drivers/net/bonding/bond_main.c

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 #include <linux/kernel.h>
  35 #include <linux/module.h>
  36 #include <linux/types.h>
  37 #include <linux/fcntl.h>
  38 #include <linux/interrupt.h>
  39 #include <linux/ptrace.h>
  40 #include <linux/ioport.h>
  41 #include <linux/in.h>
  42 #include <net/ip.h>
  43 #include <linux/ip.h>
  44 #include <linux/tcp.h>
  45 #include <linux/udp.h>
  46 #include <linux/slab.h>
  47 #include <linux/string.h>
  48 #include <linux/init.h>
  49 #include <linux/timer.h>
  50 #include <linux/socket.h>
  51 #include <linux/ctype.h>
  52 #include <linux/inet.h>
  53 #include <linux/bitops.h>
  54 #include <linux/io.h>
  55 #include <asm/dma.h>
  56 #include <linux/uaccess.h>
  57 #include <linux/errno.h>
  58 #include <linux/netdevice.h>
  59 #include <linux/inetdevice.h>
  60 #include <linux/igmp.h>
  61 #include <linux/etherdevice.h>
  62 #include <linux/skbuff.h>
  63 #include <net/sock.h>
  64 #include <linux/rtnetlink.h>
  65 #include <linux/smp.h>
  66 #include <linux/if_ether.h>
  67 #include <net/arp.h>
  68 #include <linux/mii.h>
  69 #include <linux/ethtool.h>
  70 #include <linux/if_vlan.h>
  71 #include <linux/if_bonding.h>
  72 #include <linux/jiffies.h>
  73 #include <linux/preempt.h>
  74 #include <net/route.h>
  75 #include <net/net_namespace.h>
  76 #include <net/netns/generic.h>
  77 #include <net/pkt_sched.h>
  78 #include <linux/rculist.h>
  79 #include <net/flow_dissector.h>
  80 #include <net/switchdev.h>
  81 #include <net/bonding.h>
  82 #include <net/bond_3ad.h>
  83 #include <net/bond_alb.h>
  84
  85 #include "bonding_priv.h"
  86
  87 /*---------------------------- Module parameters ----------------------------*/
  88
  89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  90
  91 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  92 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  93 static int num_peer_notif = 1;
  94 static int miimon;
  95 static int updelay;
  96 static int downdelay;
  97 static int use_carrier  = 1;
  98 static char *mode;
  99 static char *primary;
 100 static char *primary_reselect;
 101 static char *lacp_rate;
 102 static int min_links;
 103 static char *ad_select;
 104 static char *xmit_hash_policy;
 105 static int arp_interval;
 106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 107 static char *arp_validate;
 108 static char *arp_all_targets;
 109 static char *fail_over_mac;
 110 static int all_slaves_active;
 111 static struct bond_params bonding_defaults;
 112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 113 static int packets_per_slave = 1;
 114 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
 115
 116 module_param(max_bonds, int, 0);
 117 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 118 module_param(tx_queues, int, 0);
 119 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 120 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 121 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 122                                "failover event (alias of num_unsol_na)");
 123 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 124 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 125                                "failover event (alias of num_grat_arp)");
 126 module_param(miimon, int, 0);
 127 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 128 module_param(updelay, int, 0);
 129 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 130 module_param(downdelay, int, 0);
 131 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 132                             "in milliseconds");
 133 module_param(use_carrier, int, 0);
 134 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 135                               "0 for off, 1 for on (default)");
 136 module_param(mode, charp, 0);
 137 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 138                        "1 for active-backup, 2 for balance-xor, "
 139                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 140                        "6 for balance-alb");
 141 module_param(primary, charp, 0);
 142 MODULE_PARM_DESC(primary, "Primary network device to use");
 143 module_param(primary_reselect, charp, 0);
 144 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 145                                    "once it comes up; "
 146                                    "0 for always (default), "
 147                                    "1 for only if speed of primary is "
 148                                    "better, "
 149                                    "2 for only on active slave "
 150                                    "failure");
 151 module_param(lacp_rate, charp, 0);
 152 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 153                             "0 for slow, 1 for fast");
 154 module_param(ad_select, charp, 0);
 155 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
 156                             "0 for stable (default), 1 for bandwidth, "
 157                             "2 for count");
 158 module_param(min_links, int, 0);
 159 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 160
 161 module_param(xmit_hash_policy, charp, 0);
 162 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 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 unsigned int bond_net_id __read_mostly;
 203
 204 /*-------------------------- Forward declarations ---------------------------*/
 205
 206 static int bond_init(struct net_device *bond_dev);
 207 static void bond_uninit(struct net_device *bond_dev);
 208 static void bond_get_stats(struct net_device *bond_dev,
 209                            struct rtnl_link_stats64 *stats);
 210 static void bond_slave_arr_handler(struct work_struct *work);
 211 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
 212                                   int mod);
 213 static void bond_netdev_notify_work(struct work_struct *work);
 214
 215 /*---------------------------- General routines -----------------------------*/
 216
 217 const char *bond_mode_name(int mode)
 218 {
 219         static const char *names[] = {
 220                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 221                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 222                 [BOND_MODE_XOR] = "load balancing (xor)",
 223                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 224                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 225                 [BOND_MODE_TLB] = "transmit load balancing",
 226                 [BOND_MODE_ALB] = "adaptive load balancing",
 227         };
 228
 229         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 230                 return "unknown";
 231
 232         return names[mode];
 233 }
 234
 235 /*---------------------------------- VLAN -----------------------------------*/
 236
 237 /**
 238  * bond_dev_queue_xmit - Prepare skb for xmit.
 239  *
 240  * @bond: bond device that got this skb for tx.
 241  * @skb: hw accel VLAN tagged skb to transmit
 242  * @slave_dev: slave that is supposed to xmit this skbuff
 243  */
 244 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 245                         struct net_device *slave_dev)
 246 {
 247         skb->dev = slave_dev;
 248
 249         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 250                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 251         skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
 252
 253         if (unlikely(netpoll_tx_running(bond->dev)))
 254                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 255         else
 256                 dev_queue_xmit(skb);
 257 }
 258
 259 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 260  * We don't protect the slave list iteration with a lock because:
 261  * a. This operation is performed in IOCTL context,
 262  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 263  * c. Holding a lock with BH disabled while directly calling a base driver
 264  *    entry point is generally a BAD idea.
 265  *
 266  * The design of synchronization/protection for this operation in the 8021q
 267  * module is good for one or more VLAN devices over a single physical device
 268  * and cannot be extended for a teaming solution like bonding, so there is a
 269  * potential race condition here where a net device from the vlan group might
 270  * be referenced (either by a base driver or the 8021q code) while it is being
 271  * removed from the system. However, it turns out we're not making matters
 272  * worse, and if it works for regular VLAN usage it will work here too.
 273 */
 274
 275 /**
 276  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 277  * @bond_dev: bonding net device that got called
 278  * @vid: vlan id being added
 279  */
 280 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 281                                 __be16 proto, u16 vid)
 282 {
 283         struct bonding *bond = netdev_priv(bond_dev);
 284         struct slave *slave, *rollback_slave;
 285         struct list_head *iter;
 286         int res;
 287
 288         bond_for_each_slave(bond, slave, iter) {
 289                 res = vlan_vid_add(slave->dev, proto, vid);
 290                 if (res)
 291                         goto unwind;
 292         }
 293
 294         return 0;
 295
 296 unwind:
 297         /* unwind to the slave that failed */
 298         bond_for_each_slave(bond, rollback_slave, iter) {
 299                 if (rollback_slave == slave)
 300                         break;
 301
 302                 vlan_vid_del(rollback_slave->dev, proto, vid);
 303         }
 304
 305         return res;
 306 }
 307
 308 /**
 309  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 310  * @bond_dev: bonding net device that got called
 311  * @vid: vlan id being removed
 312  */
 313 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 314                                  __be16 proto, u16 vid)
 315 {
 316         struct bonding *bond = netdev_priv(bond_dev);
 317         struct list_head *iter;
 318         struct slave *slave;
 319
 320         bond_for_each_slave(bond, slave, iter)
 321                 vlan_vid_del(slave->dev, proto, vid);
 322
 323         if (bond_is_lb(bond))
 324                 bond_alb_clear_vlan(bond, vid);
 325
 326         return 0;
 327 }
 328
 329 /*------------------------------- Link status -------------------------------*/
 330
 331 /* Set the carrier state for the master according to the state of its
 332  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 333  * do special 802.3ad magic.
 334  *
 335  * Returns zero if carrier state does not change, nonzero if it does.
 336  */
 337 int bond_set_carrier(struct bonding *bond)
 338 {
 339         struct list_head *iter;
 340         struct slave *slave;
 341
 342         if (!bond_has_slaves(bond))
 343                 goto down;
 344
 345         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 346                 return bond_3ad_set_carrier(bond);
 347
 348         bond_for_each_slave(bond, slave, iter) {
 349                 if (slave->link == BOND_LINK_UP) {
 350                         if (!netif_carrier_ok(bond->dev)) {
 351                                 netif_carrier_on(bond->dev);
 352                                 return 1;
 353                         }
 354                         return 0;
 355                 }
 356         }
 357
 358 down:
 359         if (netif_carrier_ok(bond->dev)) {
 360                 netif_carrier_off(bond->dev);
 361                 return 1;
 362         }
 363         return 0;
 364 }
 365
 366 /* Get link speed and duplex from the slave's base driver
 367  * using ethtool. If for some reason the call fails or the
 368  * values are invalid, set speed and duplex to -1,
 369  * and return. Return 1 if speed or duplex settings are
 370  * UNKNOWN; 0 otherwise.
 371  */
 372 static int bond_update_speed_duplex(struct slave *slave)
 373 {
 374         struct net_device *slave_dev = slave->dev;
 375         struct ethtool_link_ksettings ecmd;
 376         int res;
 377
 378         slave->speed = SPEED_UNKNOWN;
 379         slave->duplex = DUPLEX_UNKNOWN;
 380
 381         res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
 382         if (res < 0)
 383                 return 1;
 384         if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
 385                 return 1;
 386         switch (ecmd.base.duplex) {
 387         case DUPLEX_FULL:
 388         case DUPLEX_HALF:
 389                 break;
 390         default:
 391                 return 1;
 392         }
 393
 394         slave->speed = ecmd.base.speed;
 395         slave->duplex = ecmd.base.duplex;
 396
 397         return 0;
 398 }
 399
 400 const char *bond_slave_link_status(s8 link)
 401 {
 402         switch (link) {
 403         case BOND_LINK_UP:
 404                 return "up";
 405         case BOND_LINK_FAIL:
 406                 return "going down";
 407         case BOND_LINK_DOWN:
 408                 return "down";
 409         case BOND_LINK_BACK:
 410                 return "going back";
 411         default:
 412                 return "unknown";
 413         }
 414 }
 415
 416 /* if <dev> supports MII link status reporting, check its link status.
 417  *
 418  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 419  * depending upon the setting of the use_carrier parameter.
 420  *
 421  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 422  * can't tell and just pretend it is), or 0, meaning that the link is
 423  * down.
 424  *
 425  * If reporting is non-zero, instead of faking link up, return -1 if
 426  * both ETHTOOL and MII ioctls fail (meaning the device does not
 427  * support them).  If use_carrier is set, return whatever it says.
 428  * It'd be nice if there was a good way to tell if a driver supports
 429  * netif_carrier, but there really isn't.
 430  */
 431 static int bond_check_dev_link(struct bonding *bond,
 432                                struct net_device *slave_dev, int reporting)
 433 {
 434         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 435         int (*ioctl)(struct net_device *, struct ifreq *, int);
 436         struct ifreq ifr;
 437         struct mii_ioctl_data *mii;
 438
 439         if (!reporting && !netif_running(slave_dev))
 440                 return 0;
 441
 442         if (bond->params.use_carrier)
 443                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 444
 445         /* Try to get link status using Ethtool first. */
 446         if (slave_dev->ethtool_ops->get_link)
 447                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 448                         BMSR_LSTATUS : 0;
 449
 450         /* Ethtool can't be used, fallback to MII ioctls. */
 451         ioctl = slave_ops->ndo_do_ioctl;
 452         if (ioctl) {
 453                 /* TODO: set pointer to correct ioctl on a per team member
 454                  *       bases to make this more efficient. that is, once
 455                  *       we determine the correct ioctl, we will always
 456                  *       call it and not the others for that team
 457                  *       member.
 458                  */
 459
 460                 /* We cannot assume that SIOCGMIIPHY will also read a
 461                  * register; not all network drivers (e.g., e100)
 462                  * support that.
 463                  */
 464
 465                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 466                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 467                 mii = if_mii(&ifr);
 468                 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 469                         mii->reg_num = MII_BMSR;
 470                         if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
 471                                 return mii->val_out & BMSR_LSTATUS;
 472                 }
 473         }
 474
 475         /* If reporting, report that either there's no dev->do_ioctl,
 476          * or both SIOCGMIIREG and get_link failed (meaning that we
 477          * cannot report link status).  If not reporting, pretend
 478          * we're ok.
 479          */
 480         return reporting ? -1 : BMSR_LSTATUS;
 481 }
 482
 483 /*----------------------------- Multicast list ------------------------------*/
 484
 485 /* Push the promiscuity flag down to appropriate slaves */
 486 static int bond_set_promiscuity(struct bonding *bond, int inc)
 487 {
 488         struct list_head *iter;
 489         int err = 0;
 490
 491         if (bond_uses_primary(bond)) {
 492                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 493
 494                 if (curr_active)
 495                         err = dev_set_promiscuity(curr_active->dev, inc);
 496         } else {
 497                 struct slave *slave;
 498
 499                 bond_for_each_slave(bond, slave, iter) {
 500                         err = dev_set_promiscuity(slave->dev, inc);
 501                         if (err)
 502                                 return err;
 503                 }
 504         }
 505         return err;
 506 }
 507
 508 /* Push the allmulti flag down to all slaves */
 509 static int bond_set_allmulti(struct bonding *bond, int inc)
 510 {
 511         struct list_head *iter;
 512         int err = 0;
 513
 514         if (bond_uses_primary(bond)) {
 515                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 516
 517                 if (curr_active)
 518                         err = dev_set_allmulti(curr_active->dev, inc);
 519         } else {
 520                 struct slave *slave;
 521
 522                 bond_for_each_slave(bond, slave, iter) {
 523                         err = dev_set_allmulti(slave->dev, inc);
 524                         if (err)
 525                                 return err;
 526                 }
 527         }
 528         return err;
 529 }
 530
 531 /* Retrieve the list of registered multicast addresses for the bonding
 532  * device and retransmit an IGMP JOIN request to the current active
 533  * slave.
 534  */
 535 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 536 {
 537         struct bonding *bond = container_of(work, struct bonding,
 538                                             mcast_work.work);
 539
 540         if (!rtnl_trylock()) {
 541                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 542                 return;
 543         }
 544         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 545
 546         if (bond->igmp_retrans > 1) {
 547                 bond->igmp_retrans--;
 548                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 549         }
 550         rtnl_unlock();
 551 }
 552
 553 /* Flush bond's hardware addresses from slave */
 554 static void bond_hw_addr_flush(struct net_device *bond_dev,
 555                                struct net_device *slave_dev)
 556 {
 557         struct bonding *bond = netdev_priv(bond_dev);
 558
 559         dev_uc_unsync(slave_dev, bond_dev);
 560         dev_mc_unsync(slave_dev, bond_dev);
 561
 562         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 563                 /* del lacpdu mc addr from mc list */
 564                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 565
 566                 dev_mc_del(slave_dev, lacpdu_multicast);
 567         }
 568 }
 569
 570 /*--------------------------- Active slave change ---------------------------*/
 571
 572 /* Update the hardware address list and promisc/allmulti for the new and
 573  * old active slaves (if any).  Modes that are not using primary keep all
 574  * slaves up date at all times; only the modes that use primary need to call
 575  * this function to swap these settings during a failover.
 576  */
 577 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 578                               struct slave *old_active)
 579 {
 580         if (old_active) {
 581                 if (bond->dev->flags & IFF_PROMISC)
 582                         dev_set_promiscuity(old_active->dev, -1);
 583
 584                 if (bond->dev->flags & IFF_ALLMULTI)
 585                         dev_set_allmulti(old_active->dev, -1);
 586
 587                 bond_hw_addr_flush(bond->dev, old_active->dev);
 588         }
 589
 590         if (new_active) {
 591                 /* FIXME: Signal errors upstream. */
 592                 if (bond->dev->flags & IFF_PROMISC)
 593                         dev_set_promiscuity(new_active->dev, 1);
 594
 595                 if (bond->dev->flags & IFF_ALLMULTI)
 596                         dev_set_allmulti(new_active->dev, 1);
 597
 598                 netif_addr_lock_bh(bond->dev);
 599                 dev_uc_sync(new_active->dev, bond->dev);
 600                 dev_mc_sync(new_active->dev, bond->dev);
 601                 netif_addr_unlock_bh(bond->dev);
 602         }
 603 }
 604
 605 /**
 606  * bond_set_dev_addr - clone slave's address to bond
 607  * @bond_dev: bond net device
 608  * @slave_dev: slave net device
 609  *
 610  * Should be called with RTNL held.
