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 int bond_set_dev_addr(struct net_device *bond_dev,
 613                              struct net_device *slave_dev)
 614 {
 615         int err;
 616
 617         netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n",
 618                    bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len);
 619         err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
 620         if (err)
 621                 return err;
 622
 623         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 624         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 625         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 626         return 0;
 627 }
 628
 629 static struct slave *bond_get_old_active(struct bonding *bond,
 630                                          struct slave *new_active)
 631 {
 632         struct slave *slave;
 633         struct list_head *iter;
 634
 635         bond_for_each_slave(bond, slave, iter) {
 636                 if (slave == new_active)
 637                         continue;
 638
 639                 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
 640                         return slave;
 641         }
 642
 643         return NULL;
 644 }
 645
 646 /* bond_do_fail_over_mac
 647  *
 648  * Perform special MAC address swapping for fail_over_mac settings
 649  *
 650  * Called with RTNL
 651  */
 652 static void bond_do_fail_over_mac(struct bonding *bond,
 653                                   struct slave *new_active,
 654                                   struct slave *old_active)
 655 {
 656         u8 tmp_mac[MAX_ADDR_LEN];
 657         struct sockaddr_storage ss;
 658         int rv;
 659
 660         switch (bond->params.fail_over_mac) {
 661         case BOND_FOM_ACTIVE:
 662                 if (new_active) {
 663                         rv = bond_set_dev_addr(bond->dev, new_active->dev);
 664                         if (rv)
 665                                 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 666                                            -rv, bond->dev->name);
 667                 }
 668                 break;
 669         case BOND_FOM_FOLLOW:
 670                 /* if new_active && old_active, swap them
 671                  * if just old_active, do nothing (going to no active slave)
 672                  * if just new_active, set new_active to bond's MAC
 673                  */
 674                 if (!new_active)
 675                         return;
 676
 677                 if (!old_active)
 678                         old_active = bond_get_old_active(bond, new_active);
 679
 680                 if (old_active) {
 681                         bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
 682                                           new_active->dev->addr_len);
 683                         bond_hw_addr_copy(ss.__data,
 684                                           old_active->dev->dev_addr,
 685                                           old_active->dev->addr_len);
 686                         ss.ss_family = new_active->dev->type;
 687                 } else {
 688                         bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
 689                                           bond->dev->addr_len);
 690                         ss.ss_family = bond->dev->type;
 691                 }
 692
 693                 rv = dev_set_mac_address(new_active->dev,
 694                                          (struct sockaddr *)&ss, NULL);
 695                 if (rv) {
 696                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 697                                    -rv, new_active->dev->name);
 698                         goto out;
 699                 }
 700
 701                 if (!old_active)
 702                         goto out;
 703
 704                 bond_hw_addr_copy(ss.__data, tmp_mac,
 705                                   new_active->dev->addr_len);
 706                 ss.ss_family = old_active->dev->type;
 707
 708                 rv = dev_set_mac_address(old_active->dev,
 709                                          (struct sockaddr *)&ss, NULL);
 710                 if (rv)
 711                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 712                                    -rv, new_active->dev->name);
 713 out:
 714                 break;
 715         default:
 716                 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
 717                            bond->params.fail_over_mac);
 718                 break;
 719         }
 720
 721 }
 722
 723 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
 724 {
 725         struct slave *prim = rtnl_dereference(bond->primary_slave);
 726         struct slave *curr = rtnl_dereference(bond->curr_active_slave);
 727
 728         if (!prim || prim->link != BOND_LINK_UP) {
 729                 if (!curr || curr->link != BOND_LINK_UP)
 730                         return NULL;
 731                 return curr;
 732         }
 733
 734         if (bond->force_primary) {
 735                 bond->force_primary = false;
 736                 return prim;
 737         }
 738
 739         if (!curr || curr->link != BOND_LINK_UP)
 740                 return prim;
 741
 742         /* At this point, prim and curr are both up */
 743         switch (bond->params.primary_reselect) {
 744         case BOND_PRI_RESELECT_ALWAYS:
 745                 return prim;
 746         case BOND_PRI_RESELECT_BETTER:
 747                 if (prim->speed < curr->speed)
 748                         return curr;
 749                 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
 750                         return curr;
 751                 return prim;
 752         case BOND_PRI_RESELECT_FAILURE:
 753                 return curr;
 754         default:
 755                 netdev_err(bond->dev, "impossible primary_reselect %d\n",
 756                            bond->params.primary_reselect);
 757                 return curr;
 758         }
 759 }
 760
 761 /**
 762  * bond_find_best_slave - select the best available slave to be the active one
 763  * @bond: our bonding struct
 764  */
 765 static struct slave *bond_find_best_slave(struct bonding *bond)
 766 {
 767         struct slave *slave, *bestslave = NULL;
 768         struct list_head *iter;
 769         int mintime = bond->params.updelay;
 770
 771         slave = bond_choose_primary_or_current(bond);
 772         if (slave)
 773                 return slave;
 774
 775         bond_for_each_slave(bond, slave, iter) {
 776                 if (slave->link == BOND_LINK_UP)
 777                         return slave;
 778                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
 779                     slave->delay < mintime) {
 780                         mintime = slave->delay;
 781                         bestslave = slave;
 782                 }
 783         }
 784
 785         return bestslave;
 786 }
 787
 788 static bool bond_should_notify_peers(struct bonding *bond)
 789 {
 790         struct slave *slave;
 791
 792         rcu_read_lock();
 793         slave = rcu_dereference(bond->curr_active_slave);
 794         rcu_read_unlock();
 795
 796         netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
 797                    slave ? slave->dev->name : "NULL");
 798
 799         if (!slave || !bond->send_peer_notif ||
 800             !netif_carrier_ok(bond->dev) ||
 801             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 802                 return false;
 803
 804         return true;
 805 }
 806
 807 /**
 808  * change_active_interface - change the active slave into the specified one
 809  * @bond: our bonding struct
 810  * @new: the new slave to make the active one
 811  *
 812  * Set the new slave to the bond's settings and unset them on the old
 813  * curr_active_slave.
