_dev);
3531                         dev_mc_sync(slave->dev, bond_dev);
3532                 }
3533         } else {
3534                 bond_for_each_slave_rcu(bond, slave, iter) {
3535                         dev_uc_sync_multiple(slave->dev, bond_dev);
3536                         dev_mc_sync_multiple(slave->dev, bond_dev);
3537                 }
3538         }
3539         rcu_read_unlock();
3540 }
3541
3542 static int bond_neigh_init(struct neighbour *n)
3543 {
3544         struct bonding *bond = netdev_priv(n->dev);
3545         const struct net_device_ops *slave_ops;
3546         struct neigh_parms parms;
3547         struct slave *slave;
3548         int ret;
3549
3550         slave = bond_first_slave(bond);
3551         if (!slave)
3552                 return 0;
3553         slave_ops = slave->dev->netdev_ops;
3554         if (!slave_ops->ndo_neigh_setup)
3555                 return 0;
3556
3557         parms.neigh_setup = NULL;
3558         parms.neigh_cleanup = NULL;
3559         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3560         if (ret)
3561                 return ret;
3562
3563         /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3564          * after the last slave has been detached.  Assumes that all slaves
3565          * utilize the same neigh_cleanup (true at this writing as only user
3566          * is ipoib).
3567          */
3568         n->parms->neigh_cleanup = parms.neigh_cleanup;
3569
3570         if (!parms.neigh_setup)
3571                 return 0;
3572
3573         return parms.neigh_setup(n);
3574 }
3575
3576 /* The bonding ndo_neigh_setup is called at init time beofre any
3577  * slave exists. So we must declare proxy setup function which will
3578  * be used at run time to resolve the actual slave neigh param setup.
3579  *
3580  * It's also called by master devices (such as vlans) to setup their
3581  * underlying devices. In that case - do nothing, we're already set up from
3582  * our init.
3583  */
3584 static int bond_neigh_setup(struct net_device *dev,
3585                             struct neigh_parms *parms)
3586 {
3587         /* modify only our neigh_parms */
3588         if (parms->dev == dev)
3589                 parms->neigh_setup = bond_neigh_init;
3590
3591         return 0;
3592 }
3593
3594 /* Change the MTU of all of a master's slaves to match the master */
3595 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3596 {
3597         struct bonding *bond = netdev_priv(bond_dev);
3598         struct slave *slave, *rollback_slave;
3599         struct list_head *iter;
3600         int res = 0;
3601
3602         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3603
3604         bond_for_each_slave(bond, slave, iter) {
3605                 netdev_dbg(bond_dev, "s %p c_m %p\n",
3606                            slave, slave->dev->netdev_ops->ndo_change_mtu);
3607
3608                 res = dev_set_mtu(slave->dev, new_mtu);
3609
3610                 if (res) {
3611                         /* If we failed to set the slave's mtu to the new value
3612                          * we must abort the operation even in ACTIVE_BACKUP
3613                          * mode, because if we allow the backup slaves to have
3614                          * different mtu values than the active slave we'll
3615                          * need to change their mtu when doing a failover. That
3616                          * means changing their mtu from timer context, which
3617                          * is probably not a good idea.
3618                          */
3619                         netdev_dbg(bond_dev, "err %d %s\n", res,
3620                                    slave->dev->name);
3621                         goto unwind;
3622                 }
3623         }
3624
3625         bond_dev->mtu = new_mtu;
3626
3627         return 0;
3628
3629 unwind:
3630         /* unwind from head to the slave that failed */
3631         bond_for_each_slave(bond, rollback_slave, iter) {
3632                 int tmp_res;
3633
3634                 if (rollback_slave == slave)
3635                         break;
3636
3637                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3638                 if (tmp_res) {
3639                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3640                                    tmp_res, rollback_slave->dev->name);
3641                 }
3642         }
3643
3644         return res;
3645 }
3646
3647 /* Change HW address
3648  *
3649  * Note that many devices must be down to change the HW address, and
3650  * downing the master releases all slaves.  We can make bonds full of
3651  * bonding devices to test this, however.
3652  */
3653 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3654 {
3655         struct bonding *bond = netdev_priv(bond_dev);
3656         struct slave *slave, *rollback_slave;
3657         struct sockaddr *sa = addr, tmp_sa;
3658         struct list_head *iter;
3659         int res = 0;
3660
3661         if (BOND_MODE(bond) == BOND_MODE_ALB)
3662                 return bond_alb_set_mac_address(bond_dev, addr);
3663
3664
3665         netdev_dbg(bond_dev, "bond=%p\n", bond);
3666
3667         /* If fail_over_mac is enabled, do nothing and return success.
3668          * Returning an error causes ifenslave to fail.
3669          */
3670         if (bond->params.fail_over_mac &&
3671             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3672                 return 0;
3673
3674         if (!is_valid_ether_addr(sa->sa_data))
3675                 return -EADDRNOTAVAIL;
3676
3677         bond_for_each_slave(bond, slave, iter) {
3678                 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3679                 res = dev_set_mac_address(slave->dev, addr);
3680                 if (res) {
3681                         /* TODO: consider downing the slave
3682                          * and retry ?
3683                          * User should expect communications
3684                          * breakage anyway until ARP finish
3685                          * updating, so...