 611  */
 612 static void bond_set_dev_addr(struct net_device *bond_dev,
 613                               struct net_device *slave_dev)
 614 {
 615         netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n",
 616                    bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len);
 617         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 618         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 619         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 620 }
 621
 622 static struct slave *bond_get_old_active(struct bonding *bond,
 623                                          struct slave *new_active)
 624 {
 625         struct slave *slave;
 626         struct list_head *iter;
 627
 628         bond_for_each_slave(bond, slave, iter) {
 629                 if (slave == new_active)
 630                         continue;
 631
 632                 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
 633                         return slave;
 634         }
 635
 636         return NULL;
 637 }
 638
 639 /* bond_do_fail_over_mac
 640  *
 641  * Perform special MAC address swapping for fail_over_mac settings
 642  *
 643  * Called with RTNL
 644  */
 645 static void bond_do_fail_over_mac(struct bonding *bond,
 646                                   struct slave *new_active,
 647                                   struct slave *old_active)
 648 {
 649         u8 tmp_mac[MAX_ADDR_LEN];
 650         struct sockaddr_storage ss;
 651         int rv;
 652
 653         switch (bond->params.fail_over_mac) {
 654         case BOND_FOM_ACTIVE:
 655                 if (new_active)
 656                         bond_set_dev_addr(bond->dev, new_active->dev);
 657                 break;
 658         case BOND_FOM_FOLLOW:
 659                 /* if new_active && old_active, swap them
 660                  * if just old_active, do nothing (going to no active slave)
 661                  * if just new_active, set new_active to bond's MAC
 662                  */
 663                 if (!new_active)
 664                         return;
 665
 666                 if (!old_active)
 667                         old_active = bond_get_old_active(bond, new_active);
 668
 669                 if (old_active) {
 670                         bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
 671                                           new_active->dev->addr_len);
 672                         bond_hw_addr_copy(ss.__data,
 673                                           old_active->dev->dev_addr,
 674                                           old_active->dev->addr_len);
 675                         ss.ss_family = new_active->dev->type;
 676                 } else {
 677                         bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
 678                                           bond->dev->addr_len);
 679                         ss.ss_family = bond->dev->type;
 680                 }
 681
 682                 rv = dev_set_mac_address(new_active->dev,
 683                                          (struct sockaddr *)&ss);
 684                 if (rv) {
 685                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 686                                    -rv, new_active->dev->name);
 687                         goto out;
 688                 }
 689
 690                 if (!old_active)
 691                         goto out;
 692
 693                 bond_hw_addr_copy(ss.__data, tmp_mac,
 694                                   new_active->dev->addr_len);
 695                 ss.ss_family = old_active->dev->type;
 696
 697                 rv = dev_set_mac_address(old_active->dev,
 698                                          (struct sockaddr *)&ss);
 699                 if (rv)
 700                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 701                                    -rv, new_active->dev->name);
 702 out:
 703                 break;
 704         default:
 705                 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
 706                            bond->params.fail_over_mac);
 707                 break;
 708         }
 709
 710 }
 711
 712 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
 713 {
 714         struct slave *prim = rtnl_dereference(bond->primary_slave);
 715         struct slave *curr = rtnl_dereference(bond->curr_active_slave);
 716
 717         if (!prim || prim->link != BOND_LINK_UP) {
 718                 if (!curr || curr->link != BOND_LINK_UP)
 719                         return NULL;
 720                 return curr;
 721         }
 722
 723         if (bond->force_primary) {
 724                 bond->force_primary = false;
 725                 return prim;
 726         }
 727
 728         if (!curr || curr->link != BOND_LINK_UP)
 729                 return prim;
 730
 731         /* At this point, prim and curr are both up */
 732         switch (bond->params.primary_reselect) {
 733         case BOND_PRI_RESELECT_ALWAYS:
 734                 return prim;
 735         case BOND_PRI_RESELECT_BETTER:
 736                 if (prim->speed < curr->speed)
 737                         return curr;
 738                 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
 739                         return curr;
 740                 return prim;
 741         case BOND_PRI_RESELECT_FAILURE:
 742                 return curr;
 743         default:
 744                 netdev_err(bond->dev, "impossible primary_reselect %d\n",
 745                            bond->params.primary_reselect);
 746                 return curr;
 747         }
 748 }
 749
 750 /**
 751  * bond_find_best_slave - select the best available slave to be the active one
 752  * @bond: our bonding struct
 753  */
 754 static struct slave *bond_find_best_slave(struct bonding *bond)
 755 {
 756         struct slave *slave, *bestslave = NULL;
 757         struct list_head *iter;
 758         int mintime = bond->params.updelay;
 759
 760         slave = bond_choose_primary_or_current(bond);
 761         if (slave)
 762                 return slave;
 763
 764         bond_for_each_slave(bond, slave, iter) {
 765                 if (slave->link == BOND_LINK_UP)
 766                         return slave;
 767                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
 768                     slave->delay < mintime) {
 769                         mintime = slave->delay;
 770                         bestslave = slave;
 771                 }
 772         }
 773
 774         return bestslave;
 775 }
 776
 777 static bool bond_should_notify_peers(struct bonding *bond)
 778 {
 779         struct slave *slave;
 780
 781         rcu_read_lock();
 782         slave = rcu_dereference(bond->curr_active_slave);
 783         rcu_read_unlock();
 784
 785         netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
 786                    slave ? slave->dev->name : "NULL");
 787
 788         if (!slave || !bond->send_peer_notif ||
 789             !netif_carrier_ok(bond->dev) ||
 790             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 791                 return false;
 792
 793         return true;
 794 }
 795
 796 /**
 797  * change_active_interface - change the active slave into the specified one
 798  * @bond: our bonding struct
 799  * @new: the new slave to make the active one
 800  *
 801  * Set the new slave to the bond's settings and unset them on the old
 802  * curr_active_slave.
 803  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 804  *
 805  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 806  * because it is apparently the best available slave we have, even though its
 807  * updelay hasn't timed out yet.
 808  *
 809  * Caller must hold RTNL.
 810  */
 811 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 812 {
 813         struct slave *old_active;
 814
 815         ASSERT_RTNL();
 816
 817         old_active = rtnl_dereference(bond->curr_active_slave);
 818
 819         if (old_active == new_active)
 820                 return;
 821
 822         if (new_active) {
 823                 new_active->last_link_up = jiffies;
 824
 825                 if (new_active->link == BOND_LINK_BACK) {
 826                         if (bond_uses_primary(bond)) {
 827                                 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
 828                                             new_active->dev->name,
 829                                             (bond->params.updelay - new_active->delay) * bond->params.miimon);
 830                         }
 831
 832                         new_active->delay = 0;
 833                         bond_set_slave_link_state(new_active, BOND_LINK_UP,
 834                                                   BOND_SLAVE_NOTIFY_NOW);
 835
 836                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 837                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 838
 839                         if (bond_is_lb(bond))
 840                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 841                 } else {
 842                         if (bond_uses_primary(bond)) {
 843                                 netdev_info(bond->dev, "making interface %s the new active one\n",
 844                                             new_active->dev->name);
 845                         }
 846                 }
 847         }
 848
 849         if (bond_uses_primary(bond))
 850                 bond_hw_addr_swap(bond, new_active, old_active);
 851
 852         if (bond_is_lb(bond)) {
 853                 bond_alb_handle_active_change(bond, new_active);
 854                 if (old_active)
 855                         bond_set_slave_inactive_flags(old_active,
 856                                                       BOND_SLAVE_NOTIFY_NOW);
 857                 if (new_active)
 858                         bond_set_slave_active_flags(new_active,
 859                                                     BOND_SLAVE_NOTIFY_NOW);
 860         } else {
 861                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 862         }
 863
 864         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
 865                 if (old_active)
 866                         bond_set_slave_inactive_flags(old_active,
 867                                                       BOND_SLAVE_NOTIFY_NOW);
 868
 869                 if (new_active) {
 870                         bool should_notify_peers = false;
 871
 872                         bond_set_slave_active_flags(new_active,
 873                                                     BOND_SLAVE_NOTIFY_NOW);
 874
 875                         if (bond->params.fail_over_mac)
 876                                 bond_do_fail_over_mac(bond, new_active,
 877                                                       old_active);
 878
 879                         if (netif_running(bond->dev)) {
 880                                 bond->send_peer_notif =
 881                                         bond->params.num_peer_notif;
 882                                 should_notify_peers =
 883                                         bond_should_notify_peers(bond);
 884                         }
 885
 886                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 887                         if (should_notify_peers)
 888                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 889                                                          bond->dev);
 890                 }
 891         }
 892
 893         /* resend IGMP joins since active slave has changed or
 894          * all were sent on curr_active_slave.
 895          * resend only if bond is brought up with the affected
 896          * bonding modes and the retransmission is enabled
 897          */
 898         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 899             ((bond_uses_primary(bond) && new_active) ||
 900              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
 901                 bond->igmp_retrans = bond->params.resend_igmp;
 902                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 903         }
 904 }
 905
 906 /**
 907  * bond_select_active_slave - select a new active slave, if needed
 908  * @bond: our bonding struct
 909  *
 910  * This functions should be called when one of the following occurs:
 911  * - The old curr_active_slave has been released or lost its link.
 912  * - The primary_slave has got its link back.
 913  * - A slave has got its link back and there's no old curr_active_slave.
 914  *
 915  * Caller must hold RTNL.
 916  */
 917 void bond_select_active_slave(struct bonding *bond)
 918 {
 919         struct slave *best_slave;
 920         int rv;
 921
 922         ASSERT_RTNL();
 923
 924         best_slave = bond_find_best_slave(bond);
 925         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
 926                 bond_change_active_slave(bond, best_slave);
 927                 rv = bond_set_carrier(bond);
 928                 if (!rv)
 929                         return;
 930
 931                 if (netif_carrier_ok(bond->dev))
 932                         netdev_info(bond->dev, "first active interface up!\n");
 933                 else
 934                         netdev_info(bond->dev, "now running without any active interface!\n");
 935         }
 936 }
 937
 938 #ifdef CONFIG_NET_POLL_CONTROLLER
 939 static inline int slave_enable_netpoll(struct slave *slave)
 940 {
 941         struct netpoll *np;
 942         int err = 0;
 943
 944         np = kzalloc(sizeof(*np), GFP_KERNEL);
 945         err = -ENOMEM;
 946         if (!np)
 947                 goto out;
 948
 949         err = __netpoll_setup(np, slave->dev);
 950         if (err) {
 951                 kfree(np);
 952                 goto out;
 953         }
 954         slave->np = np;
 955 out:
 956         return err;
 957 }
 958 static inline void slave_disable_netpoll(struct slave *slave)
 959 {
 960         struct netpoll *np = slave->np;
 961
 962         if (!np)
 963                 return;
 964
 965         slave->np = NULL;
 966         __netpoll_free_async(np);
 967 }
 968
 969 static void bond_poll_controller(struct net_device *bond_dev)
 970 {
 971         struct bonding *bond = netdev_priv(bond_dev);
 972         struct slave *slave = NULL;
 973         struct list_head *iter;
 974         struct ad_info ad_info;
 975
 976         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 977                 if (bond_3ad_get_active_agg_info(bond, &ad_info))
 978                         return;
 979
 980         bond_for_each_slave_rcu(bond, slave, iter) {
 981                 if (!bond_slave_is_up(slave))
 982                         continue;
 983
 984                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 985                         struct aggregator *agg =
 986                             SLAVE_AD_INFO(slave)->port.aggregator;
 987
 988                         if (agg &&
 989                             agg->aggregator_identifier != ad_info.aggregator_id)
 990                                 continue;
 991                 }
 992
 993                 netpoll_poll_dev(slave->dev);
 994         }
 995 }
 996
 997 static void bond_netpoll_cleanup(struct net_device *bond_dev)
 998 {
 999         struct bonding *bond = netdev_priv(bond_dev);
1000         struct list_head *iter;
1001         struct slave *slave;
1002
1003         bond_for_each_slave(bond, slave, iter)
1004                 if (bond_slave_is_up(slave))
1005                         slave_disable_netpoll(slave);
1006 }
1007
1008 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1009 {
1010         struct bonding *bond = netdev_priv(dev);
1011         struct list_head *iter;
1012         struct slave *slave;
1013         int err = 0;
1014
1015         bond_for_each_slave(bond, slave, iter) {
1016                 err = slave_enable_netpoll(slave);
1017                 if (err) {
1018                         bond_netpoll_cleanup(dev);
1019                         break;
1020                 }
1021         }
1022         return err;
1023 }
1024 #else
1025 static inline int slave_enable_netpoll(struct slave *slave)
1026 {
1027         return 0;
1028 }
1029 static inline void slave_disable_netpoll(struct slave *slave)
1030 {
1031 }
1032 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1033 {
1034 }
1035 #endif
1036
1037 /*---------------------------------- IOCTL ----------------------------------*/
1038
1039 static netdev_features_t bond_fix_features(struct net_device *dev,
1040                                            netdev_features_t features)
1041 {
1042         struct bonding *bond = netdev_priv(dev);
1043         struct list_head *iter;
1044         netdev_features_t mask;
1045         struct slave *slave;
1046
1047         mask = features;
1048
1049         features &= ~NETIF_F_ONE_FOR_ALL;
1050         features |= NETIF_F_ALL_FOR_ALL;
1051
1052         bond_for_each_slave(bond, slave, iter) {
1053                 features = netdev_increment_features(features,
1054                                                      slave->dev->features,
1055                                                      mask);
1056         }
1057         features = netdev_add_tso_features(features, mask);
1058
1059         return features;
1060 }
1061
1062 #define BOND_VLAN_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1063                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1064                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1065
1066 #define BOND_ENC_FEATURES       (NETIF_F_HW_CSUM | NETIF_F_SG | \
1067                                  NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1068
1069 static void bond_compute_features(struct bonding *bond)
1070 {
1071         unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1072                                         IFF_XMIT_DST_RELEASE_PERM;
1073         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1074         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1075         struct net_device *bond_dev = bond->dev;
1076         struct list_head *iter;
1077         struct slave *slave;
1078         unsigned short max_hard_header_len = ETH_HLEN;
1079         unsigned int gso_max_size = GSO_MAX_SIZE;
1080         u16 gso_max_segs = GSO_MAX_SEGS;
1081
1082         if (!bond_has_slaves(bond))
1083                 goto done;
1084         vlan_features &= NETIF_F_ALL_FOR_ALL;
1085
1086         bond_for_each_slave(bond, slave, iter) {
1087                 vlan_features = netdev_increment_features(vlan_features,
1088                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1089
1090                 enc_features = netdev_increment_features(enc_features,
1091                                                          slave->dev->hw_enc_features,
1092                                                          BOND_ENC_FEATURES);
1093                 dst_release_flag &= slave->dev->priv_flags;
1094                 if (slave->dev->hard_header_len > max_hard_header_len)
1095                         max_hard_header_len = slave->dev->hard_header_len;
1096
1097                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1098                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1099         }
1100         bond_dev->hard_header_len = max_hard_header_len;
1101
1102 done:
1103         bond_dev->vlan_features = vlan_features;
1104         bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1105                                     NETIF_F_GSO_UDP_L4;
1106         bond_dev->gso_max_segs = gso_max_segs;
1107         netif_set_gso_max_size(bond_dev, gso_max_size);
1108
1109         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1110         if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1111             dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1112                 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1113
1114         netdev_change_features(bond_dev);
1115 }
1116
1117 static void bond_setup_by_slave(struct net_device *bond_dev,
1118                                 struct net_device *slave_dev)
1119 {
1120         bond_dev->header_ops        = slave_dev->header_ops;
1121
1122         bond_dev->type              = slave_dev->type;
1123         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1124         bond_dev->addr_len          = slave_dev->addr_len;
1125
1126         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1127                 slave_dev->addr_len);
1128 }
1129
1130 /* On bonding slaves other than the currently active slave, suppress
1131  * duplicates except for alb non-mcast/bcast.
1132  */
1133 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1134                                             struct slave *slave,
1135                                             struct bonding *bond)
1136 {
1137         if (bond_is_slave_inactive(slave)) {
1138                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1139                     skb->pkt_type != PACKET_BROADCAST &&
1140                     skb->pkt_type != PACKET_MULTICAST)
1141                         return false;
1142                 return true;
1143         }
1144         return false;
1145 }
1146
1147 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1148 {
1149         struct sk_buff *skb = *pskb;
1150         struct slave *slave;
1151         struct bonding *bond;
1152         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1153                           struct slave *);
1154         int ret = RX_HANDLER_ANOTHER;
1155
1156         skb = skb_share_check(skb, GFP_ATOMIC);
1157         if (unlikely(!skb))
1158                 return RX_HANDLER_CONSUMED;
1159
1160         *pskb = skb;
1161
1162         slave = bond_slave_get_rcu(skb->dev);
1163         bond = slave->bond;
1164
1165         recv_probe = READ_ONCE(bond->recv_probe);
1166         if (recv_probe) {
1167                 ret = recv_probe(skb, bond, slave);
1168                 if (ret == RX_HANDLER_CONSUMED) {
1169                         consume_skb(skb);
1170                         return ret;
1171                 }
1172         }
1173
1174         /*
1175          * For packets determined by bond_should_deliver_exact_match() call to
1176          * be suppressed we want to make an exception for link-local packets.
1177          * This is necessary for e.g. LLDP daemons to be able to monitor
1178          * inactive slave links without being forced to bind to them
1179          * explicitly.
1180          *
1181          * At the same time, packets that are passed to the bonding master
1182          * (including link-local ones) can have their originating interface
1183          * determined via PACKET_ORIGDEV socket option.
1184          */
1185         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1186                 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1187                         return RX_HANDLER_PASS;
1188                 return RX_HANDLER_EXACT;
1189         }
1190
1191         skb->dev = bond->dev;
1192
1193         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1194             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1195             skb->pkt_type == PACKET_HOST) {
1196
1197                 if (unlikely(skb_cow_head(skb,
1198                                           skb->data - skb_mac_header(skb)))) {
1199                         kfree_skb(skb);
1200                         return RX_HANDLER_CONSUMED;
1201                 }
1202                 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1203                                   bond->dev->addr_len);
1204         }
1205
1206         return ret;
1207 }
1208
1209 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1210 {
1211         switch (BOND_MODE(bond)) {
1212         case BOND_MODE_ROUNDROBIN:
1213                 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1214         case BOND_MODE_ACTIVEBACKUP:
1215                 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1216         case BOND_MODE_BROADCAST:
1217                 return NETDEV_LAG_TX_TYPE_BROADCAST;
1218         case BOND_MODE_XOR:
1219         case BOND_MODE_8023AD:
1220                 return NETDEV_LAG_TX_TYPE_HASH;
1221         default:
1222                 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1223         }
1224 }
1225
1226 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1227                                                enum netdev_lag_tx_type type)
1228 {
1229         if (type != NETDEV_LAG_TX_TYPE_HASH)
1230                 return NETDEV_LAG_HASH_NONE;
1231
1232         switch (bond->params.xmit_policy) {
1233         case BOND_XMIT_POLICY_LAYER2:
1234                 return NETDEV_LAG_HASH_L2;
1235         case BOND_XMIT_POLICY_LAYER34:
1236                 return NETDEV_LAG_HASH_L34;
1237         case BOND_XMIT_POLICY_LAYER23:
1238                 return NETDEV_LAG_HASH_L23;
1239         case BOND_XMIT_POLICY_ENCAP23:
1240                 return NETDEV_LAG_HASH_E23;
1241         case BOND_XMIT_POLICY_ENCAP34:
1242                 return NETDEV_LAG_HASH_E34;
1243         default:
1244                 return NETDEV_LAG_HASH_UNKNOWN;
1245         }
1246 }
1247
1248 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1249                                       struct netlink_ext_ack *extack)
1250 {
1251         struct netdev_lag_upper_info lag_upper_info;
1252         enum netdev_lag_tx_type type;
1253
1254         type = bond_lag_tx_type(bond);
1255         lag_upper_info.tx_type = type;
1256         lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1257
1258         return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1259                                             &lag_upper_info, extack);
1260 }
1261
1262 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1263 {
1264         netdev_upper_dev_unlink(slave->dev, bond->dev);
1265         slave->dev->flags &= ~IFF_SLAVE;
1266 }
1267
1268 static struct slave *bond_alloc_slave(struct bonding *bond)
1269 {
1270         struct slave *slave = NULL;
1271
1272         slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1273         if (!slave)
1274                 return NULL;
1275
1276         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1277                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1278                                                GFP_KERNEL);
1279                 if (!SLAVE_AD_INFO(slave)) {
1280                         kfree(slave);
1281                         return NULL;
1282                 }
1283         }
1284         INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1285
1286         return slave;
1287 }
1288
1289 static void bond_free_slave(struct slave *slave)
1290 {
1291         struct bonding *bond = bond_get_bond_by_slave(slave);
1292
1293         cancel_delayed_work_sync(&slave->notify_work);
1294         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1295                 kfree(SLAVE_AD_INFO(slave));
1296
1297         kfree(slave);
1298 }
1299
1300 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1301 {
1302         info->bond_mode = BOND_MODE(bond);
1303         info->miimon = bond->params.miimon;
1304         info->num_slaves = bond->slave_cnt;
1305 }
1306
1307 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1308 {
1309         strcpy(info->slave_name, slave->dev->name);
1310         info->link = slave->link;
1311         info->state = bond_slave_state(slave);
1312         info->link_failure_count = slave->link_failure_count;
1313 }
1314
1315 static void bond_netdev_notify_work(struct work_struct *_work)
1316 {
1317         struct slave *slave = container_of(_work, struct slave,
1318                                            notify_work.work);
1319
1320         if (rtnl_trylock()) {
1321                 struct netdev_bonding_info binfo;
1322
1323                 bond_fill_ifslave(slave, &binfo.slave);
1324                 bond_fill_ifbond(slave->bond, &binfo.master);
1325                 netdev_bonding_info_change(slave->dev, &binfo);
1326                 rtnl_unlock();
1327         } else {
1328                 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1329         }
1330 }
1331
1332 void bond_queue_slave_event(struct slave *slave)
1333 {
1334         queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1335 }
1336
1337 void bond_lower_state_changed(struct slave *slave)
1338 {
1339         struct netdev_lag_lower_state_info info;
1340
1341         info.link_up = slave->link == BOND_LINK_UP ||
1342                        slave->link == BOND_LINK_FAIL;
1343         info.tx_enabled = bond_is_active_slave(slave);
1344         netdev_lower_state_changed(slave->dev, &info);
1345 }
1346
1347 /* enslave device <slave> to bond device <master> */
1348 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1349                  struct netlink_ext_ack *extack)
1350 {
1351         struct bonding *bond = netdev_priv(bond_dev);
1352         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1353         struct slave *new_slave = NULL, *prev_slave;
1354         struct sockaddr_storage ss;
1355         int link_reporting;
1356         int res = 0, i;
1357
1358         if (!bond->params.use_carrier &&
1359             slave_dev->ethtool_ops->get_link == NULL &&
1360             slave_ops->ndo_do_ioctl == NULL) {
1361                 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1362                             slave_dev->name);
1363         }
1364
1365         /* already in-use? */
1366         if (netdev_is_rx_handler_busy(slave_dev)) {
1367                 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1368                 netdev_err(bond_dev,
1369                            "Error: Device is in use and cannot be enslaved\n");
1370                 return -EBUSY;
1371         }
1372
1373         if (bond_dev == slave_dev) {
1374                 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1375                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1376                 return -EPERM;
1377         }
1378
1379         /* vlan challenged mutual exclusion */
1380         /* no need to lock since we're protected by rtnl_lock */
1381         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1382                 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1383                            slave_dev->name);
1384                 if (vlan_uses_dev(bond_dev)) {
1385                         NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1386                         netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1387                                    slave_dev->name, bond_dev->name);
1388                         return -EPERM;
1389                 } else {
1390                         netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1391                                     slave_dev->name, slave_dev->name,
1392                                     bond_dev->name);
1393                 }
1394         } else {
1395                 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1396                            slave_dev->name);
1397         }
1398
1399         /* Old ifenslave binaries are no longer supported.  These can
1400          * be identified with moderate accuracy by the state of the slave:
1401          * the current ifenslave will set the interface down prior to
1402          * enslaving it; the old ifenslave will not.