 814  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 815  *
 816  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 817  * because it is apparently the best available slave we have, even though its
 818  * updelay hasn't timed out yet.
 819  *
 820  * Caller must hold RTNL.
 821  */
 822 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 823 {
 824         struct slave *old_active;
 825
 826         ASSERT_RTNL();
 827
 828         old_active = rtnl_dereference(bond->curr_active_slave);
 829
 830         if (old_active == new_active)
 831                 return;
 832
 833         if (new_active) {
 834                 new_active->last_link_up = jiffies;
 835
 836                 if (new_active->link == BOND_LINK_BACK) {
 837                         if (bond_uses_primary(bond)) {
 838                                 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
 839                                             new_active->dev->name,
 840                                             (bond->params.updelay - new_active->delay) * bond->params.miimon);
 841                         }
 842
 843                         new_active->delay = 0;
 844                         bond_set_slave_link_state(new_active, BOND_LINK_UP,
 845                                                   BOND_SLAVE_NOTIFY_NOW);
 846
 847                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 848                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 849
 850                         if (bond_is_lb(bond))
 851                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 852                 } else {
 853                         if (bond_uses_primary(bond)) {
 854                                 netdev_info(bond->dev, "making interface %s the new active one\n",
 855                                             new_active->dev->name);
 856                         }
 857                 }
 858         }
 859
 860         if (bond_uses_primary(bond))
 861                 bond_hw_addr_swap(bond, new_active, old_active);
 862
 863         if (bond_is_lb(bond)) {
 864                 bond_alb_handle_active_change(bond, new_active);
 865                 if (old_active)
 866                         bond_set_slave_inactive_flags(old_active,
 867                                                       BOND_SLAVE_NOTIFY_NOW);
 868                 if (new_active)
 869                         bond_set_slave_active_flags(new_active,
 870                                                     BOND_SLAVE_NOTIFY_NOW);
 871         } else {
 872                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 873         }
 874
 875         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
 876                 if (old_active)
 877                         bond_set_slave_inactive_flags(old_active,
 878                                                       BOND_SLAVE_NOTIFY_NOW);
 879
 880                 if (new_active) {
 881                         bool should_notify_peers = false;
 882
 883                         bond_set_slave_active_flags(new_active,
 884                                                     BOND_SLAVE_NOTIFY_NOW);
 885
 886                         if (bond->params.fail_over_mac)
 887                                 bond_do_fail_over_mac(bond, new_active,
 888                                                       old_active);
 889
 890                         if (netif_running(bond->dev)) {
 891                                 bond->send_peer_notif =
 892                                         bond->params.num_peer_notif;
 893                                 should_notify_peers =
 894                                         bond_should_notify_peers(bond);
 895                         }
 896
 897                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 898                         if (should_notify_peers)
 899                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 900                                                          bond->dev);
 901                 }
 902         }
 903
 904         /* resend IGMP joins since active slave has changed or
 905          * all were sent on curr_active_slave.
 906          * resend only if bond is brought up with the affected
 907          * bonding modes and the retransmission is enabled
 908          */
 909         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 910             ((bond_uses_primary(bond) && new_active) ||
 911              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
 912                 bond->igmp_retrans = bond->params.resend_igmp;
 913                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 914         }
 915 }
 916
 917 /**
 918  * bond_select_active_slave - select a new active slave, if needed
 919  * @bond: our bonding struct
 920  *
 921  * This functions should be called when one of the following occurs:
 922  * - The old curr_active_slave has been released or lost its link.
 923  * - The primary_slave has got its link back.
 924  * - A slave has got its link back and there's no old curr_active_slave.
 925  *
 926  * Caller must hold RTNL.