3686                          */
3687                         netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3688                         goto unwind;
3689                 }
3690         }
3691
3692         /* success */
3693         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3694         return 0;
3695
3696 unwind:
3697         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3698         tmp_sa.sa_family = bond_dev->type;
3699
3700         /* unwind from head to the slave that failed */
3701         bond_for_each_slave(bond, rollback_slave, iter) {
3702                 int tmp_res;
3703
3704                 if (rollback_slave == slave)
3705                         break;
3706
3707                 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3708                 if (tmp_res) {
3709                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3710                                    tmp_res, rollback_slave->dev->name);
3711                 }
3712         }
3713
3714         return res;
3715 }
3716
3717 /**
3718  * bond_xmit_slave_id - transmit skb through slave with slave_id
3719  * @bond: bonding device that is transmitting
3720  * @skb: buffer to transmit
3721  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3722  *
3723  * This function tries to transmit through slave with slave_id but in case
3724  * it fails, it tries to find the first available slave for transmission.
3725  * The skb is consumed in all cases, thus the function is void.
3726  */
3727 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3728 {
3729         struct list_head *iter;
3730         struct slave *slave;
3731         int i = slave_id;
3732
3733         /* Here we start from the slave with slave_id */
3734         bond_for_each_slave_rcu(bond, slave, iter) {
3735                 if (--i < 0) {
3736                         if (bond_slave_can_tx(slave)) {
3737                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3738                                 return;
3739                         }
3740                 }
3741         }
3742
3743         /* Here we start from the first slave up to slave_id */
3744         i = slave_id;
3745         bond_for_each_slave_rcu(bond, slave, iter) {
3746                 if (--i < 0)
3747                         break;
3748                 if (bond_slave_can_tx(slave)) {
3749                         bond_dev_queue_xmit(bond, skb, slave->dev);
3750                         return;
3751                 }
3752         }
3753         /* no slave that can tx has been found */
3754         bond_tx_drop(bond->dev, skb);
3755 }
3756
3757 /**
3758  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3759  * @bond: bonding device to use
3760  *
3761  * Based on the value of the bonding device's packets_per_slave parameter
3762  * this function generates a slave id, which is usually used as the next
3763  * slave to transmit through.
3764  */
3765 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3766 {
3767         u32 slave_id;
3768         struct reciprocal_value reciprocal_packets_per_slave;
3769         int packets_per_slave = bond->params.packets_per_slave;
3770
3771         switch (packets_per_slave) {
3772         case 0:
3773                 slave_id = prandom_u32();
3774                 break;
3775         case 1:
3776                 slave_id = bond->rr_tx_counter;
3777                 break;
3778         default:
3779                 reciprocal_packets_per_slave =
3780                         bond->params.reciprocal_packets_per_slave;
3781                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3782                                              reciprocal_packets_per_slave);
3783                 break;
3784         }
3785         bond->rr_tx_counter++;
3786
3787         return slave_id;
3788 }
3789
3790 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3791 {
3792         struct bonding *bond = netdev_priv(bond_dev);
3793         struct slave *slave;
3794         int slave_cnt;
3795         u32 slave_id;
3796
3797         /* Start with the curr_active_slave that joined the bond as the
3798          * default for sending IGMP traffic.  For failover purposes one
3799          * needs to maintain some consistency for the interface that will
3800          * send the join/membership reports.  The curr_active_slave found
3801          * will send all of this type of traffic.
3802          */
3803         if (skb->protocol == htons(ETH_P_IP)) {
3804                 int noff = skb_network_offset(skb);
3805                 struct iphdr *iph;
3806
3807                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3808                         goto non_igmp;
3809
3810                 iph = ip_hdr(skb);
3811                 if (iph->protocol == IPPROTO_IGMP) {
3812                         slave = rcu_dereference(bond->curr_active_slave);
3813                         if (slave)
3814                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3815                         else
3816                                 bond_xmit_slave_id(bond, skb, 0);
3817                         return NETDEV_TX_OK;
3818                 }
3819         }
3820
3821 non_igmp:
3822         slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3823         if (likely(slave_cnt)) {
3824                 slave_id = bond_rr_gen_slave_id(bond);
3825                 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3826         } else {
3827                 bond_tx_drop(bond_dev, skb);
3828         }
3829         return NETDEV_TX_OK;
3830 }
3831
3832 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3833  * the bond has a usable interface.
3834  */
3835 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3836 {
3837         struct bonding *bond = netdev_priv(bond_dev);
3838         struct slave *slave;
3839
3840         slave = rcu_dereference(bond->curr_active_slave);
3841         if (slave)
3842                 bond_dev_queue_xmit(bond, skb, slave->dev);
3843         else
3844                 bond_tx_drop(bond_dev, skb);
3845
3846         return NETDEV_TX_OK;
3847 }
3848
3849 /* Use this to update slave_array when (a) it's not appropriate to update
3850  * slave_array right away (note that update_slave_array() may sleep)
3851  * and / or (b) RTNL is not held.