1403          */
1404         if (slave_dev->flags & IFF_UP) {
1405                 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
1406                 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1407                            slave_dev->name);
1408                 return -EPERM;
1409         }
1410
1411         /* set bonding device ether type by slave - bonding netdevices are
1412          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1413          * there is a need to override some of the type dependent attribs/funcs.
1414          *
1415          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1416          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1417          */
1418         if (!bond_has_slaves(bond)) {
1419                 if (bond_dev->type != slave_dev->type) {
1420                         netdev_dbg(bond_dev, "change device type from %d to %d\n",
1421                                    bond_dev->type, slave_dev->type);
1422
1423                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1424                                                        bond_dev);
1425                         res = notifier_to_errno(res);
1426                         if (res) {
1427                                 netdev_err(bond_dev, "refused to change device type\n");
1428                                 return -EBUSY;
1429                         }
1430
1431                         /* Flush unicast and multicast addresses */
1432                         dev_uc_flush(bond_dev);
1433                         dev_mc_flush(bond_dev);
1434
1435                         if (slave_dev->type != ARPHRD_ETHER)
1436                                 bond_setup_by_slave(bond_dev, slave_dev);
1437                         else {
1438                                 ether_setup(bond_dev);
1439                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1440                         }
1441
1442                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1443                                                  bond_dev);
1444                 }
1445         } else if (bond_dev->type != slave_dev->type) {
1446                 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1447                 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1448                            slave_dev->name, slave_dev->type, bond_dev->type);
1449                 return -EINVAL;
1450         }
1451
1452         if (slave_dev->type == ARPHRD_INFINIBAND &&
1453             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1454                 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1455                 netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n",
1456                             slave_dev->type);
1457                 res = -EOPNOTSUPP;
1458                 goto err_undo_flags;
1459         }
1460
1461         if (!slave_ops->ndo_set_mac_address ||
1462             slave_dev->type == ARPHRD_INFINIBAND) {
1463                 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1464                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1465                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1466                         if (!bond_has_slaves(bond)) {
1467                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1468                                 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1469                         } else {
1470                                 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1471                                 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");
1472                                 res = -EOPNOTSUPP;
1473                                 goto err_undo_flags;
1474                         }
1475                 }
1476         }
1477
1478         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1479
1480         /* If this is the first slave, then we need to set the master's hardware
1481          * address to be the same as the slave's.
1482          */
1483         if (!bond_has_slaves(bond) &&
1484             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1485                 bond_set_dev_addr(bond->dev, slave_dev);
1486
1487         new_slave = bond_alloc_slave(bond);
1488         if (!new_slave) {
1489                 res = -ENOMEM;
1490                 goto err_undo_flags;
1491         }
1492
1493         new_slave->bond = bond;
1494         new_slave->dev = slave_dev;
1495         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1496          * is set via sysfs or module option if desired.
1497          */
1498         new_slave->queue_id = 0;
1499
1500         /* Save slave's original mtu and then set it to match the bond */
1501         new_slave->original_mtu = slave_dev->mtu;
1502         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1503         if (res) {
1504                 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1505                 goto err_free;
1506         }
1507
1508         /* Save slave's original ("permanent") mac address for modes
1509          * that need it, and for restoring it upon release, and then
1510          * set it to the master's address
1511          */
1512         bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1513                           slave_dev->addr_len);
1514
1515         if (!bond->params.fail_over_mac ||
1516             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1517                 /* Set slave to master's mac address.  The application already
1518                  * set the master's mac address to that of the first slave
1519                  */
1520                 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1521                 ss.ss_family = slave_dev->type;
1522                 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1523                 if (res) {
1524                         netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1525                         goto err_restore_mtu;
1526                 }
1527         }
1528
1529         /* set slave flag before open to prevent IPv6 addrconf */
1530         slave_dev->flags |= IFF_SLAVE;
1531
1532         /* open the slave since the application closed it */
1533         res = dev_open(slave_dev);
1534         if (res) {
1535                 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1536                 goto err_restore_mac;
1537         }
1538
1539         slave_dev->priv_flags |= IFF_BONDING;
1540         /* initialize slave stats */
1541         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1542
1543         if (bond_is_lb(bond)) {
1544                 /* bond_alb_init_slave() must be called before all other stages since
1545                  * it might fail and we do not want to have to undo everything
1546                  */
1547                 res = bond_alb_init_slave(bond, new_slave);
1548                 if (res)
1549                         goto err_close;
1550         }
1551
1552         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1553         if (res) {
1554                 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1555                            slave_dev->name);
1556                 goto err_close;
1557         }
1558
1559         prev_slave = bond_last_slave(bond);
1560
1561         new_slave->delay = 0;
1562         new_slave->link_failure_count = 0;
1563
1564         if (bond_update_speed_duplex(new_slave) &&
1565             bond_needs_speed_duplex(bond))
1566                 new_slave->link = BOND_LINK_DOWN;
1567
1568         new_slave->last_rx = jiffies -
1569                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1570         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1571                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1572
1573         if (bond->params.miimon && !bond->params.use_carrier) {
1574                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1575
1576                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1577                         /* miimon is set but a bonded network driver
1578                          * does not support ETHTOOL/MII and
1579                          * arp_interval is not set.  Note: if
1580                          * use_carrier is enabled, we will never go
1581                          * here (because netif_carrier is always
1582                          * supported); thus, we don't need to change
1583                          * the messages for netif_carrier.
1584                          */
1585                         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",
1586                                     slave_dev->name);
1587                 } else if (link_reporting == -1) {
1588                         /* unable get link status using mii/ethtool */
1589                         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",
1590                                     slave_dev->name);
1591                 }
1592         }
1593
1594         /* check for initial state */
1595         new_slave->link = BOND_LINK_NOCHANGE;
1596         if (bond->params.miimon) {
1597                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1598                         if (bond->params.updelay) {
1599                                 bond_set_slave_link_state(new_slave,
1600                                                           BOND_LINK_BACK,
1601                                                           BOND_SLAVE_NOTIFY_NOW);
1602                                 new_slave->delay = bond->params.updelay;
1603                         } else {
1604                                 bond_set_slave_link_state(new_slave,
1605                                                           BOND_LINK_UP,
1606                                                           BOND_SLAVE_NOTIFY_NOW);
1607                         }
1608                 } else {
1609                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1610                                                   BOND_SLAVE_NOTIFY_NOW);
1611                 }
1612         } else if (bond->params.arp_interval) {
1613                 bond_set_slave_link_state(new_slave,
1614                                           (netif_carrier_ok(slave_dev) ?
1615                                           BOND_LINK_UP : BOND_LINK_DOWN),
1616                                           BOND_SLAVE_NOTIFY_NOW);
1617         } else {
1618                 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1619                                           BOND_SLAVE_NOTIFY_NOW);
1620         }
1621
1622         if (new_slave->link != BOND_LINK_DOWN)
1623                 new_slave->last_link_up = jiffies;
1624         netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1625                    new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1626                    (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1627
1628         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1629                 /* if there is a primary slave, remember it */
1630                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1631                         rcu_assign_pointer(bond->primary_slave, new_slave);
1632                         bond->force_primary = true;
1633                 }
1634         }
1635
1636         switch (BOND_MODE(bond)) {
1637         case BOND_MODE_ACTIVEBACKUP:
1638                 bond_set_slave_inactive_flags(new_slave,
1639                                               BOND_SLAVE_NOTIFY_NOW);
1640                 break;
1641         case BOND_MODE_8023AD:
1642                 /* in 802.3ad mode, the internal mechanism
1643                  * will activate the slaves in the selected
1644                  * aggregator
1645                  */
1646                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1647                 /* if this is the first slave */
1648                 if (!prev_slave) {
1649                         SLAVE_AD_INFO(new_slave)->id = 1;
1650                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1651                          * can be called only after the mac address of the bond is set
1652                          */
1653                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1654                 } else {
1655                         SLAVE_AD_INFO(new_slave)->id =
1656                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1657                 }
1658
1659                 bond_3ad_bind_slave(new_slave);
1660                 break;
1661         case BOND_MODE_TLB:
1662         case BOND_MODE_ALB:
1663                 bond_set_active_slave(new_slave);
1664                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1665                 break;
1666         default:
1667                 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1668
1669                 /* always active in trunk mode */
1670                 bond_set_active_slave(new_slave);
1671
1672                 /* In trunking mode there is little meaning to curr_active_slave
1673                  * anyway (it holds no special properties of the bond device),
1674                  * so we can change it without calling change_active_interface()
1675                  */
1676                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1677                     new_slave->link == BOND_LINK_UP)
1678                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1679
1680                 break;
1681         } /* switch(bond_mode) */
1682
1683 #ifdef CONFIG_NET_POLL_CONTROLLER
1684         if (bond->dev->npinfo) {
1685                 if (slave_enable_netpoll(new_slave)) {
1686                         netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1687                         res = -EBUSY;
1688                         goto err_detach;
1689                 }
1690         }
1691 #endif
1692
1693         if (!(bond_dev->features & NETIF_F_LRO))
1694                 dev_disable_lro(slave_dev);
1695
1696         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1697                                          new_slave);
1698         if (res) {
1699                 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1700                 goto err_detach;
1701         }
1702
1703         res = bond_master_upper_dev_link(bond, new_slave, extack);
1704         if (res) {
1705                 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1706                 goto err_unregister;
1707         }
1708
1709         res = bond_sysfs_slave_add(new_slave);
1710         if (res) {
1711                 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1712                 goto err_upper_unlink;
1713         }
1714
1715         bond->nest_level = dev_get_nest_level(bond_dev) + 1;
1716
1717         /* If the mode uses primary, then the following is handled by
1718          * bond_change_active_slave().
1719          */
1720         if (!bond_uses_primary(bond)) {
1721                 /* set promiscuity level to new slave */
1722                 if (bond_dev->flags & IFF_PROMISC) {
1723                         res = dev_set_promiscuity(slave_dev, 1);
1724                         if (res)
1725                                 goto err_sysfs_del;
1726                 }
1727
1728                 /* set allmulti level to new slave */
1729                 if (bond_dev->flags & IFF_ALLMULTI) {
1730                         res = dev_set_allmulti(slave_dev, 1);
1731                         if (res) {
1732                                 if (bond_dev->flags & IFF_PROMISC)
1733                                         dev_set_promiscuity(slave_dev, -1);
1734                                 goto err_sysfs_del;
1735                         }
1736                 }
1737
1738                 netif_addr_lock_bh(bond_dev);
1739                 dev_mc_sync_multiple(slave_dev, bond_dev);
1740                 dev_uc_sync_multiple(slave_dev, bond_dev);
1741                 netif_addr_unlock_bh(bond_dev);
1742
1743                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1744                         /* add lacpdu mc addr to mc list */
1745                         u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1746
1747                         dev_mc_add(slave_dev, lacpdu_multicast);
1748                 }
1749         }
1750
1751         bond->slave_cnt++;
1752         bond_compute_features(bond);
1753         bond_set_carrier(bond);
1754
1755         if (bond_uses_primary(bond)) {
1756                 block_netpoll_tx();
1757                 bond_select_active_slave(bond);
1758                 unblock_netpoll_tx();
1759         }
1760
1761         if (bond_mode_can_use_xmit_hash(bond))
1762                 bond_update_slave_arr(bond, NULL);
1763
1764
1765         netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1766                     slave_dev->name,
1767                     bond_is_active_slave(new_slave) ? "an active" : "a backup",
1768                     new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1769
1770         /* enslave is successful */
1771         bond_queue_slave_event(new_slave);
1772         return 0;
1773
1774 /* Undo stages on error */
1775 err_sysfs_del:
1776         bond_sysfs_slave_del(new_slave);
1777
1778 err_upper_unlink:
1779         bond_upper_dev_unlink(bond, new_slave);
1780
1781 err_unregister:
1782         netdev_rx_handler_unregister(slave_dev);
1783
1784 err_detach:
1785         vlan_vids_del_by_dev(slave_dev, bond_dev);
1786         if (rcu_access_pointer(bond->primary_slave) == new_slave)
1787                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1788         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1789                 block_netpoll_tx();
1790                 bond_change_active_slave(bond, NULL);
1791                 bond_select_active_slave(bond);
1792                 unblock_netpoll_tx();
1793         }
1794         /* either primary_slave or curr_active_slave might've changed */
1795         synchronize_rcu();
1796         slave_disable_netpoll(new_slave);
1797
1798 err_close:
1799         slave_dev->priv_flags &= ~IFF_BONDING;
1800         dev_close(slave_dev);
1801
1802 err_restore_mac:
1803         slave_dev->flags &= ~IFF_SLAVE;
1804         if (!bond->params.fail_over_mac ||
1805             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1806                 /* XXX TODO - fom follow mode needs to change master's
1807                  * MAC if this slave's MAC is in use by the bond, or at
1808                  * least print a warning.
1809                  */
1810                 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
1811                                   new_slave->dev->addr_len);
1812                 ss.ss_family = slave_dev->type;
1813                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1814         }
1815
1816 err_restore_mtu:
1817         dev_set_mtu(slave_dev, new_slave->original_mtu);
1818
1819 err_free:
1820         bond_free_slave(new_slave);
1821
1822 err_undo_flags:
1823         /* Enslave of first slave has failed and we need to fix master's mac */
1824         if (!bond_has_slaves(bond)) {
1825                 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1826                                             slave_dev->dev_addr))
1827                         eth_hw_addr_random(bond_dev);
1828                 if (bond_dev->type != ARPHRD_ETHER) {
1829                         dev_close(bond_dev);
1830                         ether_setup(bond_dev);
1831                         bond_dev->flags |= IFF_MASTER;
1832                         bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1833                 }
1834         }
1835
1836         return res;
1837 }
1838
1839 /* Try to release the slave device <slave> from the bond device <master>
1840  * It is legal to access curr_active_slave without a lock because all the function
1841  * is RTNL-locked. If "all" is true it means that the function is being called
1842  * while destroying a bond interface and all slaves are being released.
1843  *
1844  * The rules for slave state should be:
1845  *   for Active/Backup:
1846  *     Active stays on all backups go down
1847  *   for Bonded connections:
1848  *     The first up interface should be left on and all others downed.
1849  */
1850 static int __bond_release_one(struct net_device *bond_dev,
1851                               struct net_device *slave_dev,
1852                               bool all, bool unregister)
1853 {
1854         struct bonding *bond = netdev_priv(bond_dev);
1855         struct slave *slave, *oldcurrent;
1856         struct sockaddr_storage ss;
1857         int old_flags = bond_dev->flags;
1858         netdev_features_t old_features = bond_dev->features;
1859
1860         /* slave is not a slave or master is not master of this slave */
1861         if (!(slave_dev->flags & IFF_SLAVE) ||
1862             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1863                 netdev_dbg(bond_dev, "cannot release %s\n",
1864                            slave_dev->name);
1865                 return -EINVAL;
1866         }
1867
1868         block_netpoll_tx();
1869
1870         slave = bond_get_slave_by_dev(bond, slave_dev);
1871         if (!slave) {
1872                 /* not a slave of this bond */
1873                 netdev_info(bond_dev, "%s not enslaved\n",
1874                             slave_dev->name);
1875                 unblock_netpoll_tx();
1876                 return -EINVAL;
1877         }
1878
1879         bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1880
1881         bond_sysfs_slave_del(slave);
1882
1883         /* recompute stats just before removing the slave */
1884         bond_get_stats(bond->dev, &bond->bond_stats);
1885
1886         bond_upper_dev_unlink(bond, slave);
1887         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1888          * for this slave anymore.