 927  */
 928 void bond_select_active_slave(struct bonding *bond)
 929 {
 930         struct slave *best_slave;
 931         int rv;
 932
 933         ASSERT_RTNL();
 934
 935         best_slave = bond_find_best_slave(bond);
 936         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
 937                 bond_change_active_slave(bond, best_slave);
 938                 rv = bond_set_carrier(bond);
 939                 if (!rv)
 940                         return;
 941
 942                 if (netif_carrier_ok(bond->dev))
 943                         netdev_info(bond->dev, "first active interface up!\n");
 944                 else
 945                         netdev_info(bond->dev, "now running without any active interface!\n");
 946         }
 947 }
 948
 949 #ifdef CONFIG_NET_POLL_CONTROLLER
 950 static inline int slave_enable_netpoll(struct slave *slave)
 951 {
 952         struct netpoll *np;
 953         int err = 0;
 954
 955         np = kzalloc(sizeof(*np), GFP_KERNEL);
 956         err = -ENOMEM;
 957         if (!np)
 958                 goto out;
 959
 960         err = __netpoll_setup(np, slave->dev);
 961         if (err) {
 962                 kfree(np);
 963                 goto out;
 964         }
 965         slave->np = np;
 966 out:
 967         return err;
 968 }
 969 static inline void slave_disable_netpoll(struct slave *slave)
 970 {
 971         struct netpoll *np = slave->np;
 972
 973         if (!np)
 974                 return;
 975
 976         slave->np = NULL;
 977
 978         __netpoll_free(np);
 979 }
 980
 981 static void bond_poll_controller(struct net_device *bond_dev)
 982 {
 983         struct bonding *bond = netdev_priv(bond_dev);
 984         struct slave *slave = NULL;
 985         struct list_head *iter;
 986         struct ad_info ad_info;
 987
 988         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 989                 if (bond_3ad_get_active_agg_info(bond, &ad_info))
 990                         return;
 991
 992         bond_for_each_slave_rcu(bond, slave, iter) {
 993                 if (!bond_slave_is_up(slave))
 994                         continue;
 995
 996                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 997                         struct aggregator *agg =
 998                             SLAVE_AD_INFO(slave)->port.aggregator;
 999
1000                         if (agg &&
1001                             agg->aggregator_identifier != ad_info.aggregator_id)
1002                                 continue;
1003                 }
1004
1005                 netpoll_poll_dev(slave->dev);
1006         }
1007 }
1008
1009 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1010 {
1011         struct bonding *bond = netdev_priv(bond_dev);
1012         struct list_head *iter;
1013         struct slave *slave;
1014
1015         bond_for_each_slave(bond, slave, iter)
1016                 if (bond_slave_is_up(slave))
1017                         slave_disable_netpoll(slave);
1018 }
1019
1020 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1021 {
1022         struct bonding *bond = netdev_priv(dev);
1023         struct list_head *iter;
1024         struct slave *slave;
1025         int err = 0;
1026
1027         bond_for_each_slave(bond, slave, iter) {
1028                 err = slave_enable_netpoll(slave);
1029                 if (err) {
1030                         bond_netpoll_cleanup(dev);
1031                         break;
1032                 }
1033         }
1034         return err;
1035 }
1036 #else
1037 static inline int slave_enable_netpoll(struct slave *slave)
1038 {
1039         return 0;
1040 }
1041 static inline void slave_disable_netpoll(struct slave *slave)
1042 {
1043 }
1044 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1045 {
1046 }
1047 #endif
1048
1049 /*---------------------------------- IOCTL ----------------------------------*/
1050
1051 static netdev_features_t bond_fix_features(struct net_device *dev,
1052                                            netdev_features_t features)
1053 {
1054         struct bonding *bond = netdev_priv(dev);
1055         struct list_head *iter;
1056         netdev_features_t mask;
1057         struct slave *slave;
1058
1059         mask = features;
1060
1061         features &= ~NETIF_F_ONE_FOR_ALL;
1062         features |= NETIF_F_ALL_FOR_ALL;
1063
1064         bond_for_each_slave(bond, slave, iter) {
1065                 features = netdev_increment_features(features,
1066                                                      slave->dev->features,
1067                                                      mask);
1068         }
1069         features = netdev_add_tso_features(features, mask);
1070
1071         return features;
1072 }
1073
1074 #define BOND_VLAN_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1075                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1076                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1077
1078 #define BOND_ENC_FEATURES       (NETIF_F_HW_CSUM | NETIF_F_SG | \
1079                                  NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1080
1081 static void bond_compute_features(struct bonding *bond)