3852  */
3853 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3854 {
3855         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3856 }
3857
3858 /* Slave array work handler. Holds only RTNL */
3859 static void bond_slave_arr_handler(struct work_struct *work)
3860 {
3861         struct bonding *bond = container_of(work, struct bonding,
3862                                             slave_arr_work.work);
3863         int ret;
3864
3865         if (!rtnl_trylock())
3866                 goto err;
3867
3868         ret = bond_update_slave_arr(bond, NULL);
3869         rtnl_unlock();
3870         if (ret) {
3871                 pr_warn_ratelimited("Failed to update slave array from WT\n");
3872                 goto err;
3873         }
3874         return;
3875
3876 err:
3877         bond_slave_arr_work_rearm(bond, 1);
3878 }
3879
3880 /* Build the usable slaves array in control path for modes that use xmit-hash
3881  * to determine the slave interface -
3882  * (a) BOND_MODE_8023AD
3883  * (b) BOND_MODE_XOR
3884  * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3885  *
3886  * The caller is expected to hold RTNL only and NO other lock!
3887  */
3888 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3889 {
3890         struct slave *slave;
3891         struct list_head *iter;
3892         struct bond_up_slave *new_arr, *old_arr;
3893         int agg_id = 0;
3894         int ret = 0;
3895
3896 #ifdef CONFIG_LOCKDEP
3897         WARN_ON(lockdep_is_held(&bond->mode_lock));
3898 #endif
3899
3900         new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3901                           GFP_KERNEL);
3902         if (!new_arr) {
3903                 ret = -ENOMEM;
3904                 pr_err("Failed to build slave-array.\n");
3905                 goto out;
3906         }
3907         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3908                 struct ad_info ad_info;
3909
3910                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3911                         pr_debug("bond_3ad_get_active_agg_info failed\n");
3912                         kfree_rcu(new_arr, rcu);
3913                         /* No active aggragator means it's not safe to use
3914                          * the previous array.
3915                          */
3916                         old_arr = rtnl_dereference(bond->slave_arr);
3917                         if (old_arr) {
3918                                 RCU_INIT_POINTER(bond->slave_arr, NULL);
3919                                 kfree_rcu(old_arr, rcu);
3920                         }
3921                         goto out;
3922                 }
3923                 agg_id = ad_info.aggregator_id;
3924         }
3925         bond_for_each_slave(bond, slave, iter) {
3926                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3927                         struct aggregator *agg;
3928
3929                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
3930                         if (!agg || agg->aggregator_identifier != agg_id)
3931                                 continue;
3932                 }
3933                 if (!bond_slave_can_tx(slave))
3934                         continue;
3935                 if (skipslave == slave)
3936                         continue;
3937                 new_arr->arr[new_arr->count++] = slave;
3938         }
3939
3940         old_arr = rtnl_dereference(bond->slave_arr);
3941         rcu_assign_pointer(bond->slave_arr, new_arr);
3942         if (old_arr)
3943                 kfree_rcu(old_arr, rcu);
3944 out:
3945         if (ret != 0 && skipslave) {
3946                 int idx;
3947
3948                 /* Rare situation where caller has asked to skip a specific
3949                  * slave but allocation failed (most likely!). BTW this is
3950                  * only possible when the call is initiated from
3951                  * __bond_release_one(). In this situation; overwrite the
3952                  * skipslave entry in the array with the last entry from the
3953                  * array to avoid a situation where the xmit path may choose
3954                  * this to-be-skipped slave to send a packet out.
3955                  */
3956                 old_arr = rtnl_dereference(bond->slave_arr);
3957                 for (idx = 0; idx < old_arr->count; idx++) {
3958                         if (skipslave == old_arr->arr[idx]) {
3959                                 old_arr->arr[idx] =
3960                                     old_arr->arr[old_arr->count-1];
3961                                 old_arr->count--;
3962                                 break;
3963                         }
3964                 }
3965         }
3966         return ret;
3967 }
3968
3969 /* Use this Xmit function for 3AD as well as XOR modes. The current
3970  * usable slave array is formed in the control path. The xmit function
3971  * just calculates hash and sends the packet out.
3972  */
3973 static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3974 {
3975         struct bonding *bond = netdev_priv(dev);
3976         struct slave *slave;
3977         struct bond_up_slave *slaves;
3978         unsigned int count;
3979
3980         slaves = rcu_dereference(bond->slave_arr);
3981         count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3982         if (likely(count)) {
3983                 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3984                 bond_dev_queue_xmit(bond, skb, slave->dev);
3985         } else {
3986                 bond_tx_drop(dev, skb);
3987         }
3988
3989         return NETDEV_TX_OK;
3990 }
3991
3992 /* in broadcast mode, we send everything to all usable interfaces. */
3993 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3994 {
3995         struct bonding *bond = netdev_priv(bond_dev);
3996         struct slave *slave = NULL;
3997         struct list_head *iter;
3998
3999         bond_for_each_slave_rcu(bond, slave, iter) {
4000                 if (bond_is_last_slave(bond, slave))
4001                         break;
4002                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4003                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4004
4005                         if (!skb2) {
4006                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4007                                                     bond_dev->name, __func__);
4008                                 continue;
4009                         }
4010                         bond_dev_queue_xmit(bond, skb2, slave->dev);
4011                 }
4012         }
4013         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4014                 bond_dev_queue_xmit(bond, skb, slave->dev);
4015         else
4016                 bond_tx_drop(bond_dev, skb);
4017
4018         return NETDEV_TX_OK;
4019 }
4020
4021 /*------------------------- Device initialization ---------------------------*/
4022
4023 /* Lookup the slave that corresponds to a qid */
4024 static inline int bond_slave_override(struct bonding *bond,
4025                                       struct sk_buff *skb)
4026 {
4027         struct slave *slave = NULL;
4028         struct list_head *iter;
4029
4030         if (!skb->queue_mapping)
4031                 return 1;
4032
4033         /* Find out if any slaves have the same mapping as this skb. */
4034         bond_for_each_slave_rcu(bond, slave, iter) {
4035                 if (slave->queue_id == skb->queue_mapping) {
4036                         if (bond_slave_is_up(slave) &&
4037                             slave->link == BOND_LINK_UP) {
4038                                 bond_dev_queue_xmit(bond, skb, slave->dev);
4039                                 return 0;
4040                         }
4041                         /* If the slave isn't UP, use default transmit policy. */
4042                         break;
4043                 }
4044         }
4045
4046         return 1;
4047 }
4048
4049
4050 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4051                              void *accel_priv, select_queue_fallback_t fallback)
4052 {
4053         /* This helper function exists to help dev_pick_tx get the correct
4054          * destination queue.  Using a helper function skips a call to
4055          * skb_tx_hash and will put the skbs in the queue we expect on their
4056          * way down to the bonding driver.