1889          */
1890         netdev_rx_handler_unregister(slave_dev);
1891
1892         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1893                 bond_3ad_unbind_slave(slave);
1894
1895         if (bond_mode_can_use_xmit_hash(bond))
1896                 bond_update_slave_arr(bond, slave);
1897
1898         netdev_info(bond_dev, "Releasing %s interface %s\n",
1899                     bond_is_active_slave(slave) ? "active" : "backup",
1900                     slave_dev->name);
1901
1902         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1903
1904         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1905
1906         if (!all && (!bond->params.fail_over_mac ||
1907                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1908                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1909                     bond_has_slaves(bond))
1910                         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",
1911                                     slave_dev->name, slave->perm_hwaddr,
1912                                     bond_dev->name, slave_dev->name);
1913         }
1914
1915         if (rtnl_dereference(bond->primary_slave) == slave)
1916                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1917
1918         if (oldcurrent == slave)
1919                 bond_change_active_slave(bond, NULL);
1920
1921         if (bond_is_lb(bond)) {
1922                 /* Must be called only after the slave has been
1923                  * detached from the list and the curr_active_slave
1924                  * has been cleared (if our_slave == old_current),
1925                  * but before a new active slave is selected.
1926                  */
1927                 bond_alb_deinit_slave(bond, slave);
1928         }
1929
1930         if (all) {
1931                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1932         } else if (oldcurrent == slave) {
1933                 /* Note that we hold RTNL over this sequence, so there
1934                  * is no concern that another slave add/remove event
1935                  * will interfere.
1936                  */
1937                 bond_select_active_slave(bond);
1938         }
1939
1940         if (!bond_has_slaves(bond)) {
1941                 bond_set_carrier(bond);
1942                 eth_hw_addr_random(bond_dev);
1943                 bond->nest_level = SINGLE_DEPTH_NESTING;
1944         } else {
1945                 bond->nest_level = dev_get_nest_level(bond_dev) + 1;
1946         }
1947
1948         unblock_netpoll_tx();
1949         synchronize_rcu();
1950         bond->slave_cnt--;
1951
1952         if (!bond_has_slaves(bond)) {
1953                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1954                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1955         }
1956
1957         bond_compute_features(bond);
1958         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1959             (old_features & NETIF_F_VLAN_CHALLENGED))
1960                 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1961                             slave_dev->name, bond_dev->name);
1962
1963         vlan_vids_del_by_dev(slave_dev, bond_dev);
1964
1965         /* If the mode uses primary, then this case was handled above by
1966          * bond_change_active_slave(..., NULL)
1967          */
1968         if (!bond_uses_primary(bond)) {
1969                 /* unset promiscuity level from slave
1970                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1971                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1972                  * value of that flag before that change, as that was the value
1973                  * when this slave was attached, so we cache at the start of the
1974                  * function and use it here. Same goes for ALLMULTI below
1975                  */
1976                 if (old_flags & IFF_PROMISC)
1977                         dev_set_promiscuity(slave_dev, -1);
1978
1979                 /* unset allmulti level from slave */
1980                 if (old_flags & IFF_ALLMULTI)
1981                         dev_set_allmulti(slave_dev, -1);
1982
1983                 bond_hw_addr_flush(bond_dev, slave_dev);
1984         }
1985
1986         slave_disable_netpoll(slave);
1987
1988         /* close slave before restoring its mac address */
1989         dev_close(slave_dev);
1990
1991         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1992             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1993                 /* restore original ("permanent") mac address */
1994                 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
1995                                   slave->dev->addr_len);
1996                 ss.ss_family = slave_dev->type;
1997                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1998         }
1999
2000         if (unregister)
2001                 __dev_set_mtu(slave_dev, slave->original_mtu);
2002         else
2003                 dev_set_mtu(slave_dev, slave->original_mtu);
2004
2005         slave_dev->priv_flags &= ~IFF_BONDING;
2006
2007         bond_free_slave(slave);
2008
2009         return 0;
2010 }
2011
2012 /* A wrapper used because of ndo_del_link */
2013 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2014 {
2015         return __bond_release_one(bond_dev, slave_dev, false, false);
2016 }
2017
2018 /* First release a slave and then destroy the bond if no more slaves are left.
2019  * Must be under rtnl_lock when this function is called.
2020  */
2021 static int  bond_release_and_destroy(struct net_device *bond_dev,
2022                                      struct net_device *slave_dev)
2023 {
2024         struct bonding *bond = netdev_priv(bond_dev);
2025         int ret;
2026
2027         ret = __bond_release_one(bond_dev, slave_dev, false, true);
2028         if (ret == 0 && !bond_has_slaves(bond)) {
2029                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2030                 netdev_info(bond_dev, "Destroying bond %s\n",
2031                             bond_dev->name);
2032                 bond_remove_proc_entry(bond);
2033                 unregister_netdevice(bond_dev);
2034         }
2035         return ret;
2036 }
2037
2038 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2039 {
2040         struct bonding *bond = netdev_priv(bond_dev);
2041         bond_fill_ifbond(bond, info);
2042 }
2043
2044 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2045 {
2046         struct bonding *bond = netdev_priv(bond_dev);
2047         struct list_head *iter;
2048         int i = 0, res = -ENODEV;
2049         struct slave *slave;
2050
2051         bond_for_each_slave(bond, slave, iter) {
2052                 if (i++ == (int)info->slave_id) {
2053                         res = 0;
2054                         bond_fill_ifslave(slave, info);
2055                         break;
2056                 }
2057         }
2058
2059         return res;
2060 }
2061
2062 /*-------------------------------- Monitoring -------------------------------*/
2063
2064 /* called with rcu_read_lock() */
2065 static int bond_miimon_inspect(struct bonding *bond)
2066 {
2067         int link_state, commit = 0;
2068         struct list_head *iter;
2069         struct slave *slave;
2070         bool ignore_updelay;
2071
2072         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2073
2074         bond_for_each_slave_rcu(bond, slave, iter) {
2075                 slave->new_link = BOND_LINK_NOCHANGE;
2076                 slave->link_new_state = slave->link;
2077
2078                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2079
2080                 switch (slave->link) {
2081                 case BOND_LINK_UP:
2082                         if (link_state)
2083                                 continue;
2084
2085                         bond_propose_link_state(slave, BOND_LINK_FAIL);
2086                         commit++;
2087                         slave->delay = bond->params.downdelay;
2088                         if (slave->delay) {
2089                                 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2090                                             (BOND_MODE(bond) ==
2091                                              BOND_MODE_ACTIVEBACKUP) ?
2092                                              (bond_is_active_slave(slave) ?
2093                                               "active " : "backup ") : "",
2094                                             slave->dev->name,
2095                                             bond->params.downdelay * bond->params.miimon);
2096                         }
2097                         /*FALLTHRU*/
2098                 case BOND_LINK_FAIL:
2099                         if (link_state) {
2100                                 /* recovered before downdelay expired */
2101                                 bond_propose_link_state(slave, BOND_LINK_UP);
2102                                 slave->last_link_up = jiffies;
2103                                 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2104                                             (bond->params.downdelay - slave->delay) *
2105                                             bond->params.miimon,
2106                                             slave->dev->name);
2107                                 commit++;
2108                                 continue;
2109                         }
2110
2111                         if (slave->delay <= 0) {
2112                                 slave->new_link = BOND_LINK_DOWN;
2113                                 commit++;
2114                                 continue;
2115                         }
2116
2117                         slave->delay--;
2118                         break;
2119
2120                 case BOND_LINK_DOWN:
2121                         if (!link_state)
2122                                 continue;
2123
2124                         bond_propose_link_state(slave, BOND_LINK_BACK);
2125                         commit++;
2126                         slave->delay = bond->params.updelay;
2127
2128                         if (slave->delay) {
2129                                 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2130                                             slave->dev->name,
2131                                             ignore_updelay ? 0 :
2132                                             bond->params.updelay *
2133                                             bond->params.miimon);
2134                         }
2135                         /*FALLTHRU*/
2136                 case BOND_LINK_BACK:
2137                         if (!link_state) {
2138                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2139                                 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2140                                             (bond->params.updelay - slave->delay) *
2141                                             bond->params.miimon,
2142                                             slave->dev->name);
2143                                 commit++;
2144                                 continue;
2145                         }
2146
2147                         if (ignore_updelay)
2148                                 slave->delay = 0;
2149
2150                         if (slave->delay <= 0) {
2151                                 slave->new_link = BOND_LINK_UP;
2152                                 commit++;
2153                                 ignore_updelay = false;
2154                                 continue;
2155                         }
2156
2157                         slave->delay--;
2158                         break;
2159                 }
2160         }
2161
2162         return commit;
2163 }
2164
2165 static void bond_miimon_link_change(struct bonding *bond,
2166                                     struct slave *slave,
2167                                     char link)
2168 {
2169         switch (BOND_MODE(bond)) {
2170         case BOND_MODE_8023AD:
2171                 bond_3ad_handle_link_change(slave, link);
2172                 break;
2173         case BOND_MODE_TLB:
2174         case BOND_MODE_ALB:
2175                 bond_alb_handle_link_change(bond, slave, link);
2176                 break;
2177         case BOND_MODE_XOR:
2178                 bond_update_slave_arr(bond, NULL);
2179                 break;
2180         }
2181 }
2182
2183 static void bond_miimon_commit(struct bonding *bond)
2184 {
2185         struct list_head *iter;
2186         struct slave *slave, *primary;
2187
2188         bond_for_each_slave(bond, slave, iter) {
2189                 switch (slave->new_link) {
2190                 case BOND_LINK_NOCHANGE:
2191                         continue;
2192
2193                 case BOND_LINK_UP:
2194                         if (bond_update_speed_duplex(slave) &&
2195                             bond_needs_speed_duplex(bond)) {
2196                                 slave->link = BOND_LINK_DOWN;
2197                                 if (net_ratelimit())
2198                                         netdev_warn(bond->dev,
2199                                                     "failed to get link speed/duplex for %s\n",
2200                                                     slave->dev->name);
2201                                 continue;
2202                         }
2203                         bond_set_slave_link_state(slave, BOND_LINK_UP,
2204                                                   BOND_SLAVE_NOTIFY_NOW);
2205                         slave->last_link_up = jiffies;
2206
2207                         primary = rtnl_dereference(bond->primary_slave);
2208                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2209                                 /* prevent it from being the active one */
2210                                 bond_set_backup_slave(slave);
2211                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2212                                 /* make it immediately active */
2213                                 bond_set_active_slave(slave);
2214                         } else if (slave != primary) {
2215                                 /* prevent it from being the active one */
2216                                 bond_set_backup_slave(slave);
2217                         }
2218
2219                         netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2220                                     slave->dev->name,
2221                                     slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2222                                     slave->duplex ? "full" : "half");
2223
2224                         bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2225
2226                         if (!bond->curr_active_slave || slave == primary)
2227                                 goto do_failover;
2228
2229                         continue;
2230
2231                 case BOND_LINK_DOWN:
2232                         if (slave->link_failure_count < UINT_MAX)
2233                                 slave->link_failure_count++;
2234
2235                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2236                                                   BOND_SLAVE_NOTIFY_NOW);
2237
2238                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2239                             BOND_MODE(bond) == BOND_MODE_8023AD)
2240                                 bond_set_slave_inactive_flags(slave,
2241                                                               BOND_SLAVE_NOTIFY_NOW);
2242
2243                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2244                                     slave->dev->name);
2245
2246                         bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2247
2248                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2249                                 goto do_failover;
2250
2251                         continue;
2252
2253                 default:
2254                         netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2255                                    slave->new_link, slave->dev->name);
2256                         slave->new_link = BOND_LINK_NOCHANGE;
2257
2258                         continue;
2259                 }
2260
2261 do_failover:
2262                 block_netpoll_tx();
2263                 bond_select_active_slave(bond);
2264                 unblock_netpoll_tx();
2265         }
2266
2267         bond_set_carrier(bond);
2268 }
2269
2270 /* bond_mii_monitor
2271  *
2272  * Really a wrapper that splits the mii monitor into two phases: an
2273  * inspection, then (if inspection indicates something needs to be done)
2274  * an acquisition of appropriate locks followed by a commit phase to
2275  * implement whatever link state changes are indicated.
2276  */
2277 static void bond_mii_monitor(struct work_struct *work)
2278 {
2279         struct bonding *bond = container_of(work, struct bonding,
2280                                             mii_work.work);
2281         bool should_notify_peers = false;
2282         unsigned long delay;
2283         struct slave *slave;
2284         struct list_head *iter;
2285
2286         delay = msecs_to_jiffies(bond->params.miimon);
2287
2288         if (!bond_has_slaves(bond))
2289                 goto re_arm;
2290
2291         rcu_read_lock();
2292
2293         should_notify_peers = bond_should_notify_peers(bond);
2294
2295         if (bond_miimon_inspect(bond)) {
2296                 rcu_read_unlock();
2297
2298                 /* Race avoidance with bond_close cancel of workqueue */
2299                 if (!rtnl_trylock()) {
2300                         delay = 1;
2301                         should_notify_peers = false;
2302                         goto re_arm;
2303                 }
2304
2305                 bond_for_each_slave(bond, slave, iter) {
2306                         bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2307                 }
2308                 bond_miimon_commit(bond);
2309
2310                 rtnl_unlock();  /* might sleep, hold no other locks */
2311         } else
2312                 rcu_read_unlock();
2313
2314 re_arm:
2315         if (bond->params.miimon)
2316                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2317
2318         if (should_notify_peers) {
2319                 if (!rtnl_trylock())
2320                         return;
2321                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2322                 rtnl_unlock();
2323         }
2324 }
2325
2326 static int bond_upper_dev_walk(struct net_device *upper, void *data)
2327 {
2328         __be32 ip = *((__be32 *)data);
2329
2330         return ip == bond_confirm_addr(upper, 0, ip);
2331 }
2332
2333 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2334 {
2335         bool ret = false;
2336
2337         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2338                 return true;
2339
2340         rcu_read_lock();
2341         if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip))
2342                 ret = true;
2343         rcu_read_unlock();
2344
2345         return ret;
2346 }
2347
2348 /* We go to the (large) trouble of VLAN tagging ARP frames because
2349  * switches in VLAN mode (especially if ports are configured as
2350  * "native" to a VLAN) might not pass non-tagged frames.
2351  */
2352 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2353                           __be32 dest_ip, __be32 src_ip,
2354                           struct bond_vlan_tag *tags)
2355 {
2356         struct sk_buff *skb;
2357         struct bond_vlan_tag *outer_tag = tags;
2358
2359         netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2360                    arp_op, slave_dev->name, &dest_ip, &src_ip);
2361
2362         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2363                          NULL, slave_dev->dev_addr, NULL);
2364
2365         if (!skb) {
2366                 net_err_ratelimited("ARP packet allocation failed\n");
2367                 return;
2368         }
2369
2370         if (!tags || tags->vlan_proto == VLAN_N_VID)
2371                 goto xmit;
2372
2373         tags++;
2374
2375         /* Go through all the tags backwards and add them to the packet */
2376         while (tags->vlan_proto != VLAN_N_VID) {
2377                 if (!tags->vlan_id) {
2378                         tags++;
2379                         continue;
2380                 }
2381
2382                 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2383                            ntohs(outer_tag->vlan_proto), tags->vlan_id);
2384                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2385                                                 tags->vlan_id);
2386                 if (!skb) {
2387                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2388                         return;
2389                 }
2390
2391                 tags++;
2392         }
2393         /* Set the outer tag */
2394         if (outer_tag->vlan_id) {
2395                 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2396                            ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2397                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2398                                        outer_tag->vlan_id);
2399         }
2400
2401 xmit:
2402         arp_xmit(skb);
2403 }
2404
2405 /* Validate the device path between the @start_dev and the @end_dev.
2406  * The path is valid if the @end_dev is reachable through device
2407  * stacking.
2408  * When the path is validated, collect any vlan information in the
2409  * path.
2410  */
2411 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2412                                               struct net_device *end_dev,
2413                                               int level)
2414 {
2415         struct bond_vlan_tag *tags;
2416         struct net_device *upper;
2417         struct list_head  *iter;
2418
2419         if (start_dev == end_dev) {
2420                 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2421                 if (!tags)
2422                         return ERR_PTR(-ENOMEM);
2423                 tags[level].vlan_proto = VLAN_N_VID;
2424                 return tags;
2425         }
2426
2427         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2428                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2429                 if (IS_ERR_OR_NULL(tags)) {
2430                         if (IS_ERR(tags))
2431                                 return tags;
2432                         continue;
2433                 }
2434                 if (is_vlan_dev(upper)) {
2435                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2436                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2437                 }
2438
2439                 return tags;
2440         }
2441
2442         return NULL;
2443 }
2444
2445 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2446 {
2447         struct rtable *rt;
2448         struct bond_vlan_tag *tags;
2449         __be32 *targets = bond->params.arp_targets, addr;
2450         int i;
2451
2452         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2453                 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2454                 tags = NULL;
2455
2456                 /* Find out through which dev should the packet go */
2457                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2458                                      RTO_ONLINK, 0);
2459                 if (IS_ERR(rt)) {
2460                         /* there's no route to target - try to send arp
2461                          * probe to generate any traffic (arp_validate=0)
2462                          */
2463                         if (bond->params.arp_validate)
2464                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2465                                                      bond->dev->name,
2466                                                      &targets[i]);
2467                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2468                                       0, tags);
2469                         continue;
2470                 }
2471
2472                 /* bond device itself */
2473                 if (rt->dst.dev == bond->dev)
2474                         goto found;
2475
2476                 rcu_read_lock();
2477                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2478                 rcu_read_unlock();
2479
2480                 if (!IS_ERR_OR_NULL(tags))
2481                         goto found;
2482
2483                 /* Not our device - skip */
2484                 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2485                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2486
2487                 ip_rt_put(rt);
2488                 continue;
2489
2490 found:
2491                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2492                 ip_rt_put(rt);
2493                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2494                               addr, tags);
2495                 kfree(tags);
2496         }
2497 }
2498
2499 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2500 {
2501         int i;
2502
2503         if (!sip || !bond_has_this_ip(bond, tip)) {
2504                 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2505                            &sip, &tip);
2506                 return;
2507         }
2508
2509         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2510         if (i == -1) {
2511                 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2512                            &sip);
2513                 return;
2514         }
2515         slave->last_rx = jiffies;
2516         slave->target_last_arp_rx[i] = jiffies;
2517 }
2518
2519 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2520                  struct slave *slave)
2521 {
2522         struct arphdr *arp = (struct arphdr *)skb->data;
2523         struct slave *curr_active_slave, *curr_arp_slave;
2524         unsigned char *arp_ptr;
2525         __be32 sip, tip;
2526         int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2527         unsigned int alen;
2528
2529         if (!slave_do_arp_validate(bond, slave)) {
2530                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2531                     !slave_do_arp_validate_only(bond))
2532                         slave->last_rx = jiffies;
2533                 return RX_HANDLER_ANOTHER;
2534         } else if (!is_arp) {
2535                 return RX_HANDLER_ANOTHER;
2536         }
2537
2538         alen = arp_hdr_len(bond->dev);
2539
2540         netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2541                    skb->dev->name);
2542
2543         if (alen > skb_headlen(skb)) {
2544                 arp = kmalloc(alen, GFP_ATOMIC);
2545                 if (!arp)
2546                         goto out_unlock;
2547                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2548                         goto out_unlock;
2549         }
2550
2551         if (arp->ar_hln != bond->dev->addr_len ||
2552             skb->pkt_type == PACKET_OTHERHOST ||
2553             skb->pkt_type == PACKET_LOOPBACK ||
2554             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2555             arp->ar_pro != htons(ETH_P_IP) ||
2556             arp->ar_pln != 4)
2557                 goto out_unlock;
2558
2559         arp_ptr = (unsigned char *)(arp + 1);
2560         arp_ptr += bond->dev->addr_len;
2561         memcpy(&sip, arp_ptr, 4);
2562         arp_ptr += 4 + bond->dev->addr_len;
2563         memcpy(&tip, arp_ptr, 4);
2564
2565         netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2566                    slave->dev->name, bond_slave_state(slave),
2567                      bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2568                      &sip, &tip);
2569
2570         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2571         curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2572
2573         /* We 'trust' the received ARP enough to validate it if:
2574          *
2575          * (a) the slave receiving the ARP is active (which includes the
2576          * current ARP slave, if any), or
2577          *
2578          * (b) the receiving slave isn't active, but there is a currently
2579          * active slave and it received valid arp reply(s) after it became
2580          * the currently active slave, or
2581          *
2582          * (c) there is an ARP slave that sent an ARP during the prior ARP
2583          * interval, and we receive an ARP reply on any slave.  We accept
2584          * these because switch FDB update delays may deliver the ARP
2585          * reply to a slave other than the sender of the ARP request.