1082 {
1083         unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1084                                         IFF_XMIT_DST_RELEASE_PERM;
1085         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1086         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1087         struct net_device *bond_dev = bond->dev;
1088         struct list_head *iter;
1089         struct slave *slave;
1090         unsigned short max_hard_header_len = ETH_HLEN;
1091         unsigned int gso_max_size = GSO_MAX_SIZE;
1092         u16 gso_max_segs = GSO_MAX_SEGS;
1093
1094         if (!bond_has_slaves(bond))
1095                 goto done;
1096         vlan_features &= NETIF_F_ALL_FOR_ALL;
1097
1098         bond_for_each_slave(bond, slave, iter) {
1099                 vlan_features = netdev_increment_features(vlan_features,
1100                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1101
1102                 enc_features = netdev_increment_features(enc_features,
1103                                                          slave->dev->hw_enc_features,
1104                                                          BOND_ENC_FEATURES);
1105                 dst_release_flag &= slave->dev->priv_flags;
1106                 if (slave->dev->hard_header_len > max_hard_header_len)
1107                         max_hard_header_len = slave->dev->hard_header_len;
1108
1109                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1110                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1111         }
1112         bond_dev->hard_header_len = max_hard_header_len;
1113
1114 done:
1115         bond_dev->vlan_features = vlan_features;
1116         bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1117                                     NETIF_F_GSO_UDP_L4;
1118         bond_dev->gso_max_segs = gso_max_segs;
1119         netif_set_gso_max_size(bond_dev, gso_max_size);
1120
1121         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1122         if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1123             dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1124                 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1125
1126         netdev_change_features(bond_dev);
1127 }
1128
1129 static void bond_setup_by_slave(struct net_device *bond_dev,
1130                                 struct net_device *slave_dev)
1131 {
1132         bond_dev->header_ops        = slave_dev->header_ops;
1133
1134         bond_dev->type              = slave_dev->type;
1135         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1136         bond_dev->addr_len          = slave_dev->addr_len;
1137
1138         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1139                 slave_dev->addr_len);
1140 }
1141
1142 /* On bonding slaves other than the currently active slave, suppress
1143  * duplicates except for alb non-mcast/bcast.
1144  */
1145 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1146                                             struct slave *slave,
1147                                             struct bonding *bond)
1148 {
1149         if (bond_is_slave_inactive(slave)) {
1150                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1151                     skb->pkt_type != PACKET_BROADCAST &&
1152                     skb->pkt_type != PACKET_MULTICAST)
1153                         return false;
1154                 return true;
1155         }
1156         return false;
1157 }
1158
1159 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1160 {
1161         struct sk_buff *skb = *pskb;
1162         struct slave *slave;
1163         struct bonding *bond;
1164         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1165                           struct slave *);
1166         int ret = RX_HANDLER_ANOTHER;
1167
1168         skb = skb_share_check(skb, GFP_ATOMIC);
1169         if (unlikely(!skb))
1170                 return RX_HANDLER_CONSUMED;
1171
1172         *pskb = skb;
1173
1174         slave = bond_slave_get_rcu(skb->dev);
1175         bond = slave->bond;
1176
1177         recv_probe = READ_ONCE(bond->recv_probe);
1178         if (recv_probe) {
1179                 ret = recv_probe(skb, bond, slave);
1180                 if (ret == RX_HANDLER_CONSUMED) {
1181                         consume_skb(skb);
1182                         return ret;
1183                 }
1184         }
1185
1186         /* Link-local multicast packets should be passed to the
1187          * stack on the link they arrive as well as pass them to the
1188          * bond-master device. These packets are mostly usable when
1189          * stack receives it with the link on which they arrive
1190          * (e.g. LLDP) they also must be available on master. Some of
1191          * the use cases include (but are not limited to): LLDP agents
1192          * that must be able to operate both on enslaved interfaces as
1193          * well as on bonds themselves; linux bridges that must be able
1194          * to process/pass BPDUs from attached bonds when any kind of
1195          * STP version is enabled on the network.