4057          */
4058         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4059
4060         /* Save the original txq to restore before passing to the driver */
4061         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4062
4063         if (unlikely(txq >= dev->real_num_tx_queues)) {
4064                 do {
4065                         txq -= dev->real_num_tx_queues;
4066                 } while (txq >= dev->real_num_tx_queues);
4067         }
4068         return txq;
4069 }
4070
4071 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4072 {
4073         struct bonding *bond = netdev_priv(dev);
4074
4075         if (bond_should_override_tx_queue(bond) &&
4076             !bond_slave_override(bond, skb))
4077                 return NETDEV_TX_OK;
4078
4079         switch (BOND_MODE(bond)) {
4080         case BOND_MODE_ROUNDROBIN:
4081                 return bond_xmit_roundrobin(skb, dev);
4082         case BOND_MODE_ACTIVEBACKUP:
4083                 return bond_xmit_activebackup(skb, dev);
4084         case BOND_MODE_8023AD:
4085         case BOND_MODE_XOR:
4086                 return bond_3ad_xor_xmit(skb, dev);
4087         case BOND_MODE_BROADCAST:
4088                 return bond_xmit_broadcast(skb, dev);
4089         case BOND_MODE_ALB:
4090                 return bond_alb_xmit(skb, dev);
4091         case BOND_MODE_TLB:
4092                 return bond_tlb_xmit(skb, dev);
4093         default:
4094                 /* Should never happen, mode already checked */
4095                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4096                 WARN_ON_ONCE(1);
4097                 bond_tx_drop(dev, skb);
4098                 return NETDEV_TX_OK;
4099         }
4100 }
4101
4102 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4103 {
4104         struct bonding *bond = netdev_priv(dev);
4105         netdev_tx_t ret = NETDEV_TX_OK;
4106
4107         /* If we risk deadlock from transmitting this in the
4108          * netpoll path, tell netpoll to queue the frame for later tx
4109          */
4110         if (unlikely(is_netpoll_tx_blocked(dev)))
4111                 return NETDEV_TX_BUSY;
4112
4113         rcu_read_lock();
4114         if (bond_has_slaves(bond))
4115                 ret = __bond_start_xmit(skb, dev);
4116         else
4117                 bond_tx_drop(dev, skb);
4118         rcu_read_unlock();
4119
4120         return ret;
4121 }
4122
4123 static int bond_ethtool_get_settings(struct net_device *bond_dev,
4124                                      struct ethtool_cmd *ecmd)
4125 {
4126         struct bonding *bond = netdev_priv(bond_dev);
4127         unsigned long speed = 0;
4128         struct list_head *iter;
4129         struct slave *slave;
4130
4131         ecmd->duplex = DUPLEX_UNKNOWN;
4132         ecmd->port = PORT_OTHER;
4133
4134         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4135          * do not need to check mode.  Though link speed might not represent
4136          * the true receive or transmit bandwidth (not all modes are symmetric)
4137          * this is an accurate maximum.