2586          *
2587          * Note: for (b), backup slaves are receiving the broadcast ARP
2588          * request, not a reply.  This request passes from the sending
2589          * slave through the L2 switch(es) to the receiving slave.  Since
2590          * this is checking the request, sip/tip are swapped for
2591          * validation.
2592          *
2593          * This is done to avoid endless looping when we can't reach the
2594          * arp_ip_target and fool ourselves with our own arp requests.
2595          */
2596         if (bond_is_active_slave(slave))
2597                 bond_validate_arp(bond, slave, sip, tip);
2598         else if (curr_active_slave &&
2599                  time_after(slave_last_rx(bond, curr_active_slave),
2600                             curr_active_slave->last_link_up))
2601                 bond_validate_arp(bond, slave, tip, sip);
2602         else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2603                  bond_time_in_interval(bond,
2604                                        dev_trans_start(curr_arp_slave->dev), 1))
2605                 bond_validate_arp(bond, slave, sip, tip);
2606
2607 out_unlock:
2608         if (arp != (struct arphdr *)skb->data)
2609                 kfree(arp);
2610         return RX_HANDLER_ANOTHER;
2611 }
2612
2613 /* function to verify if we're in the arp_interval timeslice, returns true if
2614  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2615  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2616  */
2617 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2618                                   int mod)
2619 {
2620         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2621
2622         return time_in_range(jiffies,
2623                              last_act - delta_in_ticks,
2624                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2625 }
2626
2627 /* This function is called regularly to monitor each slave's link
2628  * ensuring that traffic is being sent and received when arp monitoring
2629  * is used in load-balancing mode. if the adapter has been dormant, then an
2630  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2631  * arp monitoring in active backup mode.
2632  */
2633 static void bond_loadbalance_arp_mon(struct bonding *bond)
2634 {
2635         struct slave *slave, *oldcurrent;
2636         struct list_head *iter;
2637         int do_failover = 0, slave_state_changed = 0;
2638
2639         if (!bond_has_slaves(bond))
2640                 goto re_arm;
2641
2642         rcu_read_lock();
2643
2644         oldcurrent = rcu_dereference(bond->curr_active_slave);
2645         /* see if any of the previous devices are up now (i.e. they have
2646          * xmt and rcv traffic). the curr_active_slave does not come into
2647          * the picture unless it is null. also, slave->last_link_up is not
2648          * needed here because we send an arp on each slave and give a slave
2649          * as long as it needs to get the tx/rx within the delta.
2650          * TODO: what about up/down delay in arp mode? it wasn't here before
2651          *       so it can wait
2652          */
2653         bond_for_each_slave_rcu(bond, slave, iter) {
2654                 unsigned long trans_start = dev_trans_start(slave->dev);
2655
2656                 slave->new_link = BOND_LINK_NOCHANGE;
2657
2658                 if (slave->link != BOND_LINK_UP) {
2659                         if (bond_time_in_interval(bond, trans_start, 1) &&
2660                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2661
2662                                 slave->new_link = BOND_LINK_UP;
2663                                 slave_state_changed = 1;
2664
2665                                 /* primary_slave has no meaning in round-robin
2666                                  * mode. the window of a slave being up and
2667                                  * curr_active_slave being null after enslaving
2668                                  * is closed.
2669                                  */
2670                                 if (!oldcurrent) {
2671                                         netdev_info(bond->dev, "link status definitely up for interface %s\n",
2672                                                     slave->dev->name);
2673                                         do_failover = 1;
2674                                 } else {
2675                                         netdev_info(bond->dev, "interface %s is now up\n",
2676                                                     slave->dev->name);
2677                                 }
2678                         }
2679                 } else {
2680                         /* slave->link == BOND_LINK_UP */
2681
2682                         /* not all switches will respond to an arp request
2683                          * when the source ip is 0, so don't take the link down
2684                          * if we don't know our ip yet
2685                          */
2686                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2687                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2688
2689                                 slave->new_link = BOND_LINK_DOWN;
2690                                 slave_state_changed = 1;
2691
2692                                 if (slave->link_failure_count < UINT_MAX)
2693                                         slave->link_failure_count++;
2694
2695                                 netdev_info(bond->dev, "interface %s is now down\n",
2696                                             slave->dev->name);
2697
2698                                 if (slave == oldcurrent)
2699                                         do_failover = 1;
2700                         }
2701                 }
2702
2703                 /* note: if switch is in round-robin mode, all links
2704                  * must tx arp to ensure all links rx an arp - otherwise
2705                  * links may oscillate or not come up at all; if switch is
2706                  * in something like xor mode, there is nothing we can
2707                  * do - all replies will be rx'ed on same link causing slaves
2708                  * to be unstable during low/no traffic periods
2709                  */
2710                 if (bond_slave_is_up(slave))
2711                         bond_arp_send_all(bond, slave);
2712         }
2713
2714         rcu_read_unlock();
2715
2716         if (do_failover || slave_state_changed) {
2717                 if (!rtnl_trylock())
2718                         goto re_arm;
2719
2720                 bond_for_each_slave(bond, slave, iter) {
2721                         if (slave->new_link != BOND_LINK_NOCHANGE)
2722                                 slave->link = slave->new_link;
2723                 }
2724
2725                 if (slave_state_changed) {
2726                         bond_slave_state_change(bond);
2727                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2728                                 bond_update_slave_arr(bond, NULL);
2729                 }
2730                 if (do_failover) {
2731                         block_netpoll_tx();
2732                         bond_select_active_slave(bond);
2733                         unblock_netpoll_tx();
2734                 }
2735                 rtnl_unlock();
2736         }
2737
2738 re_arm:
2739         if (bond->params.arp_interval)
2740                 queue_delayed_work(bond->wq, &bond->arp_work,
2741                                    msecs_to_jiffies(bond->params.arp_interval));
2742 }
2743
2744 /* Called to inspect slaves for active-backup mode ARP monitor link state
2745  * changes.  Sets new_link in slaves to specify what action should take
2746  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2747  * to link states must be committed.
2748  *
2749  * Called with rcu_read_lock held.
2750  */
2751 static int bond_ab_arp_inspect(struct bonding *bond)
2752 {
2753         unsigned long trans_start, last_rx;
2754         struct list_head *iter;
2755         struct slave *slave;
2756         int commit = 0;
2757
2758         bond_for_each_slave_rcu(bond, slave, iter) {
2759                 slave->new_link = BOND_LINK_NOCHANGE;
2760                 last_rx = slave_last_rx(bond, slave);
2761
2762                 if (slave->link != BOND_LINK_UP) {
2763                         if (bond_time_in_interval(bond, last_rx, 1)) {
2764                                 slave->new_link = BOND_LINK_UP;
2765                                 commit++;
2766                         }
2767                         continue;
2768                 }
2769
2770                 /* Give slaves 2*delta after being enslaved or made
2771                  * active.  This avoids bouncing, as the last receive
2772                  * times need a full ARP monitor cycle to be updated.
2773                  */
2774                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2775                         continue;
2776
2777                 /* Backup slave is down if:
2778                  * - No current_arp_slave AND
2779                  * - more than 3*delta since last receive AND
2780                  * - the bond has an IP address
2781                  *
2782                  * Note: a non-null current_arp_slave indicates
2783                  * the curr_active_slave went down and we are
2784                  * searching for a new one; under this condition
2785                  * we only take the curr_active_slave down - this
2786                  * gives each slave a chance to tx/rx traffic
2787                  * before being taken out
2788                  */
2789                 if (!bond_is_active_slave(slave) &&
2790                     !rcu_access_pointer(bond->current_arp_slave) &&
2791                     !bond_time_in_interval(bond, last_rx, 3)) {
2792                         slave->new_link = BOND_LINK_DOWN;
2793                         commit++;
2794                 }
2795
2796                 /* Active slave is down if:
2797                  * - more than 2*delta since transmitting OR
2798                  * - (more than 2*delta since receive AND
2799                  *    the bond has an IP address)
2800                  */
2801                 trans_start = dev_trans_start(slave->dev);
2802                 if (bond_is_active_slave(slave) &&
2803                     (!bond_time_in_interval(bond, trans_start, 2) ||
2804                      !bond_time_in_interval(bond, last_rx, 2))) {
2805                         slave->new_link = BOND_LINK_DOWN;
2806                         commit++;
2807                 }
2808         }
2809
2810         return commit;
2811 }
2812
2813 /* Called to commit link state changes noted by inspection step of
2814  * active-backup mode ARP monitor.
2815  *
2816  * Called with RTNL hold.
2817  */
2818 static void bond_ab_arp_commit(struct bonding *bond)
2819 {
2820         unsigned long trans_start;
2821         struct list_head *iter;
2822         struct slave *slave;
2823
2824         bond_for_each_slave(bond, slave, iter) {
2825                 switch (slave->new_link) {
2826                 case BOND_LINK_NOCHANGE:
2827                         continue;
2828
2829                 case BOND_LINK_UP:
2830                         trans_start = dev_trans_start(slave->dev);
2831                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
2832                             (!rtnl_dereference(bond->curr_active_slave) &&
2833                              bond_time_in_interval(bond, trans_start, 1))) {
2834                                 struct slave *current_arp_slave;
2835
2836                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2837                                 bond_set_slave_link_state(slave, BOND_LINK_UP,
2838                                                           BOND_SLAVE_NOTIFY_NOW);
2839                                 if (current_arp_slave) {
2840                                         bond_set_slave_inactive_flags(
2841                                                 current_arp_slave,
2842                                                 BOND_SLAVE_NOTIFY_NOW);
2843                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2844                                 }
2845
2846                                 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2847                                             slave->dev->name);
2848
2849                                 if (!rtnl_dereference(bond->curr_active_slave) ||
2850                                     slave == rtnl_dereference(bond->primary_slave))
2851                                         goto do_failover;
2852
2853                         }
2854
2855                         continue;
2856
2857                 case BOND_LINK_DOWN:
2858                         if (slave->link_failure_count < UINT_MAX)
2859                                 slave->link_failure_count++;
2860
2861                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2862                                                   BOND_SLAVE_NOTIFY_NOW);
2863                         bond_set_slave_inactive_flags(slave,
2864                                                       BOND_SLAVE_NOTIFY_NOW);
2865
2866                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2867                                     slave->dev->name);
2868
2869                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
2870                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2871                                 goto do_failover;
2872                         }
2873
2874                         continue;
2875
2876                 default:
2877                         netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2878                                    slave->new_link, slave->dev->name);
2879                         continue;
2880                 }
2881
2882 do_failover:
2883                 block_netpoll_tx();
2884                 bond_select_active_slave(bond);
2885                 unblock_netpoll_tx();
2886         }
2887
2888         bond_set_carrier(bond);
2889 }
2890
2891 /* Send ARP probes for active-backup mode ARP monitor.
2892  *
2893  * Called with rcu_read_lock held.
2894  */
2895 static bool bond_ab_arp_probe(struct bonding *bond)
2896 {
2897         struct slave *slave, *before = NULL, *new_slave = NULL,
2898                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2899                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2900         struct list_head *iter;
2901         bool found = false;
2902         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2903
2904         if (curr_arp_slave && curr_active_slave)
2905                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2906                             curr_arp_slave->dev->name,
2907                             curr_active_slave->dev->name);
2908
2909         if (curr_active_slave) {
2910                 bond_arp_send_all(bond, curr_active_slave);
2911                 return should_notify_rtnl;
2912         }
2913
2914         /* if we don't have a curr_active_slave, search for the next available
2915          * backup slave from the current_arp_slave and make it the candidate
2916          * for becoming the curr_active_slave
2917          */
2918
2919         if (!curr_arp_slave) {
2920                 curr_arp_slave = bond_first_slave_rcu(bond);
2921                 if (!curr_arp_slave)
2922                         return should_notify_rtnl;
2923         }
2924
2925         bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2926
2927         bond_for_each_slave_rcu(bond, slave, iter) {
2928                 if (!found && !before && bond_slave_is_up(slave))
2929                         before = slave;
2930
2931                 if (found && !new_slave && bond_slave_is_up(slave))
2932                         new_slave = slave;
2933                 /* if the link state is up at this point, we
2934                  * mark it down - this can happen if we have
2935                  * simultaneous link failures and
2936                  * reselect_active_interface doesn't make this
2937                  * one the current slave so it is still marked
2938                  * up when it is actually down
2939                  */
2940                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2941                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2942                                                   BOND_SLAVE_NOTIFY_LATER);
2943                         if (slave->link_failure_count < UINT_MAX)
2944                                 slave->link_failure_count++;
2945
2946                         bond_set_slave_inactive_flags(slave,
2947                                                       BOND_SLAVE_NOTIFY_LATER);
2948
2949                         netdev_info(bond->dev, "backup interface %s is now down\n",
2950                                     slave->dev->name);
2951                 }
2952                 if (slave == curr_arp_slave)
2953                         found = true;
2954         }
2955
2956         if (!new_slave && before)
2957                 new_slave = before;
2958
2959         if (!new_slave)
2960                 goto check_state;
2961
2962         bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2963                                   BOND_SLAVE_NOTIFY_LATER);
2964         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2965         bond_arp_send_all(bond, new_slave);
2966         new_slave->last_link_up = jiffies;
2967         rcu_assign_pointer(bond->current_arp_slave, new_slave);
2968
2969 check_state:
2970         bond_for_each_slave_rcu(bond, slave, iter) {
2971                 if (slave->should_notify || slave->should_notify_link) {
2972                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2973                         break;
2974                 }
2975         }
2976         return should_notify_rtnl;
2977 }
2978
2979 static void bond_activebackup_arp_mon(struct bonding *bond)
2980 {
2981         bool should_notify_peers = false;
2982         bool should_notify_rtnl = false;
2983         int delta_in_ticks;
2984
2985         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2986
2987         if (!bond_has_slaves(bond))
2988                 goto re_arm;
2989
2990         rcu_read_lock();
2991
2992         should_notify_peers = bond_should_notify_peers(bond);
2993
2994         if (bond_ab_arp_inspect(bond)) {
2995                 rcu_read_unlock();
2996
2997                 /* Race avoidance with bond_close flush of workqueue */
2998                 if (!rtnl_trylock()) {
2999                         delta_in_ticks = 1;
3000                         should_notify_peers = false;
3001                         goto re_arm;
3002                 }
3003
3004                 bond_ab_arp_commit(bond);
3005
3006                 rtnl_unlock();
3007                 rcu_read_lock();
3008         }
3009
3010         should_notify_rtnl = bond_ab_arp_probe(bond);
3011         rcu_read_unlock();
3012
3013 re_arm:
3014         if (bond->params.arp_interval)
3015                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3016
3017         if (should_notify_peers || should_notify_rtnl) {
3018                 if (!rtnl_trylock())
3019                         return;
3020
3021                 if (should_notify_peers)
3022                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3023                                                  bond->dev);
3024                 if (should_notify_rtnl) {
3025                         bond_slave_state_notify(bond);
3026                         bond_slave_link_notify(bond);
3027                 }
3028
3029                 rtnl_unlock();
3030         }
3031 }
3032
3033 static void bond_arp_monitor(struct work_struct *work)
3034 {
3035         struct bonding *bond = container_of(work, struct bonding,
3036                                             arp_work.work);
3037
3038         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3039                 bond_activebackup_arp_mon(bond);
3040         else
3041                 bond_loadbalance_arp_mon(bond);
3042 }
3043
3044 /*-------------------------- netdev event handling --------------------------*/
3045
3046 /* Change device name */
3047 static int bond_event_changename(struct bonding *bond)
3048 {
3049         bond_remove_proc_entry(bond);
3050         bond_create_proc_entry(bond);
3051
3052         bond_debug_reregister(bond);
3053
3054         return NOTIFY_DONE;
3055 }
3056
3057 static int bond_master_netdev_event(unsigned long event,
3058                                     struct net_device *bond_dev)
3059 {
3060         struct bonding *event_bond = netdev_priv(bond_dev);
3061
3062         switch (event) {
3063         case NETDEV_CHANGENAME:
3064                 return bond_event_changename(event_bond);
3065         case NETDEV_UNREGISTER:
3066                 bond_remove_proc_entry(event_bond);
3067                 break;
3068         case NETDEV_REGISTER:
3069                 bond_create_proc_entry(event_bond);
3070                 break;
3071         case NETDEV_NOTIFY_PEERS:
3072                 if (event_bond->send_peer_notif)
3073                         event_bond->send_peer_notif--;
3074                 break;
3075         default:
3076                 break;
3077         }
3078
3079         return NOTIFY_DONE;
3080 }
3081
3082 static int bond_slave_netdev_event(unsigned long event,
3083                                    struct net_device *slave_dev)
3084 {
3085         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3086         struct bonding *bond;
3087         struct net_device *bond_dev;
3088
3089         /* A netdev event can be generated while enslaving a device
3090          * before netdev_rx_handler_register is called in which case
3091          * slave will be NULL
3092          */
3093         if (!slave)
3094                 return NOTIFY_DONE;
3095         bond_dev = slave->bond->dev;
3096         bond = slave->bond;
3097         primary = rtnl_dereference(bond->primary_slave);
3098
3099         switch (event) {
3100         case NETDEV_UNREGISTER:
3101                 if (bond_dev->type != ARPHRD_ETHER)
3102                         bond_release_and_destroy(bond_dev, slave_dev);
3103                 else
3104                         __bond_release_one(bond_dev, slave_dev, false, true);
3105                 break;
3106         case NETDEV_UP:
3107         case NETDEV_CHANGE:
3108                 /* For 802.3ad mode only:
3109                  * Getting invalid Speed/Duplex values here will put slave
3110                  * in weird state. Mark it as link-fail if the link was
3111                  * previously up or link-down if it hasn't yet come up, and
3112                  * let link-monitoring (miimon) set it right when correct
3113                  * speeds/duplex are available.