1196          */
1197         if (is_link_local_ether_addr(eth_hdr(skb)->h_dest)) {
1198                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1199
1200                 if (nskb) {
1201                         nskb->dev = bond->dev;
1202                         nskb->queue_mapping = 0;
1203                         netif_rx(nskb);
1204                 }
1205                 return RX_HANDLER_PASS;
1206         }
1207         if (bond_should_deliver_exact_match(skb, slave, bond))
1208                 return RX_HANDLER_EXACT;
1209
1210         skb->dev = bond->dev;
1211
1212         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1213             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1214             skb->pkt_type == PACKET_HOST) {
1215
1216                 if (unlikely(skb_cow_head(skb,
1217                                           skb->data - skb_mac_header(skb)))) {
1218                         kfree_skb(skb);
1219                         return RX_HANDLER_CONSUMED;
1220                 }
1221                 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1222                                   bond->dev->addr_len);
1223         }
1224
1225         return ret;
1226 }
1227
1228 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1229 {
1230         switch (BOND_MODE(bond)) {
1231         case BOND_MODE_ROUNDROBIN:
1232                 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1233         case BOND_MODE_ACTIVEBACKUP:
1234                 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1235         case BOND_MODE_BROADCAST:
1236                 return NETDEV_LAG_TX_TYPE_BROADCAST;
1237         case BOND_MODE_XOR:
1238         case BOND_MODE_8023AD:
1239                 return NETDEV_LAG_TX_TYPE_HASH;
1240         default:
1241                 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1242         }
1243 }
1244
1245 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1246                                                enum netdev_lag_tx_type type)
1247 {
1248         if (type != NETDEV_LAG_TX_TYPE_HASH)
1249                 return NETDEV_LAG_HASH_NONE;
1250
1251         switch (bond->params.xmit_policy) {
1252         case BOND_XMIT_POLICY_LAYER2:
1253                 return NETDEV_LAG_HASH_L2;
1254         case BOND_XMIT_POLICY_LAYER34:
1255                 return NETDEV_LAG_HASH_L34;
1256         case BOND_XMIT_POLICY_LAYER23:
1257                 return NETDEV_LAG_HASH_L23;
1258         case BOND_XMIT_POLICY_ENCAP23:
1259                 return NETDEV_LAG_HASH_E23;
1260         case BOND_XMIT_POLICY_ENCAP34:
1261                 return NETDEV_LAG_HASH_E34;
1262         default:
1263                 return NETDEV_LAG_HASH_UNKNOWN;
1264         }
1265 }
1266
1267 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1268                                       struct netlink_ext_ack *extack)
1269 {
1270         struct netdev_lag_upper_info lag_upper_info;
1271         enum netdev_lag_tx_type type;
1272
1273         type = bond_lag_tx_type(bond);
1274         lag_upper_info.tx_type = type;
1275         lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1276
1277         return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1278                                             &lag_upper_info, extack);
1279 }
1280
1281 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1282 {
1283         netdev_upper_dev_unlink(slave->dev, bond->dev);
1284         slave->dev->flags &= ~IFF_SLAVE;
1285 }
1286
1287 static struct slave *bond_alloc_slave(struct bonding *bond)
1288 {
1289         struct slave *slave = NULL;
1290
1291         slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1292         if (!slave)
1293                 return NULL;
1294
1295         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1296                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1297                                                GFP_KERNEL);
1298                 if (!SLAVE_AD_INFO(slave)) {
1299                         kfree(slave);
1300                         return NULL;
1301                 }
1302         }
1303         INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1304
1305         return slave;
1306 }
1307
1308 static void bond_free_slave(struct slave *slave)
1309 {
1310         struct bonding *bond = bond_get_bond_by_slave(slave);
1311
1312         cancel_delayed_work_sync(&slave->notify_work);
1313         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1314                 kfree(SLAVE_AD_INFO(slave));
1315
1316         kfree(slave);
1317 }
1318
1319 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1320 {
1321         info->bond_mode = BOND_MODE(bond);
1322         info->miimon = bond->params.miimon;
1323         info->num_slaves = bond->slave_cnt;
1324 }
1325
1326 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1327 {
1328         strcpy(info->slave_name, slave->dev->name);
1329         info->link = slave->link;
1330         info->state = bond_slave_state(slave);
1331         info->link_failure_count = slave->link_failure_count;
1332 }
1333
1334 static void bond_netdev_notify_work(struct work_struct *_work)
1335 {
1336         struct slave *slave = container_of(_work, struct slave,
1337                                            notify_work.work);
1338
1339         if (rtnl_trylock()) {
1340                 struct netdev_bonding_info binfo;
1341
1342                 bond_fill_ifslave(slave, &binfo.slave);
1343                 bond_fill_ifbond(slave->bond, &binfo.master);
1344                 netdev_bonding_info_change(slave->dev, &binfo);
1345                 rtnl_unlock();
1346         } else {
1347                 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1348         }
1349 }
1350
1351 void bond_queue_slave_event(struct slave *slave)
1352 {
1353         queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1354 }
1355
1356 void bond_lower_state_changed(struct slave *slave)
1357 {
1358         struct netdev_lag_lower_state_info info;
1359
1360         info.link_up = slave->link == BOND_LINK_UP ||
1361                        slave->link == BOND_LINK_FAIL;
1362         info.tx_enabled = bond_is_active_slave(slave);
1363         netdev_lower_state_changed(slave->dev, &info);
1364 }
1365
1366 /* enslave device <slave> to bond device <master> */
1367 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1368                  struct netlink_ext_ack *extack)
1369 {
1370         struct bonding *bond = netdev_priv(bond_dev);
1371         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1372         struct slave *new_slave = NULL, *prev_slave;
1373         struct sockaddr_storage ss;
1374         int link_reporting;
1375         int res = 0, i;
1376
1377         if (!bond->params.use_carrier &&
1378             slave_dev->ethtool_ops->get_link == NULL &&
1379             slave_ops->ndo_do_ioctl == NULL) {
1380                 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1381                             slave_dev->name);
1382         }
1383
1384         /* already in-use? */
1385         if (netdev_is_rx_handler_busy(slave_dev)) {
1386                 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1387                 netdev_err(bond_dev,
1388                            "Error: Device is in use and cannot be enslaved\n");
1389                 return -EBUSY;
1390         }
1391
1392         if (bond_dev == slave_dev) {
1393                 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1394                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1395                 return -EPERM;
1396         }
1397
1398         /* vlan challenged mutual exclusion */
1399         /* no need to lock since we're protected by rtnl_lock */
1400         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1401                 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1402                            slave_dev->name);
1403                 if (vlan_uses_dev(bond_dev)) {
1404                         NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1405                         netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1406                                    slave_dev->name, bond_dev->name);
1407                         return -EPERM;
1408                 } else {
1409                         netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1410                                     slave_dev->name, slave_dev->name,
1411                                     bond_dev->name);
1412                 }
1413         } else {
1414                 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1415                            slave_dev->name);
1416         }
1417
1418         /* Old ifenslave binaries are no longer supported.  These can
1419          * be identified with moderate accuracy by the state of the slave:
1420          * the current ifenslave will set the interface down prior to
1421          * enslaving it; the old ifenslave will not.