4138          */
4139         bond_for_each_slave(bond, slave, iter) {
4140                 if (bond_slave_can_tx(slave)) {
4141                         if (slave->speed != SPEED_UNKNOWN)
4142                                 speed += slave->speed;
4143                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
4144                             slave->duplex != DUPLEX_UNKNOWN)
4145                                 ecmd->duplex = slave->duplex;
4146                 }
4147         }
4148         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
4149
4150         return 0;
4151 }
4152
4153 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4154                                      struct ethtool_drvinfo *drvinfo)
4155 {
4156         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4157         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4158         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4159                  BOND_ABI_VERSION);
4160 }
4161
4162 static const struct ethtool_ops bond_ethtool_ops = {
4163         .get_drvinfo            = bond_ethtool_get_drvinfo,
4164         .get_settings           = bond_ethtool_get_settings,
4165         .get_link               = ethtool_op_get_link,
4166 };
4167
4168 static const struct net_device_ops bond_netdev_ops = {
4169         .ndo_init               = bond_init,
4170         .ndo_uninit             = bond_uninit,
4171         .ndo_open               = bond_open,
4172         .ndo_stop               = bond_close,
4173         .ndo_start_xmit         = bond_start_xmit,
4174         .ndo_select_queue       = bond_select_queue,
4175         .ndo_get_stats64        = bond_get_stats,
4176         .ndo_do_ioctl           = bond_do_ioctl,
4177         .ndo_change_rx_flags    = bond_change_rx_flags,
4178         .ndo_set_rx_mode        = bond_set_rx_mode,
4179         .ndo_change_mtu         = bond_change_mtu,
4180         .ndo_set_mac_address    = bond_set_mac_address,
4181         .ndo_neigh_setup        = bond_neigh_setup,
4182         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4183         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4184         .ndo_get_lock_subclass  = bond_get_nest_level,
4185 #ifdef CONFIG_NET_POLL_CONTROLLER
4186         .ndo_netpoll_setup      = bond_netpoll_setup,
4187         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4188         .ndo_poll_controller    = bond_poll_controller,
4189 #endif
4190         .ndo_add_slave          = bond_enslave,
4191         .ndo_del_slave          = bond_release,
4192         .ndo_fix_features       = bond_fix_features,
4193         .ndo_neigh_construct    = netdev_default_l2upper_neigh_construct,
4194         .ndo_neigh_destroy      = netdev_default_l2upper_neigh_destroy,
4195         .ndo_bridge_setlink     = switchdev_port_bridge_setlink,
4196         .ndo_bridge_getlink     = switchdev_port_bridge_getlink,
4197         .ndo_bridge_dellink     = switchdev_port_bridge_dellink,
4198         .ndo_fdb_add            = switchdev_port_fdb_add,
4199         .ndo_fdb_del            = switchdev_port_fdb_del,
4200         .ndo_fdb_dump           = switchdev_port_fdb_dump,
4201         .ndo_features_check     = passthru_features_check,
4202 };
4203
4204 static const struct device_type bond_type = {
4205         .name = "bond",
4206 };
4207
4208 static void bond_destructor(struct net_device *bond_dev)
4209 {
4210         struct bonding *bond = netdev_priv(bond_dev);
4211         if (bond->wq)
4212                 destroy_workqueue(bond->wq);
4213         free_netdev(bond_dev);
4214 }
4215
4216 void bond_setup(struct net_device *bond_dev)
4217 {
4218         struct bonding *bond = netdev_priv(bond_dev);
4219
4220         spin_lock_init(&bond->mode_lock);
4221         spin_lock_init(&bond->stats_lock);
4222         bond->params = bonding_defaults;
4223
4224         /* Initialize pointers */
4225         bond->dev = bond_dev;
4226
4227         /* Initialize the device entry points */
4228         ether_setup(bond_dev);
4229         bond_dev->netdev_ops = &bond_netdev_ops;
4230         bond_dev->ethtool_ops = &bond_ethtool_ops;
4231
4232         bond_dev->destructor = bond_destructor;
4233
4234         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4235
4236         /* Initialize the device options */
4237         bond_dev->flags |= IFF_MASTER;
4238         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4239         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4240
4241         /* don't acquire bond device's netif_tx_lock when transmitting */
4242         bond_dev->features |= NETIF_F_LLTX;
4243
4244         /* By default, we declare the bond to be fully
4245          * VLAN hardware accelerated capable. Special
4246          * care is taken in the various xmit functions
4247          * when there are slaves that are not hw accel
4248          * capable
4249          */
4250
4251         /* Don't allow bond devices to change network namespaces. */
4252         bond_dev->features |= NETIF_F_NETNS_LOCAL;
4253
4254         bond_dev->hw_features = BOND_VLAN_FEATURES |
4255                                 NETIF_F_HW_VLAN_CTAG_RX |
4256                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4257
4258         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4259         bond_dev->features |= bond_dev->hw_features;
4260         bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX;
4261 }
4262
4263 /* Destroy a bonding device.
4264  * Must be under rtnl_lock when this function is called.
4265  */
4266 static void bond_uninit(struct net_device *bond_dev)
4267 {
4268         struct bonding *bond = netdev_priv(bond_dev);
4269         struct list_head *iter;
4270         struct slave *slave;
4271         struct bond_up_slave *arr;
4272
4273         bond_netpoll_cleanup(bond_dev);
4274
4275         /* Release the bonded slaves */
4276         bond_for_each_slave(bond, slave, iter)
4277                 __bond_release_one(bond_dev, slave->dev, true);
4278         netdev_info(bond_dev, "Released all slaves\n");
4279
4280         arr = rtnl_dereference(bond->slave_arr);
4281         if (arr) {
4282                 RCU_INIT_POINTER(bond->slave_arr, NULL);
4283                 kfree_rcu(arr, rcu);
4284         }
4285
4286         list_del(&bond->bond_list);
4287
4288         bond_debug_unregister(bond);
4289 }
4290
4291 /*------------------------- Module initialization ---------------------------*/
4292
4293 static int bond_check_params(struct bond_params *params)
4294 {
4295         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4296         struct bond_opt_value newval;
4297         const struct bond_opt_value *valptr;
4298         int arp_all_targets_value;
4299         u16 ad_actor_sys_prio = 0;
4300         u16 ad_user_port_key = 0;
4301
4302         /* Convert string parameters. */
4303         if (mode) {
4304                 bond_opt_initstr(&newval, mode);
4305                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4306                 if (!valptr) {
4307                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4308                         return -EINVAL;
4309                 }
4310                 bond_mode = valptr->value;
4311         }
4312
4313         if (xmit_hash_policy) {
4314                 if ((bond_mode != BOND_MODE_XOR) &&
4315                     (bond_mode != BOND_MODE_8023AD) &&
4316                     (bond_mode != BOND_MODE_TLB)) {
4317                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4318                                 bond_mode_name(bond_mode));
4319                 } else {
4320                         bond_opt_initstr(&newval, xmit_hash_policy);
4321                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4322                                                 &newval);
4323                         if (!valptr) {
4324                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4325                                        xmit_hash_policy);