3114                  */
3115                 if (bond_update_speed_duplex(slave) &&
3116                     BOND_MODE(bond) == BOND_MODE_8023AD) {
3117                         if (slave->last_link_up)
3118                                 slave->link = BOND_LINK_FAIL;
3119                         else
3120                                 slave->link = BOND_LINK_DOWN;
3121                 }
3122
3123                 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3124                         bond_3ad_adapter_speed_duplex_changed(slave);
3125                 /* Fallthrough */
3126         case NETDEV_DOWN:
3127                 /* Refresh slave-array if applicable!
3128                  * If the setup does not use miimon or arpmon (mode-specific!),
3129                  * then these events will not cause the slave-array to be
3130                  * refreshed. This will cause xmit to use a slave that is not
3131                  * usable. Avoid such situation by refeshing the array at these
3132                  * events. If these (miimon/arpmon) parameters are configured
3133                  * then array gets refreshed twice and that should be fine!
3134                  */
3135                 if (bond_mode_can_use_xmit_hash(bond))
3136                         bond_update_slave_arr(bond, NULL);
3137                 break;
3138         case NETDEV_CHANGEMTU:
3139                 /* TODO: Should slaves be allowed to
3140                  * independently alter their MTU?  For
3141                  * an active-backup bond, slaves need
3142                  * not be the same type of device, so
3143                  * MTUs may vary.  For other modes,
3144                  * slaves arguably should have the
3145                  * same MTUs. To do this, we'd need to
3146                  * take over the slave's change_mtu
3147                  * function for the duration of their
3148                  * servitude.
3149                  */
3150                 break;
3151         case NETDEV_CHANGENAME:
3152                 /* we don't care if we don't have primary set */
3153                 if (!bond_uses_primary(bond) ||
3154                     !bond->params.primary[0])
3155                         break;
3156
3157                 if (slave == primary) {
3158                         /* slave's name changed - he's no longer primary */
3159                         RCU_INIT_POINTER(bond->primary_slave, NULL);
3160                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3161                         /* we have a new primary slave */
3162                         rcu_assign_pointer(bond->primary_slave, slave);
3163                 } else { /* we didn't change primary - exit */
3164                         break;
3165                 }
3166
3167                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3168                             primary ? slave_dev->name : "none");
3169
3170                 block_netpoll_tx();
3171                 bond_select_active_slave(bond);
3172                 unblock_netpoll_tx();
3173                 break;
3174         case NETDEV_FEAT_CHANGE:
3175                 bond_compute_features(bond);
3176                 break;
3177         case NETDEV_RESEND_IGMP:
3178                 /* Propagate to master device */
3179                 call_netdevice_notifiers(event, slave->bond->dev);
3180                 break;
3181         default:
3182                 break;
3183         }
3184
3185         return NOTIFY_DONE;
3186 }
3187
3188 /* bond_netdev_event: handle netdev notifier chain events.
3189  *
3190  * This function receives events for the netdev chain.  The caller (an
3191  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3192  * locks for us to safely manipulate the slave devices (RTNL lock,
3193  * dev_probe_lock).
3194  */
3195 static int bond_netdev_event(struct notifier_block *this,
3196                              unsigned long event, void *ptr)
3197 {
3198         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3199
3200         netdev_dbg(event_dev, "event: %lx\n", event);
3201
3202         if (!(event_dev->priv_flags & IFF_BONDING))
3203                 return NOTIFY_DONE;
3204
3205         if (event_dev->flags & IFF_MASTER) {
3206                 int ret;
3207
3208                 netdev_dbg(event_dev, "IFF_MASTER\n");
3209                 ret = bond_master_netdev_event(event, event_dev);
3210                 if (ret != NOTIFY_DONE)
3211                         return ret;
3212         }
3213
3214         if (event_dev->flags & IFF_SLAVE) {
3215                 netdev_dbg(event_dev, "IFF_SLAVE\n");
3216                 return bond_slave_netdev_event(event, event_dev);
3217         }
3218
3219         return NOTIFY_DONE;
3220 }
3221
3222 static struct notifier_block bond_netdev_notifier = {
3223         .notifier_call = bond_netdev_event,
3224 };
3225
3226 /*---------------------------- Hashing Policies -----------------------------*/
3227
3228 /* L2 hash helper */
3229 static inline u32 bond_eth_hash(struct sk_buff *skb)
3230 {
3231         struct ethhdr *ep, hdr_tmp;
3232
3233         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3234         if (ep)
3235                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3236         return 0;
3237 }
3238
3239 /* Extract the appropriate headers based on bond's xmit policy */
3240 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3241                               struct flow_keys *fk)
3242 {
3243         const struct ipv6hdr *iph6;
3244         const struct iphdr *iph;
3245         int noff, proto = -1;
3246
3247         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3248                 return skb_flow_dissect_flow_keys(skb, fk, 0);
3249
3250         fk->ports.ports = 0;
3251         noff = skb_network_offset(skb);
3252         if (skb->protocol == htons(ETH_P_IP)) {
3253                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3254                         return false;
3255                 iph = ip_hdr(skb);
3256                 iph_to_flow_copy_v4addrs(fk, iph);
3257                 noff += iph->ihl << 2;
3258                 if (!ip_is_fragment(iph))
3259                         proto = iph->protocol;
3260         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3261                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3262                         return false;
3263                 iph6 = ipv6_hdr(skb);
3264                 iph_to_flow_copy_v6addrs(fk, iph6);
3265                 noff += sizeof(*iph6);
3266                 proto = iph6->nexthdr;
3267         } else {
3268                 return false;
3269         }
3270         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3271                 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3272
3273         return true;
3274 }
3275
3276 /**
3277  * bond_xmit_hash - generate a hash value based on the xmit policy
3278  * @bond: bonding device
3279  * @skb: buffer to use for headers
3280  *
3281  * This function will extract the necessary headers from the skb buffer and use
3282  * them to generate a hash based on the xmit_policy set in the bonding device
3283  */
3284 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3285 {
3286         struct flow_keys flow;
3287         u32 hash;
3288
3289         if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3290             skb->l4_hash)
3291                 return skb->hash;
3292
3293         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3294             !bond_flow_dissect(bond, skb, &flow))
3295                 return bond_eth_hash(skb);
3296
3297         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3298             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3299                 hash = bond_eth_hash(skb);
3300         else
3301                 hash = (__force u32)flow.ports.ports;
3302         hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3303                 (__force u32)flow_get_u32_src(&flow);
3304         hash ^= (hash >> 16);
3305         hash ^= (hash >> 8);
3306
3307         return hash >> 1;
3308 }
3309
3310 /*-------------------------- Device entry points ----------------------------*/
3311
3312 void bond_work_init_all(struct bonding *bond)
3313 {
3314         INIT_DELAYED_WORK(&bond->mcast_work,
3315                           bond_resend_igmp_join_requests_delayed);
3316         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3317         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3318         INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3319         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3320         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3321 }
3322
3323 static void bond_work_cancel_all(struct bonding *bond)
3324 {
3325         cancel_delayed_work_sync(&bond->mii_work);
3326         cancel_delayed_work_sync(&bond->arp_work);
3327         cancel_delayed_work_sync(&bond->alb_work);
3328         cancel_delayed_work_sync(&bond->ad_work);
3329         cancel_delayed_work_sync(&bond->mcast_work);
3330         cancel_delayed_work_sync(&bond->slave_arr_work);
3331 }
3332
3333 static int bond_open(struct net_device *bond_dev)
3334 {
3335         struct bonding *bond = netdev_priv(bond_dev);
3336         struct list_head *iter;
3337         struct slave *slave;
3338
3339         /* reset slave->backup and slave->inactive */
3340         if (bond_has_slaves(bond)) {
3341                 bond_for_each_slave(bond, slave, iter) {
3342                         if (bond_uses_primary(bond) &&
3343                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3344                                 bond_set_slave_inactive_flags(slave,
3345                                                               BOND_SLAVE_NOTIFY_NOW);
3346                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3347                                 bond_set_slave_active_flags(slave,
3348                                                             BOND_SLAVE_NOTIFY_NOW);
3349                         }
3350                 }
3351         }
3352
3353         if (bond_is_lb(bond)) {
3354                 /* bond_alb_initialize must be called before the timer
3355                  * is started.
3356                  */
3357                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3358                         return -ENOMEM;
3359                 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
3360                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3361         }
3362
3363         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3364                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3365
3366         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3367                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3368                 bond->recv_probe = bond_arp_rcv;
3369         }
3370
3371         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3372                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3373                 /* register to receive LACPDUs */
3374                 bond->recv_probe = bond_3ad_lacpdu_recv;
3375                 bond_3ad_initiate_agg_selection(bond, 1);
3376         }
3377
3378         if (bond_mode_can_use_xmit_hash(bond))
3379                 bond_update_slave_arr(bond, NULL);
3380
3381         return 0;
3382 }
3383
3384 static int bond_close(struct net_device *bond_dev)
3385 {
3386         struct bonding *bond = netdev_priv(bond_dev);
3387
3388         bond_work_cancel_all(bond);
3389         bond->send_peer_notif = 0;
3390         if (bond_is_lb(bond))
3391                 bond_alb_deinitialize(bond);
3392         bond->recv_probe = NULL;
3393
3394         return 0;
3395 }
3396
3397 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3398  * that some drivers can provide 32bit values only.
3399  */
3400 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3401                             const struct rtnl_link_stats64 *_new,
3402                             const struct rtnl_link_stats64 *_old)
3403 {
3404         const u64 *new = (const u64 *)_new;
3405         const u64 *old = (const u64 *)_old;
3406         u64 *res = (u64 *)_res;
3407         int i;
3408
3409         for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3410                 u64 nv = new[i];
3411                 u64 ov = old[i];
3412                 s64 delta = nv - ov;
3413
3414                 /* detects if this particular field is 32bit only */
3415                 if (((nv | ov) >> 32) == 0)
3416                         delta = (s64)(s32)((u32)nv - (u32)ov);
3417
3418                 /* filter anomalies, some drivers reset their stats
3419                  * at down/up events.
3420                  */
3421                 if (delta > 0)
3422                         res[i] += delta;
3423         }
3424 }
3425
3426 static int bond_get_nest_level(struct net_device *bond_dev)
3427 {
3428         struct bonding *bond = netdev_priv(bond_dev);
3429
3430         return bond->nest_level;
3431 }
3432
3433 static void bond_get_stats(struct net_device *bond_dev,
3434                            struct rtnl_link_stats64 *stats)
3435 {
3436         struct bonding *bond = netdev_priv(bond_dev);
3437         struct rtnl_link_stats64 temp;
3438         struct list_head *iter;
3439         struct slave *slave;
3440
3441         spin_lock_nested(&bond->stats_lock, bond_get_nest_level(bond_dev));
3442         memcpy(stats, &bond->bond_stats, sizeof(*stats));
3443
3444         rcu_read_lock();
3445         bond_for_each_slave_rcu(bond, slave, iter) {
3446                 const struct rtnl_link_stats64 *new =
3447                         dev_get_stats(slave->dev, &temp);
3448
3449                 bond_fold_stats(stats, new, &slave->slave_stats);
3450
3451                 /* save off the slave stats for the next run */
3452                 memcpy(&slave->slave_stats, new, sizeof(*new));
3453         }
3454         rcu_read_unlock();
3455
3456         memcpy(&bond->bond_stats, stats, sizeof(*stats));
3457         spin_unlock(&bond->stats_lock);
3458 }
3459
3460 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3461 {
3462         struct bonding *bond = netdev_priv(bond_dev);
3463         struct net_device *slave_dev = NULL;
3464         struct ifbond k_binfo;
3465         struct ifbond __user *u_binfo = NULL;
3466         struct ifslave k_sinfo;
3467         struct ifslave __user *u_sinfo = NULL;
3468         struct mii_ioctl_data *mii = NULL;
3469         struct bond_opt_value newval;
3470         struct net *net;
3471         int res = 0;
3472
3473         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3474
3475         switch (cmd) {
3476         case SIOCGMIIPHY:
3477                 mii = if_mii(ifr);
3478                 if (!mii)
3479                         return -EINVAL;
3480
3481                 mii->phy_id = 0;
3482                 /* Fall Through */
3483         case SIOCGMIIREG:
3484                 /* We do this again just in case we were called by SIOCGMIIREG
3485                  * instead of SIOCGMIIPHY.
3486                  */
3487                 mii = if_mii(ifr);
3488                 if (!mii)
3489                         return -EINVAL;
3490
3491                 if (mii->reg_num == 1) {
3492                         mii->val_out = 0;
3493                         if (netif_carrier_ok(bond->dev))
3494                                 mii->val_out = BMSR_LSTATUS;
3495                 }
3496
3497                 return 0;
3498         case BOND_INFO_QUERY_OLD:
3499         case SIOCBONDINFOQUERY:
3500                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3501
3502                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3503                         return -EFAULT;
3504
3505                 bond_info_query(bond_dev, &k_binfo);
3506                 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3507                         return -EFAULT;
3508
3509                 return 0;
3510         case BOND_SLAVE_INFO_QUERY_OLD:
3511         case SIOCBONDSLAVEINFOQUERY:
3512                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3513
3514                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3515                         return -EFAULT;
3516
3517                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3518                 if (res == 0 &&
3519                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3520                         return -EFAULT;
3521
3522                 return res;
3523         default:
3524                 break;
3525         }
3526
3527         net = dev_net(bond_dev);
3528
3529         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3530                 return -EPERM;
3531
3532         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3533
3534         netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3535
3536         if (!slave_dev)
3537                 return -ENODEV;
3538
3539         netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3540         switch (cmd) {
3541         case BOND_ENSLAVE_OLD:
3542         case SIOCBONDENSLAVE:
3543                 res = bond_enslave(bond_dev, slave_dev, NULL);
3544                 break;
3545         case BOND_RELEASE_OLD:
3546         case SIOCBONDRELEASE:
3547                 res = bond_release(bond_dev, slave_dev);
3548                 break;
3549         case BOND_SETHWADDR_OLD:
3550         case SIOCBONDSETHWADDR:
3551                 bond_set_dev_addr(bond_dev, slave_dev);
3552                 res = 0;
3553                 break;
3554         case BOND_CHANGE_ACTIVE_OLD:
3555         case SIOCBONDCHANGEACTIVE:
3556                 bond_opt_initstr(&newval, slave_dev->name);
3557                 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
3558                                             &newval);
3559                 break;
3560         default:
3561                 res = -EOPNOTSUPP;
3562         }
3563
3564         return res;
3565 }
3566
3567 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3568 {
3569         struct bonding *bond = netdev_priv(bond_dev);
3570
3571         if (change & IFF_PROMISC)
3572                 bond_set_promiscuity(bond,
3573                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3574
3575         if (change & IFF_ALLMULTI)
3576                 bond_set_allmulti(bond,
3577                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3578 }
3579
3580 static void bond_set_rx_mode(struct net_device *bond_dev)
3581 {
3582         struct bonding *bond = netdev_priv(bond_dev);
3583         struct list_head *iter;
3584         struct slave *slave;
3585
3586         rcu_read_lock();
3587         if (bond_uses_primary(bond)) {
3588                 slave = rcu_dereference(bond->curr_active_slave);
3589                 if (slave) {
3590                         dev_uc_sync(slave->dev, bond_dev);
3591                         dev_mc_sync(slave->dev, bond_dev);
3592                 }
3593         } else {
3594                 bond_for_each_slave_rcu(bond, slave, iter) {
3595                         dev_uc_sync_multiple(slave->dev, bond_dev);
3596                         dev_mc_sync_multiple(slave->dev, bond_dev);
3597                 }
3598         }
3599         rcu_read_unlock();
3600 }
3601
3602 static int bond_neigh_init(struct neighbour *n)
3603 {
3604         struct bonding *bond = netdev_priv(n->dev);
3605         const struct net_device_ops *slave_ops;
3606         struct neigh_parms parms;
3607         struct slave *slave;
3608         int ret;
3609
3610         slave = bond_first_slave(bond);
3611         if (!slave)
3612                 return 0;
3613         slave_ops = slave->dev->netdev_ops;
3614         if (!slave_ops->ndo_neigh_setup)
3615                 return 0;
3616
3617         parms.neigh_setup = NULL;
3618         parms.neigh_cleanup = NULL;
3619         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3620         if (ret)
3621                 return ret;
3622
3623         /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3624          * after the last slave has been detached.  Assumes that all slaves
3625          * utilize the same neigh_cleanup (true at this writing as only user
3626          * is ipoib).
3627          */
3628         n->parms->neigh_cleanup = parms.neigh_cleanup;
3629
3630         if (!parms.neigh_setup)
3631                 return 0;
3632
3633         return parms.neigh_setup(n);
3634 }
3635
3636 /* The bonding ndo_neigh_setup is called at init time beofre any
3637  * slave exists. So we must declare proxy setup function which will
3638  * be used at run time to resolve the actual slave neigh param setup.
3639  *
3640  * It's also called by master devices (such as vlans) to setup their
3641  * underlying devices. In that case - do nothing, we're already set up from
3642  * our init.
3643  */
3644 static int bond_neigh_setup(struct net_device *dev,
3645                             struct neigh_parms *parms)
3646 {
3647         /* modify only our neigh_parms */
3648         if (parms->dev == dev)
3649                 parms->neigh_setup = bond_neigh_init;
3650
3651         return 0;
3652 }
3653
3654 /* Change the MTU of all of a master's slaves to match the master */
3655 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3656 {
3657         struct bonding *bond = netdev_priv(bond_dev);
3658         struct slave *slave, *rollback_slave;
3659         struct list_head *iter;
3660         int res = 0;
3661
3662         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3663
3664         bond_for_each_slave(bond, slave, iter) {
3665                 netdev_dbg(bond_dev, "s %p c_m %p\n",
3666                            slave, slave->dev->netdev_ops->ndo_change_mtu);
3667
3668                 res = dev_set_mtu(slave->dev, new_mtu);
3669
3670                 if (res) {
3671                         /* If we failed to set the slave's mtu to the new value
3672                          * we must abort the operation even in ACTIVE_BACKUP
3673                          * mode, because if we allow the backup slaves to have
3674                          * different mtu values than the active slave we'll
3675                          * need to change their mtu when doing a failover. That
3676                          * means changing their mtu from timer context, which
3677                          * is probably not a good idea.
3678                          */
3679                         netdev_dbg(bond_dev, "err %d %s\n", res,
3680                                    slave->dev->name);
3681                         goto unwind;
3682                 }
3683         }
3684
3685         bond_dev->mtu = new_mtu;
3686
3687         return 0;
3688
3689 unwind:
3690         /* unwind from head to the slave that failed */
3691         bond_for_each_slave(bond, rollback_slave, iter) {
3692                 int tmp_res;
3693
3694                 if (rollback_slave == slave)
3695                         break;
3696
3697                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3698                 if (tmp_res) {
3699                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3700                                    tmp_res, rollback_slave->dev->name);
3701                 }
3702         }
3703
3704         return res;
3705 }
3706
3707 /* Change HW address
3708  *
3709  * Note that many devices must be down to change the HW address, and
3710  * downing the master releases all slaves.  We can make bonds full of
3711  * bonding devices to test this, however.