1422          */
1423         if (slave_dev->flags & IFF_UP) {
1424                 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
1425                 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1426                            slave_dev->name);
1427                 return -EPERM;
1428         }
1429
1430         /* set bonding device ether type by slave - bonding netdevices are
1431          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1432          * there is a need to override some of the type dependent attribs/funcs.
1433          *
1434          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1435          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1436          */
1437         if (!bond_has_slaves(bond)) {
1438                 if (bond_dev->type != slave_dev->type) {
1439                         netdev_dbg(bond_dev, "change device type from %d to %d\n",
1440                                    bond_dev->type, slave_dev->type);
1441
1442                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1443                                                        bond_dev);
1444                         res = notifier_to_errno(res);
1445                         if (res) {
1446                                 netdev_err(bond_dev, "refused to change device type\n");
1447                                 return -EBUSY;
1448                         }
1449
1450                         /* Flush unicast and multicast addresses */
1451                         dev_uc_flush(bond_dev);
1452                         dev_mc_flush(bond_dev);
1453
1454                         if (slave_dev->type != ARPHRD_ETHER)
1455                                 bond_setup_by_slave(bond_dev, slave_dev);
1456                         else {
1457                                 ether_setup(bond_dev);
1458                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1459                         }
1460
1461                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1462                                                  bond_dev);
1463                 }
1464         } else if (bond_dev->type != slave_dev->type) {
1465                 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1466                 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1467                            slave_dev->name, slave_dev->type, bond_dev->type);
1468                 return -EINVAL;
1469         }
1470
1471         if (slave_dev->type == ARPHRD_INFINIBAND &&
1472             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1473                 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1474                 netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n",
1475                             slave_dev->type);
1476                 res = -EOPNOTSUPP;
1477                 goto err_undo_flags;
1478         }
1479
1480         if (!slave_ops->ndo_set_mac_address ||
1481             slave_dev->type == ARPHRD_INFINIBAND) {
1482                 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1483                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1484                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1485                         if (!bond_has_slaves(bond)) {
1486                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1487                                 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1488                         } else {
1489                                 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1490                                 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");
1491                                 res = -EOPNOTSUPP;
1492                                 goto err_undo_flags;
1493                         }
1494                 }
1495         }
1496
1497         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1498
1499         /* If this is the first slave, then we need to set the master's hardware
1500          * address to be the same as the slave's.
1501          */
1502         if (!bond_has_slaves(bond) &&
1503             bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1504                 res = bond_set_dev_addr(bond->dev, slave_dev);
1505                 if (res)
1506                         goto err_undo_flags;
1507         }
1508
1509         new_slave = bond_alloc_slave(bond);
1510         if (!new_slave) {
1511                 res = -ENOMEM;
1512                 goto err_undo_flags;
1513         }
1514
1515         new_slave->bond = bond;
1516         new_slave->dev = slave_dev;
1517         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1518          * is set via sysfs or module option if desired.