4326                                 return -EINVAL;
4327                         }
4328                         xmit_hashtype = valptr->value;
4329                 }
4330         }
4331
4332         if (lacp_rate) {
4333                 if (bond_mode != BOND_MODE_8023AD) {
4334                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4335                                 bond_mode_name(bond_mode));
4336                 } else {
4337                         bond_opt_initstr(&newval, lacp_rate);
4338                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4339                                                 &newval);
4340                         if (!valptr) {
4341                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4342                                        lacp_rate);
4343                                 return -EINVAL;
4344                         }
4345                         lacp_fast = valptr->value;
4346                 }
4347         }
4348
4349         if (ad_select) {
4350                 bond_opt_initstr(&newval, ad_select);
4351                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4352                                         &newval);
4353                 if (!valptr) {
4354                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4355                         return -EINVAL;
4356                 }
4357                 params->ad_select = valptr->value;
4358                 if (bond_mode != BOND_MODE_8023AD)
4359                         pr_warn("ad_select param only affects 802.3ad mode\n");
4360         } else {
4361                 params->ad_select = BOND_AD_STABLE;
4362         }
4363
4364         if (max_bonds < 0) {
4365                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4366                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4367                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4368         }
4369
4370         if (miimon < 0) {
4371                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4372                         miimon, INT_MAX);
4373                 miimon = 0;
4374         }
4375
4376         if (updelay < 0) {
4377                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4378                         updelay, INT_MAX);
4379                 updelay = 0;
4380         }
4381
4382         if (downdelay < 0) {
4383                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4384                         downdelay, INT_MAX);
4385                 downdelay = 0;
4386         }
4387
4388         if ((use_carrier != 0) && (use_carrier != 1)) {
4389                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4390                         use_carrier);
4391                 use_carrier = 1;
4392         }
4393
4394         if (num_peer_notif < 0 || num_peer_notif > 255) {
4395                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4396                         num_peer_notif);
4397                 num_peer_notif = 1;
4398         }
4399
4400         /* reset values for 802.3ad/TLB/ALB */
4401         if (!bond_mode_uses_arp(bond_mode)) {
4402                 if (!miimon) {
4403                         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");
4404                         pr_warn("Forcing miimon to 100msec\n");
4405                         miimon = BOND_DEFAULT_MIIMON;
4406                 }
4407         }
4408
4409         if (tx_queues < 1 || tx_queues > 255) {
4410                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4411                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4412                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4413         }
4414
4415         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4416                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4417                         all_slaves_active);
4418                 all_slaves_active = 0;
4419         }
4420
4421         if (resend_igmp < 0 || resend_igmp > 255) {
4422                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4423                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4424                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4425         }
4426
4427         bond_opt_initval(&newval, packets_per_slave);
4428         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4429                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4430                         packets_per_slave, USHRT_MAX);
4431                 packets_per_slave = 1;
4432         }
4433
4434         if (bond_mode == BOND_MODE_ALB) {
4435                 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",
4436                           updelay);
4437         }
4438
4439         if (!miimon) {
4440                 if (updelay || downdelay) {
4441                         /* just warn the user the up/down delay will have
4442                          * no effect since miimon is zero...
4443                          */
4444                         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",
4445                                 updelay, downdelay);
4446                 }
4447         } else {
4448                 /* don't allow arp monitoring */
4449                 if (arp_interval) {
4450                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4451                                 miimon, arp_interval);
4452                         arp_interval = 0;
4453                 }
4454
4455                 if ((updelay % miimon) != 0) {
4456                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4457                                 updelay, miimon, (updelay / miimon) * miimon);
4458                 }
4459
4460                 updelay /= miimon;
4461
4462                 if ((downdelay % miimon) != 0) {
4463                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4464                                 downdelay, miimon,
4465                                 (downdelay / miimon) * miimon);
4466                 }
4467
4468                 downdelay /= miimon;
4469         }
4470
4471         if (arp_interval < 0) {
4472                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4473                         arp_interval, INT_MAX);
4474                 arp_interval = 0;
4475         }
4476
4477         for (arp_ip_count = 0, i = 0;
4478              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4479                 __be32 ip;
4480
4481                 /* not a complete check, but good enough to catch mistakes */
4482                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4483                     !bond_is_ip_target_ok(ip)) {
4484                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4485                                 arp_ip_target[i]);
4486                         arp_interval = 0;
4487                 } else {
4488                         if (bond_get_targets_ip(arp_target, ip) == -1)
4489                                 arp_target[arp_ip_count++] = ip;
4490                         else
4491                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4492                                         &ip);
4493                 }
4494         }
4495
4496         if (arp_interval && !arp_ip_count) {
4497                 /* don't allow arping if no arp_ip_target given... */
4498                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4499                         arp_interval);
4500                 arp_interval = 0;
4501         }
4502
4503         if (arp_validate) {
4504                 if (!arp_interval) {
4505                         pr_err("arp_validate requires arp_interval\n");
4506                         return -EINVAL;
4507                 }
4508
4509                 bond_opt_initstr(&newval, arp_validate);
4510                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4511                                         &newval);