3712  */
3713 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3714 {
3715         struct bonding *bond = netdev_priv(bond_dev);
3716         struct slave *slave, *rollback_slave;
3717         struct sockaddr_storage *ss = addr, tmp_ss;
3718         struct list_head *iter;
3719         int res = 0;
3720
3721         if (BOND_MODE(bond) == BOND_MODE_ALB)
3722                 return bond_alb_set_mac_address(bond_dev, addr);
3723
3724
3725         netdev_dbg(bond_dev, "bond=%p\n", bond);
3726
3727         /* If fail_over_mac is enabled, do nothing and return success.
3728          * Returning an error causes ifenslave to fail.
3729          */
3730         if (bond->params.fail_over_mac &&
3731             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3732                 return 0;
3733
3734         if (!is_valid_ether_addr(ss->__data))
3735                 return -EADDRNOTAVAIL;
3736
3737         bond_for_each_slave(bond, slave, iter) {
3738                 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3739                 res = dev_set_mac_address(slave->dev, addr);
3740                 if (res) {
3741                         /* TODO: consider downing the slave
3742                          * and retry ?
3743                          * User should expect communications
3744                          * breakage anyway until ARP finish
3745                          * updating, so...
3746                          */
3747                         netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3748                         goto unwind;
3749                 }
3750         }
3751
3752         /* success */
3753         memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
3754         return 0;
3755
3756 unwind:
3757         memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
3758         tmp_ss.ss_family = bond_dev->type;
3759
3760         /* unwind from head to the slave that failed */
3761         bond_for_each_slave(bond, rollback_slave, iter) {
3762                 int tmp_res;
3763
3764                 if (rollback_slave == slave)
3765                         break;
3766
3767                 tmp_res = dev_set_mac_address(rollback_slave->dev,
3768                                               (struct sockaddr *)&tmp_ss);
3769                 if (tmp_res) {
3770                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3771                                    tmp_res, rollback_slave->dev->name);
3772                 }
3773         }
3774
3775         return res;
3776 }
3777
3778 /**
3779  * bond_xmit_slave_id - transmit skb through slave with slave_id
3780  * @bond: bonding device that is transmitting
3781  * @skb: buffer to transmit
3782  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3783  *
3784  * This function tries to transmit through slave with slave_id but in case
3785  * it fails, it tries to find the first available slave for transmission.
3786  * The skb is consumed in all cases, thus the function is void.
3787  */
3788 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3789 {
3790         struct list_head *iter;
3791         struct slave *slave;
3792         int i = slave_id;
3793
3794         /* Here we start from the slave with slave_id */
3795         bond_for_each_slave_rcu(bond, slave, iter) {
3796                 if (--i < 0) {
3797                         if (bond_slave_can_tx(slave)) {
3798                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3799                                 return;
3800                         }
3801                 }
3802         }
3803
3804         /* Here we start from the first slave up to slave_id */
3805         i = slave_id;
3806         bond_for_each_slave_rcu(bond, slave, iter) {
3807                 if (--i < 0)
3808                         break;
3809                 if (bond_slave_can_tx(slave)) {
3810                         bond_dev_queue_xmit(bond, skb, slave->dev);
3811                         return;
3812                 }
3813         }
3814         /* no slave that can tx has been found */
3815         bond_tx_drop(bond->dev, skb);
3816 }
3817
3818 /**
3819  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3820  * @bond: bonding device to use
3821  *
3822  * Based on the value of the bonding device's packets_per_slave parameter
3823  * this function generates a slave id, which is usually used as the next
3824  * slave to transmit through.
3825  */
3826 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3827 {
3828         u32 slave_id;
3829         struct reciprocal_value reciprocal_packets_per_slave;
3830         int packets_per_slave = bond->params.packets_per_slave;
3831
3832         switch (packets_per_slave) {
3833         case 0:
3834                 slave_id = prandom_u32();
3835                 break;
3836         case 1:
3837                 slave_id = bond->rr_tx_counter;
3838                 break;
3839         default:
3840                 reciprocal_packets_per_slave =
3841                         bond->params.reciprocal_packets_per_slave;
3842                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3843                                              reciprocal_packets_per_slave);
3844                 break;
3845         }
3846         bond->rr_tx_counter++;
3847
3848         return slave_id;
3849 }
3850
3851 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
3852                                         struct net_device *bond_dev)
3853 {
3854         struct bonding *bond = netdev_priv(bond_dev);
3855         struct iphdr *iph = ip_hdr(skb);
3856         struct slave *slave;
3857         u32 slave_id;
3858
3859         /* Start with the curr_active_slave that joined the bond as the
3860          * default for sending IGMP traffic.  For failover purposes one
3861          * needs to maintain some consistency for the interface that will
3862          * send the join/membership reports.  The curr_active_slave found
3863          * will send all of this type of traffic.
3864          */
3865         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3866                 slave = rcu_dereference(bond->curr_active_slave);
3867                 if (slave)
3868                         bond_dev_queue_xmit(bond, skb, slave->dev);
3869                 else
3870                         bond_xmit_slave_id(bond, skb, 0);
3871         } else {
3872                 int slave_cnt = READ_ONCE(bond->slave_cnt);
3873
3874                 if (likely(slave_cnt)) {
3875                         slave_id = bond_rr_gen_slave_id(bond);
3876                         bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3877                 } else {
3878                         bond_tx_drop(bond_dev, skb);
3879                 }
3880         }
3881
3882         return NETDEV_TX_OK;
3883 }
3884
3885 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3886  * the bond has a usable interface.
3887  */
3888 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
3889                                           struct net_device *bond_dev)
3890 {
3891         struct bonding *bond = netdev_priv(bond_dev);
3892         struct slave *slave;
3893
3894         slave = rcu_dereference(bond->curr_active_slave);
3895         if (slave)
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 /* Use this to update slave_array when (a) it's not appropriate to update
3904  * slave_array right away (note that update_slave_array() may sleep)
3905  * and / or (b) RTNL is not held.
3906  */
3907 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3908 {
3909         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3910 }
3911
3912 /* Slave array work handler. Holds only RTNL */
3913 static void bond_slave_arr_handler(struct work_struct *work)
3914 {
3915         struct bonding *bond = container_of(work, struct bonding,
3916                                             slave_arr_work.work);
3917         int ret;
3918
3919         if (!rtnl_trylock())
3920                 goto err;
3921
3922         ret = bond_update_slave_arr(bond, NULL);
3923         rtnl_unlock();
3924         if (ret) {
3925                 pr_warn_ratelimited("Failed to update slave array from WT\n");
3926                 goto err;
3927         }
3928         return;
3929
3930 err:
3931         bond_slave_arr_work_rearm(bond, 1);
3932 }
3933
3934 /* Build the usable slaves array in control path for modes that use xmit-hash
3935  * to determine the slave interface -
3936  * (a) BOND_MODE_8023AD
3937  * (b) BOND_MODE_XOR
3938  * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
3939  *
3940  * The caller is expected to hold RTNL only and NO other lock!
3941  */
3942 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3943 {
3944         struct slave *slave;
3945         struct list_head *iter;
3946         struct bond_up_slave *new_arr, *old_arr;
3947         int agg_id = 0;
3948         int ret = 0;
3949
3950 #ifdef CONFIG_LOCKDEP
3951         WARN_ON(lockdep_is_held(&bond->mode_lock));
3952 #endif
3953
3954         new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3955                           GFP_KERNEL);
3956         if (!new_arr) {
3957                 ret = -ENOMEM;
3958                 pr_err("Failed to build slave-array.\n");
3959                 goto out;
3960         }
3961         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3962                 struct ad_info ad_info;
3963
3964                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3965                         pr_debug("bond_3ad_get_active_agg_info failed\n");
3966                         kfree_rcu(new_arr, rcu);
3967                         /* No active aggragator means it's not safe to use
3968                          * the previous array.
3969                          */
3970                         old_arr = rtnl_dereference(bond->slave_arr);
3971                         if (old_arr) {
3972                                 RCU_INIT_POINTER(bond->slave_arr, NULL);
3973                                 kfree_rcu(old_arr, rcu);
3974                         }
3975                         goto out;
3976                 }
3977                 agg_id = ad_info.aggregator_id;
3978         }
3979         bond_for_each_slave(bond, slave, iter) {
3980                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3981                         struct aggregator *agg;
3982
3983                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
3984                         if (!agg || agg->aggregator_identifier != agg_id)
3985                                 continue;
3986                 }
3987                 if (!bond_slave_can_tx(slave))
3988                         continue;
3989                 if (skipslave == slave)
3990                         continue;
3991
3992                 netdev_dbg(bond->dev,
3993                            "Adding slave dev %s to tx hash array[%d]\n",
3994                            slave->dev->name, new_arr->count);
3995
3996                 new_arr->arr[new_arr->count++] = slave;
3997         }
3998
3999         old_arr = rtnl_dereference(bond->slave_arr);
4000         rcu_assign_pointer(bond->slave_arr, new_arr);
4001         if (old_arr)
4002                 kfree_rcu(old_arr, rcu);
4003 out:
4004         if (ret != 0 && skipslave) {
4005                 int idx;
4006
4007                 /* Rare situation where caller has asked to skip a specific
4008                  * slave but allocation failed (most likely!). BTW this is
4009                  * only possible when the call is initiated from
4010                  * __bond_release_one(). In this situation; overwrite the
4011                  * skipslave entry in the array with the last entry from the
4012                  * array to avoid a situation where the xmit path may choose
4013                  * this to-be-skipped slave to send a packet out.
4014                  */
4015                 old_arr = rtnl_dereference(bond->slave_arr);
4016                 for (idx = 0; idx < old_arr->count; idx++) {
4017                         if (skipslave == old_arr->arr[idx]) {
4018                                 old_arr->arr[idx] =
4019                                     old_arr->arr[old_arr->count-1];
4020                                 old_arr->count--;
4021                                 break;
4022                         }
4023                 }
4024         }
4025         return ret;
4026 }
4027
4028 /* Use this Xmit function for 3AD as well as XOR modes. The current
4029  * usable slave array is formed in the control path. The xmit function
4030  * just calculates hash and sends the packet out.
4031  */
4032 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
4033                                      struct net_device *dev)
4034 {
4035         struct bonding *bond = netdev_priv(dev);
4036         struct slave *slave;
4037         struct bond_up_slave *slaves;
4038         unsigned int count;
4039
4040         slaves = rcu_dereference(bond->slave_arr);
4041         count = slaves ? READ_ONCE(slaves->count) : 0;
4042         if (likely(count)) {
4043                 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
4044                 bond_dev_queue_xmit(bond, skb, slave->dev);
4045         } else {
4046                 bond_tx_drop(dev, skb);
4047         }
4048
4049         return NETDEV_TX_OK;
4050 }
4051
4052 /* in broadcast mode, we send everything to all usable interfaces. */
4053 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
4054                                        struct net_device *bond_dev)
4055 {
4056         struct bonding *bond = netdev_priv(bond_dev);
4057         struct slave *slave = NULL;
4058         struct list_head *iter;
4059
4060         bond_for_each_slave_rcu(bond, slave, iter) {
4061                 if (bond_is_last_slave(bond, slave))
4062                         break;
4063                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4064                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4065
4066                         if (!skb2) {
4067                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4068                                                     bond_dev->name, __func__);
4069                                 continue;
4070                         }
4071                         bond_dev_queue_xmit(bond, skb2, slave->dev);
4072                 }
4073         }
4074         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4075                 bond_dev_queue_xmit(bond, skb, slave->dev);
4076         else
4077                 bond_tx_drop(bond_dev, skb);
4078
4079         return NETDEV_TX_OK;
4080 }
4081
4082 /*------------------------- Device initialization ---------------------------*/
4083
4084 /* Lookup the slave that corresponds to a qid */
4085 static inline int bond_slave_override(struct bonding *bond,
4086                                       struct sk_buff *skb)
4087 {
4088         struct slave *slave = NULL;
4089         struct list_head *iter;
4090
4091         if (!skb_rx_queue_recorded(skb))
4092                 return 1;
4093
4094         /* Find out if any slaves have the same mapping as this skb. */
4095         bond_for_each_slave_rcu(bond, slave, iter) {
4096                 if (slave->queue_id == skb_get_queue_mapping(skb)) {
4097                         if (bond_slave_is_up(slave) &&
4098                             slave->link == BOND_LINK_UP) {
4099                                 bond_dev_queue_xmit(bond, skb, slave->dev);
4100                                 return 0;
4101                         }
4102                         /* If the slave isn't UP, use default transmit policy. */
4103                         break;
4104                 }
4105         }
4106
4107         return 1;
4108 }
4109
4110
4111 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4112                              struct net_device *sb_dev,
4113                              select_queue_fallback_t fallback)
4114 {
4115         /* This helper function exists to help dev_pick_tx get the correct
4116          * destination queue.  Using a helper function skips a call to
4117          * skb_tx_hash and will put the skbs in the queue we expect on their
4118          * way down to the bonding driver.
4119          */
4120         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4121
4122         /* Save the original txq to restore before passing to the driver */
4123         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
4124
4125         if (unlikely(txq >= dev->real_num_tx_queues)) {
4126                 do {
4127                         txq -= dev->real_num_tx_queues;
4128                 } while (txq >= dev->real_num_tx_queues);
4129         }
4130         return txq;
4131 }
4132
4133 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4134 {
4135         struct bonding *bond = netdev_priv(dev);
4136
4137         if (bond_should_override_tx_queue(bond) &&
4138             !bond_slave_override(bond, skb))
4139                 return NETDEV_TX_OK;
4140
4141         switch (BOND_MODE(bond)) {
4142         case BOND_MODE_ROUNDROBIN:
4143                 return bond_xmit_roundrobin(skb, dev);
4144         case BOND_MODE_ACTIVEBACKUP:
4145                 return bond_xmit_activebackup(skb, dev);
4146         case BOND_MODE_8023AD:
4147         case BOND_MODE_XOR:
4148                 return bond_3ad_xor_xmit(skb, dev);
4149         case BOND_MODE_BROADCAST:
4150                 return bond_xmit_broadcast(skb, dev);
4151         case BOND_MODE_ALB:
4152                 return bond_alb_xmit(skb, dev);
4153         case BOND_MODE_TLB:
4154                 return bond_tlb_xmit(skb, dev);
4155         default:
4156                 /* Should never happen, mode already checked */
4157                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4158                 WARN_ON_ONCE(1);
4159                 bond_tx_drop(dev, skb);
4160                 return NETDEV_TX_OK;
4161         }
4162 }
4163
4164 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4165 {
4166         struct bonding *bond = netdev_priv(dev);
4167         netdev_tx_t ret = NETDEV_TX_OK;
4168
4169         /* If we risk deadlock from transmitting this in the
4170          * netpoll path, tell netpoll to queue the frame for later tx
4171          */
4172         if (unlikely(is_netpoll_tx_blocked(dev)))
4173                 return NETDEV_TX_BUSY;
4174
4175         rcu_read_lock();
4176         if (bond_has_slaves(bond))
4177                 ret = __bond_start_xmit(skb, dev);
4178         else
4179                 bond_tx_drop(dev, skb);
4180         rcu_read_unlock();
4181
4182         return ret;
4183 }
4184
4185 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4186                                            struct ethtool_link_ksettings *cmd)
4187 {
4188         struct bonding *bond = netdev_priv(bond_dev);
4189         unsigned long speed = 0;
4190         struct list_head *iter;
4191         struct slave *slave;
4192
4193         cmd->base.duplex = DUPLEX_UNKNOWN;
4194         cmd->base.port = PORT_OTHER;
4195
4196         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4197          * do not need to check mode.  Though link speed might not represent
4198          * the true receive or transmit bandwidth (not all modes are symmetric)
4199          * this is an accurate maximum.
4200          */
4201         bond_for_each_slave(bond, slave, iter) {
4202                 if (bond_slave_can_tx(slave)) {
4203                         if (slave->speed != SPEED_UNKNOWN)
4204                                 speed += slave->speed;
4205                         if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4206                             slave->duplex != DUPLEX_UNKNOWN)
4207                                 cmd->base.duplex = slave->duplex;
4208                 }
4209         }
4210         cmd->base.speed = speed ? : SPEED_UNKNOWN;
4211
4212         return 0;
4213 }
4214
4215 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4216                                      struct ethtool_drvinfo *drvinfo)
4217 {
4218         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4219         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4220         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4221                  BOND_ABI_VERSION);
4222 }
4223
4224 static const struct ethtool_ops bond_ethtool_ops = {
4225         .get_drvinfo            = bond_ethtool_get_drvinfo,
4226         .get_link               = ethtool_op_get_link,
4227         .get_link_ksettings     = bond_ethtool_get_link_ksettings,
4228 };
4229
4230 static const struct net_device_ops bond_netdev_ops = {
4231         .ndo_init               = bond_init,
4232         .ndo_uninit             = bond_uninit,
4233         .ndo_open               = bond_open,
4234         .ndo_stop               = bond_close,
4235         .ndo_start_xmit         = bond_start_xmit,
4236         .ndo_select_queue       = bond_select_queue,
4237         .ndo_get_stats64        = bond_get_stats,
4238         .ndo_do_ioctl           = bond_do_ioctl,
4239         .ndo_change_rx_flags    = bond_change_rx_flags,
4240         .ndo_set_rx_mode        = bond_set_rx_mode,
4241         .ndo_change_mtu         = bond_change_mtu,
4242         .ndo_set_mac_address    = bond_set_mac_address,
4243         .ndo_neigh_setup        = bond_neigh_setup,
4244         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4245         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4246         .ndo_get_lock_subclass  = bond_get_nest_level,
4247 #ifdef CONFIG_NET_POLL_CONTROLLER
4248         .ndo_netpoll_setup      = bond_netpoll_setup,
4249         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4250         .ndo_poll_controller    = bond_poll_controller,
4251 #endif
4252         .ndo_add_slave          = bond_enslave,
4253         .ndo_del_slave          = bond_release,
4254         .ndo_fix_features       = bond_fix_features,
4255         .ndo_features_check     = passthru_features_check,
4256 };
4257
4258 static const struct device_type bond_type = {
4259         .name = "bond",
4260 };
4261
4262 static void bond_destructor(struct net_device *bond_dev)
4263 {
4264         struct bonding *bond = netdev_priv(bond_dev);
4265         if (bond->wq)
4266                 destroy_workqueue(bond->wq);
4267 }
4268
4269 void bond_setup(struct net_device *bond_dev)
4270 {
4271         struct bonding *bond = netdev_priv(bond_dev);
4272
4273         spin_lock_init(&bond->mode_lock);
4274         spin_lock_init(&bond->stats_lock);
4275         bond->params = bonding_defaults;
4276
4277         /* Initialize pointers */
4278         bond->dev = bond_dev;
4279
4280         /* Initialize the device entry points */
4281         ether_setup(bond_dev);
4282         bond_dev->max_mtu = ETH_MAX_MTU;
4283         bond_dev->netdev_ops = &bond_netdev_ops;
4284         bond_dev->ethtool_ops = &bond_ethtool_ops;
4285
4286         bond_dev->needs_free_netdev = true;
4287         bond_dev->priv_destructor = bond_destructor;
4288
4289         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4290
4291         /* Initialize the device options */
4292         bond_dev->flags |= IFF_MASTER;
4293         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4294         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4295
4296         /* don't acquire bond device's netif_tx_lock when transmitting */
4297         bond_dev->features |= NETIF_F_LLTX;
4298
4299         /* By default, we declare the bond to be fully
4300          * VLAN hardware accelerated capable. Special
4301          * care is taken in the various xmit functions
4302          * when there are slaves that are not hw accel
4303          * capable
4304          */
4305
4306         /* Don't allow bond devices to change network namespaces. */
4307         bond_dev->features |= NETIF_F_NETNS_LOCAL;
4308
4309         bond_dev->hw_features = BOND_VLAN_FEATURES |
4310                                 NETIF_F_HW_VLAN_CTAG_RX |
4311                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4312
4313         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4;
4314         bond_dev->features |= bond_dev->hw_features;
4315         bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
4316 }
4317
4318 /* Destroy a bonding device.
