drivers/net/bonding/bond_main.c

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