4512                 if (!valptr) {
4513                         pr_err("Error: invalid arp_validate \"%s\"\n",
4514                                arp_validate);
4515                         return -EINVAL;
4516                 }
4517                 arp_validate_value = valptr->value;
4518         } else {
4519                 arp_validate_value = 0;
4520         }
4521
4522         arp_all_targets_value = 0;
4523         if (arp_all_targets) {
4524                 bond_opt_initstr(&newval, arp_all_targets);
4525                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4526                                         &newval);
4527                 if (!valptr) {
4528                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4529                                arp_all_targets);
4530                         arp_all_targets_value = 0;
4531                 } else {
4532                         arp_all_targets_value = valptr->value;
4533                 }
4534         }
4535
4536         if (miimon) {
4537                 pr_info("MII link monitoring set to %d ms\n", miimon);
4538         } else if (arp_interval) {
4539                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4540                                           arp_validate_value);
4541                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4542                         arp_interval, valptr->string, arp_ip_count);
4543
4544                 for (i = 0; i < arp_ip_count; i++)
4545                         pr_cont(" %s", arp_ip_target[i]);
4546
4547                 pr_cont("\n");
4548
4549         } else if (max_bonds) {
4550                 /* miimon and arp_interval not set, we need one so things
4551                  * work as expected, see bonding.txt for details
4552                  */
4553                 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");
4554         }
4555
4556         if (primary && !bond_mode_uses_primary(bond_mode)) {
4557                 /* currently, using a primary only makes sense
4558                  * in active backup, TLB or ALB modes
4559                  */
4560                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4561                         primary, bond_mode_name(bond_mode));
4562                 primary = NULL;
4563         }
4564
4565         if (primary && primary_reselect) {
4566                 bond_opt_initstr(&newval, primary_reselect);
4567                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4568                                         &newval);
4569                 if (!valptr) {
4570                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4571                                primary_reselect);
4572                         return -EINVAL;
4573                 }
4574                 primary_reselect_value = valptr->value;
4575         } else {
4576                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4577         }
4578
4579         if (fail_over_mac) {
4580                 bond_opt_initstr(&newval, fail_over_mac);
4581                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4582                                         &newval);
4583                 if (!valptr) {
4584                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4585                                fail_over_mac);
4586                         return -EINVAL;
4587                 }
4588                 fail_over_mac_value = valptr->value;
4589                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4590                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4591         } else {
4592                 fail_over_mac_value = BOND_FOM_NONE;
4593         }
4594
4595         bond_opt_initstr(&newval, "default");
4596         valptr = bond_opt_parse(
4597                         bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4598                                      &newval);
4599         if (!valptr) {
4600                 pr_err("Error: No ad_actor_sys_prio default value");
4601                 return -EINVAL;
4602         }
4603         ad_actor_sys_prio = valptr->value;
4604
4605         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4606                                 &newval);
4607         if (!valptr) {
4608                 pr_err("Error: No ad_user_port_key default value");
4609                 return -EINVAL;
4610         }
4611         ad_user_port_key = valptr->value;
4612
4613         if (lp_interval == 0) {
4614                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4615                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4616                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4617         }
4618
4619         /* fill params struct with the proper values */
4620         params->mode = bond_mode;
4621         params->xmit_policy = xmit_hashtype;
4622         params->miimon = miimon;
4623         params->num_peer_notif = num_peer_notif;
4624         params->arp_interval = arp_interval;
4625         params->arp_validate = arp_validate_value;
4626         params->arp_all_targets = arp_all_targets_value;
4627         params->updelay = updelay;
4628         params->downdelay = downdelay;
4629         params->use_carrier = use_carrier;
4630         params->lacp_fast = lacp_fast;
4631         params->primary[0] = 0;
4632         params->primary_reselect = primary_reselect_value;
4633         params->fail_over_mac = fail_over_mac_value;
4634         params->tx_queues = tx_queues;
4635         params->all_slaves_active = all_slaves_active;
4636         params->resend_igmp = resend_igmp;
4637         params->min_links = min_links;
4638         params->lp_interval = lp_interval;
4639         params->packets_per_slave = packets_per_slave;
4640         params->tlb_dynamic_lb = 1; /* Default value */
4641         params->ad_actor_sys_prio = ad_actor_sys_prio;
4642         eth_zero_addr(params->ad_actor_system);
4643         params->ad_user_port_key = ad_user_port_key;
4644         if (packets_per_slave > 0) {
4645                 params->reciprocal_packets_per_slave =
4646                         reciprocal_value(packets_per_slave);
4647         } else {
4648                 /* reciprocal_packets_per_slave is unused if
4649                  * packets_per_slave is 0 or 1, just initialize it
4650                  */
4651                 params->reciprocal_packets_per_slave =
4652                         (struct reciprocal_value) { 0 };
4653         }
4654
4655         if (primary) {
4656                 strncpy(params->primary, primary, IFNAMSIZ);
4657                 params->primary[IFNAMSIZ - 1] = 0;
4658         }
4659
4660         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4661
4662         return 0;
4663 }
4664
4665 /* Called from registration process */
4666 static int bond_init(struct net_device *bond_dev)
4667 {
4668         struct bonding *bond = netdev_priv(bond_dev);
4669         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4670
4671         netdev_dbg(bond_dev, "Begin bond_init\n");
4672
4673         bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
4674         if (!bond->wq)
4675                 return -ENOMEM;
4676
4677         bond->nest_level = SINGLE_DEPTH_NESTING;
4678         netdev_lockdep_set_classes(bond_dev);
4679
4680         list_add_tail(&bond->bond_list, &bn->dev_list);
4681
4682         bond_prepare_sysfs_group(bond);
4683
4684         bond_debug_register(bond);
4685
4686         /* Ensure valid dev_addr */
4687         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4688             bond_dev->addr_assign_type == NET_ADDR_PERM)
4689                 eth_hw_addr_random(bond_dev);
4690
4691         return 0;
4692 }
4693
4694 unsigned int bond_get_num_tx_queues(void)
4695 {
4696         return tx_queues;
4697 }
4698
4699 /* Create a new bond based on the specified name and bonding parameters.