4319  * Must be under rtnl_lock when this function is called.
4320  */
4321 static void bond_uninit(struct net_device *bond_dev)
4322 {
4323         struct bonding *bond = netdev_priv(bond_dev);
4324         struct list_head *iter;
4325         struct slave *slave;
4326         struct bond_up_slave *arr;
4327
4328         bond_netpoll_cleanup(bond_dev);
4329
4330         /* Release the bonded slaves */
4331         bond_for_each_slave(bond, slave, iter)
4332                 __bond_release_one(bond_dev, slave->dev, true, true);
4333         netdev_info(bond_dev, "Released all slaves\n");
4334
4335         arr = rtnl_dereference(bond->slave_arr);
4336         if (arr) {
4337                 RCU_INIT_POINTER(bond->slave_arr, NULL);
4338                 kfree_rcu(arr, rcu);
4339         }
4340
4341         list_del(&bond->bond_list);
4342
4343         bond_debug_unregister(bond);
4344 }
4345
4346 /*------------------------- Module initialization ---------------------------*/
4347
4348 static int bond_check_params(struct bond_params *params)
4349 {
4350         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4351         struct bond_opt_value newval;
4352         const struct bond_opt_value *valptr;
4353         int arp_all_targets_value = 0;
4354         u16 ad_actor_sys_prio = 0;
4355         u16 ad_user_port_key = 0;
4356         __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
4357         int arp_ip_count;
4358         int bond_mode   = BOND_MODE_ROUNDROBIN;
4359         int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4360         int lacp_fast = 0;
4361         int tlb_dynamic_lb;
4362
4363         /* Convert string parameters. */
4364         if (mode) {
4365                 bond_opt_initstr(&newval, mode);
4366                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4367                 if (!valptr) {
4368                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4369                         return -EINVAL;
4370                 }
4371                 bond_mode = valptr->value;
4372         }
4373
4374         if (xmit_hash_policy) {
4375                 if (bond_mode == BOND_MODE_ROUNDROBIN ||
4376                     bond_mode == BOND_MODE_ACTIVEBACKUP ||
4377                     bond_mode == BOND_MODE_BROADCAST) {
4378                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4379                                 bond_mode_name(bond_mode));
4380                 } else {
4381                         bond_opt_initstr(&newval, xmit_hash_policy);
4382                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4383                                                 &newval);
4384                         if (!valptr) {
4385                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4386                                        xmit_hash_policy);
4387                                 return -EINVAL;
4388                         }
4389                         xmit_hashtype = valptr->value;
4390                 }
4391         }
4392
4393         if (lacp_rate) {
4394                 if (bond_mode != BOND_MODE_8023AD) {
4395                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4396                                 bond_mode_name(bond_mode));
4397                 } else {
4398                         bond_opt_initstr(&newval, lacp_rate);
4399                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4400                                                 &newval);
4401                         if (!valptr) {
4402                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4403                                        lacp_rate);
4404                                 return -EINVAL;
4405                         }
4406                         lacp_fast = valptr->value;
4407                 }
4408         }
4409
4410         if (ad_select) {
4411                 bond_opt_initstr(&newval, ad_select);
4412                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4413                                         &newval);
4414                 if (!valptr) {
4415                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4416                         return -EINVAL;
4417                 }
4418                 params->ad_select = valptr->value;
4419                 if (bond_mode != BOND_MODE_8023AD)
4420                         pr_warn("ad_select param only affects 802.3ad mode\n");
4421         } else {
4422                 params->ad_select = BOND_AD_STABLE;
4423         }
4424
4425         if (max_bonds < 0) {
4426                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4427                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4428                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4429         }
4430
4431         if (miimon < 0) {
4432                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4433                         miimon, INT_MAX);
4434                 miimon = 0;
4435         }
4436
4437         if (updelay < 0) {
4438                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4439                         updelay, INT_MAX);
4440                 updelay = 0;
4441         }
4442
4443         if (downdelay < 0) {
4444                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4445                         downdelay, INT_MAX);
4446                 downdelay = 0;
4447         }
4448
4449         if ((use_carrier != 0) && (use_carrier != 1)) {
4450                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4451                         use_carrier);
4452                 use_carrier = 1;
4453         }
4454
4455         if (num_peer_notif < 0 || num_peer_notif > 255) {
4456                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4457                         num_peer_notif);
4458                 num_peer_notif = 1;
4459         }
4460
4461         /* reset values for 802.3ad/TLB/ALB */
4462         if (!bond_mode_uses_arp(bond_mode)) {
4463                 if (!miimon) {
4464                         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");
4465                         pr_warn("Forcing miimon to 100msec\n");
4466                         miimon = BOND_DEFAULT_MIIMON;
4467                 }
4468         }
4469
4470         if (tx_queues < 1 || tx_queues > 255) {
4471                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4472                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4473                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4474         }
4475
4476         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4477                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4478                         all_slaves_active);
4479                 all_slaves_active = 0;
4480         }
4481
4482         if (resend_igmp < 0 || resend_igmp > 255) {
4483                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4484                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4485                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4486         }
4487
4488         bond_opt_initval(&newval, packets_per_slave);
4489         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4490                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4491                         packets_per_slave, USHRT_MAX);
4492                 packets_per_slave = 1;
4493         }
4494
4495         if (bond_mode == BOND_MODE_ALB) {
4496                 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",
4497                           updelay);
4498         }
4499
4500         if (!miimon) {
4501                 if (updelay || downdelay) {
4502                         /* just warn the user the up/down delay will have
4503                          * no effect since miimon is zero...
4504                          */
4505                         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",
4506                                 updelay, downdelay);
4507                 }
4508         } else {
4509                 /* don't allow arp monitoring */
4510                 if (arp_interval) {
4511                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4512                                 miimon, arp_interval);
4513                         arp_interval = 0;
4514                 }
4515
4516                 if ((updelay % miimon) != 0) {
4517                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4518                                 updelay, miimon, (updelay / miimon) * miimon);
4519                 }
4520
4521                 updelay /= miimon;
4522
4523                 if ((downdelay % miimon) != 0) {
4524                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4525                                 downdelay, miimon,
4526                                 (downdelay / miimon) * miimon);
4527                 }
4528
4529                 downdelay /= miimon;
4530         }
4531
4532         if (arp_interval < 0) {
4533                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4534                         arp_interval, INT_MAX);
4535                 arp_interval = 0;
4536         }
4537
4538         for (arp_ip_count = 0, i = 0;
4539              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4540                 __be32 ip;
4541
4542                 /* not a complete check, but good enough to catch mistakes */
4543                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4544                     !bond_is_ip_target_ok(ip)) {
4545                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4546                                 arp_ip_target[i]);
4547                         arp_interval = 0;
4548                 } else {
4549                         if (bond_get_targets_ip(arp_target, ip) == -1)
4550                                 arp_target[arp_ip_count++] = ip;
4551                         else
4552                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4553                                         &ip);
4554                 }
4555         }
4556
4557         if (arp_interval && !arp_ip_count) {
4558                 /* don't allow arping if no arp_ip_target given... */
4559                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4560                         arp_interval);
4561                 arp_interval = 0;
4562         }
4563
4564         if (arp_validate) {
4565                 if (!arp_interval) {
4566                         pr_err("arp_validate requires arp_interval\n");
4567                         return -EINVAL;
4568                 }
4569
4570                 bond_opt_initstr(&newval, arp_validate);
4571                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4572                                         &newval);
4573                 if (!valptr) {
4574                         pr_err("Error: invalid arp_validate \"%s\"\n",
4575                                arp_validate);
4576                         return -EINVAL;
4577                 }
4578                 arp_validate_value = valptr->value;
4579         } else {
4580                 arp_validate_value = 0;
4581         }
4582
4583         if (arp_all_targets) {
4584                 bond_opt_initstr(&newval, arp_all_targets);
4585                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4586                                         &newval);
4587                 if (!valptr) {
4588                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4589                                arp_all_targets);
4590                         arp_all_targets_value = 0;
4591                 } else {
4592                         arp_all_targets_value = valptr->value;
4593                 }
4594         }
4595
4596         if (miimon) {
4597                 pr_info("MII link monitoring set to %d ms\n", miimon);
4598         } else if (arp_interval) {
4599                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4600                                           arp_validate_value);
4601                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4602                         arp_interval, valptr->string, arp_ip_count);
4603
4604                 for (i = 0; i < arp_ip_count; i++)
4605                         pr_cont(" %s", arp_ip_target[i]);
4606
4607                 pr_cont("\n");
4608
4609         } else if (max_bonds) {
4610                 /* miimon and arp_interval not set, we need one so things
4611                  * work as expected, see bonding.txt for details
4612                  */
4613                 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");
4614         }
4615
4616         if (primary && !bond_mode_uses_primary(bond_mode)) {
4617                 /* currently, using a primary only makes sense
4618                  * in active backup, TLB or ALB modes
4619                  */
4620                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4621                         primary, bond_mode_name(bond_mode));
4622                 primary = NULL;
4623         }
4624
4625         if (primary && primary_reselect) {
4626                 bond_opt_initstr(&newval, primary_reselect);
4627                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4628                                         &newval);
4629                 if (!valptr) {
4630                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4631                                primary_reselect);
4632                         return -EINVAL;
4633                 }
4634                 primary_reselect_value = valptr->value;
4635         } else {
4636                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4637         }
4638
4639         if (fail_over_mac) {
4640                 bond_opt_initstr(&newval, fail_over_mac);
4641                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4642                                         &newval);
4643                 if (!valptr) {
4644                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4645                                fail_over_mac);
4646                         return -EINVAL;
4647                 }
4648                 fail_over_mac_value = valptr->value;
4649                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4650                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4651         } else {
4652                 fail_over_mac_value = BOND_FOM_NONE;
4653         }
4654
4655         bond_opt_initstr(&newval, "default");
4656         valptr = bond_opt_parse(
4657                         bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4658                                      &newval);
4659         if (!valptr) {
4660                 pr_err("Error: No ad_actor_sys_prio default value");
4661                 return -EINVAL;
4662         }
4663         ad_actor_sys_prio = valptr->value;
4664
4665         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4666                                 &newval);
4667         if (!valptr) {
4668                 pr_err("Error: No ad_user_port_key default value");
4669                 return -EINVAL;
4670         }
4671         ad_user_port_key = valptr->value;
4672
4673         bond_opt_initstr(&newval, "default");
4674         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
4675         if (!valptr) {
4676                 pr_err("Error: No tlb_dynamic_lb default value");
4677                 return -EINVAL;
4678         }
4679         tlb_dynamic_lb = valptr->value;
4680
4681         if (lp_interval == 0) {
4682                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4683                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4684                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4685         }
4686
4687         /* fill params struct with the proper values */
4688         params->mode = bond_mode;
4689         params->xmit_policy = xmit_hashtype;
4690         params->miimon = miimon;
4691         params->num_peer_notif = num_peer_notif;
4692         params->arp_interval = arp_interval;
4693         params->arp_validate = arp_validate_value;
4694         params->arp_all_targets = arp_all_targets_value;
4695         params->updelay = updelay;
4696         params->downdelay = downdelay;
4697         params->use_carrier = use_carrier;
4698         params->lacp_fast = lacp_fast;
4699         params->primary[0] = 0;
4700         params->primary_reselect = primary_reselect_value;
4701         params->fail_over_mac = fail_over_mac_value;
4702         params->tx_queues = tx_queues;
4703         params->all_slaves_active = all_slaves_active;
4704         params->resend_igmp = resend_igmp;
4705         params->min_links = min_links;
4706         params->lp_interval = lp_interval;
4707         params->packets_per_slave = packets_per_slave;
4708         params->tlb_dynamic_lb = tlb_dynamic_lb;
4709         params->ad_actor_sys_prio = ad_actor_sys_prio;
4710         eth_zero_addr(params->ad_actor_system);
4711         params->ad_user_port_key = ad_user_port_key;
4712         if (packets_per_slave > 0) {
4713                 params->reciprocal_packets_per_slave =
4714                         reciprocal_value(packets_per_slave);
4715         } else {
4716                 /* reciprocal_packets_per_slave is unused if
4717                  * packets_per_slave is 0 or 1, just initialize it
4718                  */
4719                 params->reciprocal_packets_per_slave =
4720                         (struct reciprocal_value) { 0 };
4721         }
4722
4723         if (primary) {
4724                 strncpy(params->primary, primary, IFNAMSIZ);
4725                 params->primary[IFNAMSIZ - 1] = 0;
4726         }
4727
4728         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4729
4730         return 0;
4731 }
4732
4733 /* Called from registration process */
4734 static int bond_init(struct net_device *bond_dev)
4735 {
4736         struct bonding *bond = netdev_priv(bond_dev);
4737         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4738
4739         netdev_dbg(bond_dev, "Begin bond_init\n");
4740
4741         bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
4742         if (!bond->wq)
4743                 return -ENOMEM;
4744
4745         bond->nest_level = SINGLE_DEPTH_NESTING;
4746         netdev_lockdep_set_classes(bond_dev);
4747
4748         list_add_tail(&bond->bond_list, &bn->dev_list);
4749
4750         bond_prepare_sysfs_group(bond);
4751
4752         bond_debug_register(bond);
4753
4754         /* Ensure valid dev_addr */
4755         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4756             bond_dev->addr_assign_type == NET_ADDR_PERM)
4757                 eth_hw_addr_random(bond_dev);
4758
4759         return 0;
4760 }
4761
4762 unsigned int bond_get_num_tx_queues(void)
4763 {
4764         return tx_queues;
4765 }
4766
4767 /* Create a new bond based on the specified name and bonding parameters.
4768  * If name is NULL, obtain a suitable "bond%d" name for us.
4769  * Caller must NOT hold rtnl_lock; we need to release it here before we
4770  * set up our sysfs entries.
4771  */
4772 int bond_create(struct net *net, const char *name)
4773 {
4774         struct net_device *bond_dev;
4775         struct bonding *bond;
4776         struct alb_bond_info *bond_info;
4777         int res;
4778
4779         rtnl_lock();
4780
4781         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4782                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
4783                                    bond_setup, tx_queues);
4784         if (!bond_dev) {
4785                 pr_err("%s: eek! can't alloc netdev!\n", name);
4786                 rtnl_unlock();
4787                 return -ENOMEM;
4788         }
4789
4790         /*
4791          * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4792          * It is set to 0 by default which is wrong.
4793          */
4794         bond = netdev_priv(bond_dev);
4795         bond_info = &(BOND_ALB_INFO(bond));
4796         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4797
4798         dev_net_set(bond_dev, net);
4799         bond_dev->rtnl_link_ops = &bond_link_ops;
4800
4801         res = register_netdevice(bond_dev);
4802
4803         netif_carrier_off(bond_dev);
4804
4805         bond_work_init_all(bond);
4806
4807         rtnl_unlock();
4808         if (res < 0)
4809                 free_netdev(bond_dev);
4810         return res;
4811 }
4812
4813 static int __net_init bond_net_init(struct net *net)
4814 {
4815         struct bond_net *bn = net_generic(net, bond_net_id);
4816
4817         bn->net = net;
4818         INIT_LIST_HEAD(&bn->dev_list);
4819
4820         bond_create_proc_dir(bn);
4821         bond_create_sysfs(bn);
4822
4823         return 0;
4824 }
4825
4826 static void __net_exit bond_net_exit(struct net *net)
4827 {
4828         struct bond_net *bn = net_generic(net, bond_net_id);
4829         struct bonding *bond, *tmp_bond;
4830         LIST_HEAD(list);
4831
4832         bond_destroy_sysfs(bn);
4833
4834         /* Kill off any bonds created after unregistering bond rtnl ops */
4835         rtnl_lock();
4836         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4837                 unregister_netdevice_queue(bond->dev, &list);
4838         unregister_netdevice_many(&list);
4839         rtnl_unlock();
4840
4841         bond_destroy_proc_dir(bn);
4842 }
4843
4844 static struct pernet_operations bond_net_ops = {
4845         .init = bond_net_init,
4846         .exit = bond_net_exit,
4847         .id   = &bond_net_id,
4848         .size = sizeof(struct bond_net),
4849 };
4850
4851 static int __init bonding_init(void)
4852 {
4853         int i;
4854         int res;
4855
4856         pr_info("%s", bond_version);
4857
4858         res = bond_check_params(&bonding_defaults);
4859         if (res)
4860                 goto out;
4861
4862         res = register_pernet_subsys(&bond_net_ops);
4863         if (res)
4864                 goto out;
4865
4866         res = bond_netlink_init();
4867         if (res)
4868                 goto err_link;
4869
4870         bond_create_debugfs();
4871
4872         for (i = 0; i < max_bonds; i++) {
4873                 res = bond_create(&init_net, NULL);
4874                 if (res)
4875                         goto err;
4876         }
4877
4878         register_netdevice_notifier(&bond_netdev_notifier);
4879 out:
4880         return res;
4881 err:
4882         bond_destroy_debugfs();
4883         bond_netlink_fini();
4884 err_link:
4885         unregister_pernet_subsys(&bond_net_ops);
4886         goto out;
4887
4888 }
4889
4890 static void __exit bonding_exit(void)
4891 {
4892         unregister_netdevice_notifier(&bond_netdev_notifier);
4893
4894         bond_destroy_debugfs();
4895
4896         bond_netlink_fini();
4897         unregister_pernet_subsys(&bond_net_ops);
4898
4899 #ifdef CONFIG_NET_POLL_CONTROLLER
4900         /* Make sure we don't have an imbalance on our netpoll blocking */
4901         WARN_ON(atomic_read(&netpoll_block_tx));
4902 #endif
4903 }
4904
4905 module_init(bonding_init);
4906 module_exit(bonding_exit);
4907 MODULE_LICENSE("GPL");
4908 MODULE_VERSION(DRV_VERSION);
4909 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4910 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");