1519          */
1520         new_slave->queue_id = 0;
1521
1522         /* Save slave's original mtu and then set it to match the bond */
1523         new_slave->original_mtu = slave_dev->mtu;
1524         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1525         if (res) {
1526                 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1527                 goto err_free;
1528         }
1529
1530         /* Save slave's original ("permanent") mac address for modes
1531          * that need it, and for restoring it upon release, and then
1532          * set it to the master's address
1533          */
1534         bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1535                           slave_dev->addr_len);
1536
1537         if (!bond->params.fail_over_mac ||
1538             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1539                 /* Set slave to master's mac address.  The application already
1540                  * set the master's mac address to that of the first slave
1541                  */
1542                 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1543                 ss.ss_family = slave_dev->type;
1544                 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
1545                                           extack);
1546                 if (res) {
1547                         netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1548                         goto err_restore_mtu;
1549                 }
1550         }
1551
1552         /* set slave flag before open to prevent IPv6 addrconf */
1553         slave_dev->flags |= IFF_SLAVE;
1554
1555         /* open the slave since the application closed it */
1556         res = dev_open(slave_dev, extack);
1557         if (res) {
1558                 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1559                 goto err_restore_mac;
1560         }
1561
1562         slave_dev->priv_flags |= IFF_BONDING;
1563         /* initialize slave stats */
1564         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1565
1566         if (bond_is_lb(bond)) {
1567                 /* bond_alb_init_slave() must be called before all other stages since
1568                  * it might fail and we do not want to have to undo everything
1569                  */
1570                 res = bond_alb_init_slave(bond, new_slave);
1571                 if (res)
1572                         goto err_close;
1573         }
1574
1575         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1576         if (res) {
1577                 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1578                            slave_dev->name);
1579                 goto err_close;
1580         }
1581
1582         prev_slave = bond_last_slave(bond);
1583
1584         new_slave->delay = 0;
1585         new_slave->link_failure_count = 0;
1586
1587         if (bond_update_speed_duplex(new_slave) &&
1588             bond_needs_speed_duplex(bond))
1589                 new_slave->link = BOND_LINK_DOWN;
1590
1591         new_slave->last_rx = jiffies -
1592                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1593         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1594                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1595
1596         if (bond->params.miimon && !bond->params.use_carrier) {
1597                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1598
1599                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1600                         /* miimon is set but a bonded network driver
1601                          * does not support ETHTOOL/MII and
1602                          * arp_interval is not set.  Note: if
1603                          * use_carrier is enabled, we will never go
1604                          * here (because netif_carrier is always
1605                          * supported); thus, we don't need to change
1606                          * the messages for netif_carrier.
1607                          */
1608                         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",
1609                                     slave_dev->name);
1610                 } else if (link_reporting == -1) {
1611                         /* unable get link status using mii/ethtool */
1612                         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",
1613                                     slave_dev->name);
1614                 }
1615         }
1616
1617         /* check for initial state */
1618         new_slave->link = BOND_LINK_NOCHANGE;
1619         if (bond->params.miimon) {
1620                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1621                         if (bond->params.updelay) {
1622                                 bond_set_slave_link_state(new_slave,
1623                                                           BOND_LINK_BACK,
1624                                                           BOND_SLAVE_NOTIFY_NOW);
1625                                 new_slave->delay = bond->params.updelay;
1626                         } else {
1627                                 bond_set_slave_link_state(new_slave,
1628                                                           BOND_LINK_UP,
1629                                                           BOND_SLAVE_NOTIFY_NOW);
1630                         }
1631                 } else {
1632                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1633                                                   BOND_SLAVE_NOTIFY_NOW);
1634                 }
1635         } else if (bond->params.arp_interval) {
1636                 bond_set_slave_link_state(new_slave,
1637                                           (netif_carrier_ok(slave_dev) ?
1638                                           BOND_LINK_UP : BOND_LINK_DOWN),
1639                                           BOND_SLAVE_NOTIFY_NOW);
1640         } else {
1641                 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1642                                           BOND_SLAVE_NOTIFY_NOW);
1643         }
1644
1645         if (new_slave->link != BOND_LINK_DOWN)
1646                 new_slave->last_link_up = jiffies;
1647         netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1648                    new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1649                    (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1650
1651         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1652                 /* if there is a primary slave, remember it */
1653                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1654                         rcu_assign_pointer(bond->primary_slave, new_slave);
1655                         bond->force_primary = true;
1656                 }
1657         }
1658
1659         switch (BOND_MODE(bond)) {
1660         case BOND_MODE_ACTIVEBACKUP:
1661                 bond_set_slave_inactive_flags(new_slave,
1662                                               BOND_SLAVE_NOTIFY_NOW);
1663                 break;
1664         case BOND_MODE_8023AD:
1665                 /* in 802.3ad mode, the internal mechanism
1666                  * will activate the slaves in the selected
1667                  * aggregator
1668                  */
1669                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1670                 /* if this is the first slave */
1671                 if (!prev_slave) {
1672                         SLAVE_AD_INFO(new_slave)->id = 1;
1673                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1674                          * can be called only after the mac address of the bond is set
1675                          */
1676                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1677                 } else {
1678                         SLAVE_AD_INFO(new_slave)->id =
1679                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1680                 }
1681
1682                 bond_3ad_bind_slave(new_slave);
1683                 break;
1684         case BOND_MODE_TLB:
1685         case BOND_MODE_ALB:
1686                 bond_set_active_slave(new_slave);
1687                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1688                 break;
1689         default:
1690                 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1691
1692                 /* always active in trunk mode */
1693                 bond_set_active_slave(new_slave);
1694
1695                 /* In trunking mode there is little meaning to curr_active_slave
1696                  * anyway (it holds no special properties of the bond device),
1697                  * so we can change it without calling change_active_interface()
1698                  */
1699                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1700                     new_slave->link == BOND_LINK_UP)
1701                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1702
1703                 break;
1704         } /* switch(bond_mode) */
1705
1706 #ifdef CONFIG_NET_POLL_CONTROLLER
1707         if (bond->dev->npinfo) {
1708                 if (slave_enable_netpoll(new_slave)) {
1709                         netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1710                         res = -EBUSY;
1711                         goto err_detach;
1712                 }
1713         }
1714 #endif
1715
1716         if (!(bond_dev->features & NETIF_F_LRO))
1717                 dev_disable_lro(slave_dev);
1718
1719         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1720                                          new_slave);
1721         if (res) {
1722                 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1723                 goto err_detach;
1724         }
1725
1726         res = bond_master_upper_dev_link(bond, new_slave, extack);
1727         if (res) {
1728                 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1729                 goto err_unregister;
1730         }
1731
1732         res = bond_sysfs_slave_add(new_slave);
1733         if (res) {
1734                 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1735                 goto err_upper_unlink;
1736         }
1737
1738         bond->nest_level = dev_get_nest_level(bond_dev) + 1;
1739
1740         /* If the mode uses primary, then the following is handled by
1741          * bond_change_active_slave().