4700  * If name is NULL, obtain a suitable "bond%d" name for us.
4701  * Caller must NOT hold rtnl_lock; we need to release it here before we
4702  * set up our sysfs entries.
4703  */
4704 int bond_create(struct net *net, const char *name)
4705 {
4706         struct net_device *bond_dev;
4707         struct bonding *bond;
4708         struct alb_bond_info *bond_info;
4709         int res;
4710
4711         rtnl_lock();
4712
4713         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4714                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
4715                                    bond_setup, tx_queues);
4716         if (!bond_dev) {
4717                 pr_err("%s: eek! can't alloc netdev!\n", name);
4718                 rtnl_unlock();
4719                 return -ENOMEM;
4720         }
4721
4722         /*
4723          * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4724          * It is set to 0 by default which is wrong.
4725          */
4726         bond = netdev_priv(bond_dev);
4727         bond_info = &(BOND_ALB_INFO(bond));
4728         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4729
4730         dev_net_set(bond_dev, net);
4731         bond_dev->rtnl_link_ops = &bond_link_ops;
4732
4733         res = register_netdevice(bond_dev);
4734
4735         netif_carrier_off(bond_dev);
4736
4737         rtnl_unlock();
4738         if (res < 0)
4739                 bond_destructor(bond_dev);
4740         return res;
4741 }
4742
4743 static int __net_init bond_net_init(struct net *net)
4744 {
4745         struct bond_net *bn = net_generic(net, bond_net_id);
4746
4747         bn->net = net;
4748         INIT_LIST_HEAD(&bn->dev_list);
4749
4750         bond_create_proc_dir(bn);
4751         bond_create_sysfs(bn);
4752
4753         return 0;
4754 }
4755
4756 static void __net_exit bond_net_exit(struct net *net)
4757 {
4758         struct bond_net *bn = net_generic(net, bond_net_id);
4759         struct bonding *bond, *tmp_bond;
4760         LIST_HEAD(list);
4761
4762         bond_destroy_sysfs(bn);
4763
4764         /* Kill off any bonds created after unregistering bond rtnl ops */
4765         rtnl_lock();
4766         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4767                 unregister_netdevice_queue(bond->dev, &list);
4768         unregister_netdevice_many(&list);
4769         rtnl_unlock();
4770
4771         bond_destroy_proc_dir(bn);
4772 }
4773
4774 static struct pernet_operations bond_net_ops = {
4775         .init = bond_net_init,
4776         .exit = bond_net_exit,
4777         .id   = &bond_net_id,
4778         .size = sizeof(struct bond_net),
4779 };
4780
4781 static int __init bonding_init(void)
4782 {
4783         int i;
4784         int res;
4785
4786         pr_info("%s", bond_version);
4787
4788         res = bond_check_params(&bonding_defaults);
4789         if (res)
4790                 goto out;
4791
4792         res = register_pernet_subsys(&bond_net_ops);
4793         if (res)
4794                 goto out;
4795
4796         res = bond_netlink_init();
4797         if (res)
4798                 goto err_link;
4799
4800         bond_create_debugfs();
4801
4802         for (i = 0; i < max_bonds; i++) {
4803                 res = bond_create(&init_net, NULL);
4804                 if (res)
4805                         goto err;
4806         }
4807
4808         register_netdevice_notifier(&bond_netdev_notifier);
4809 out:
4810         return res;
4811 err:
4812         bond_destroy_debugfs();
4813         bond_netlink_fini();
4814 err_link:
4815         unregister_pernet_subsys(&bond_net_ops);
4816         goto out;
4817
4818 }
4819
4820 static void __exit bonding_exit(void)
4821 {
4822         unregister_netdevice_notifier(&bond_netdev_notifier);
4823
4824         bond_destroy_debugfs();
4825
4826         bond_netlink_fini();
4827         unregister_pernet_subsys(&bond_net_ops);
4828
4829 #ifdef CONFIG_NET_POLL_CONTROLLER
4830         /* Make sure we don't have an imbalance on our netpoll blocking */
4831         WARN_ON(atomic_read(&netpoll_block_tx));
4832 #endif
4833 }
4834
4835 module_init(bonding_init);
4836 module_exit(bonding_exit);
4837 MODULE_LICENSE("GPL");
4838 MODULE_VERSION(DRV_VERSION);
4839 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4840 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");