1742          */
1743         if (!bond_uses_primary(bond)) {
1744                 /* set promiscuity level to new slave */
1745                 if (bond_dev->flags & IFF_PROMISC) {
1746                         res = dev_set_promiscuity(slave_dev, 1);
1747                         if (res)
1748                                 goto err_sysfs_del;
1749                 }
1750
1751                 /* set allmulti level to new slave */
1752                 if (bond_dev->flags & IFF_ALLMULTI) {
1753                         res = dev_set_allmulti(slave_dev, 1);
1754                         if (res) {
1755                                 if (bond_dev->flags & IFF_PROMISC)
1756                                         dev_set_promiscuity(slave_dev, -1);
1757                                 goto err_sysfs_del;
1758                         }
1759                 }
1760
1761                 netif_addr_lock_bh(bond_dev);
1762                 dev_mc_sync_multiple(slave_dev, bond_dev);
1763                 dev_uc_sync_multiple(slave_dev, bond_dev);
1764                 netif_addr_unlock_bh(bond_dev);
1765
1766                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1767                         /* add lacpdu mc addr to mc list */
1768                         u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1769
1770                         dev_mc_add(slave_dev, lacpdu_multicast);
1771                 }
1772         }
1773
1774         bond->slave_cnt++;
1775         bond_compute_features(bond);
1776         bond_set_carrier(bond);
1777
1778         if (bond_uses_primary(bond)) {
1779                 block_netpoll_tx();
1780                 bond_select_active_slave(bond);
1781                 unblock_netpoll_tx();
1782         }
1783
1784         if (bond_mode_can_use_xmit_hash(bond))
1785                 bond_update_slave_arr(bond, NULL);
1786
1787
1788         netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1789                     slave_dev->name,
1790                     bond_is_active_slave(new_slave) ? "an active" : "a backup",
1791                     new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1792
1793         /* enslave is successful */
1794         bond_queue_slave_event(new_slave);
1795         return 0;
1796
1797 /* Undo stages on error */
1798 err_sysfs_del:
1799         bond_sysfs_slave_del(new_slave);
1800
1801 err_upper_unlink:
1802         bond_upper_dev_unlink(bond, new_slave);
1803
1804 err_unregister:
1805         netdev_rx_handler_unregister(slave_dev);
1806
1807 err_detach:
1808         vlan_vids_del_by_dev(slave_dev, bond_dev);
1809         if (rcu_access_pointer(bond->primary_slave) == new_slave)
1810                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1811         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1812                 block_netpoll_tx();
1813                 bond_change_active_slave(bond, NULL);
1814                 bond_select_active_slave(bond);
1815                 unblock_netpoll_tx();
1816         }
1817         /* either primary_slave or curr_active_slave might've changed */
1818         synchronize_rcu();
1819         slave_disable_netpoll(new_slave);
1820
1821 err_close:
1822         slave_dev->priv_flags &= ~IFF_BONDING;
1823         dev_close(slave_dev);
1824
1825 err_restore_mac:
1826         slave_dev->flags &= ~IFF_SLAVE;
1827         if (!bond->params.fail_over_mac ||
1828             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1829                 /* XXX TODO - fom follow mode needs to change master's
1830                  * MAC if this slave's MAC is in use by the bond, or at
1831                  * least print a warning.
1832                  */
1833                 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
1834                                   new_slave->dev->addr_len);
1835                 ss.ss_family = slave_dev->type;
1836                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
1837         }
1838
1839 err_restore_mtu:
1840         dev_set_mtu(slave_dev, new_slave->original_mtu);
1841
1842 err_free:
1843         bond_free_slave(new_slave);
1844
1845 err_undo_flags:
1846         /* Enslave of first slave has failed and we need to fix master's mac */
1847         if (!bond_has_slaves(bond)) {
1848                 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1849                                             slave_dev->dev_addr))
1850                         eth_hw_addr_random(bond_dev);
1851                 if (bond_dev->type != ARPHRD_ETHER) {
1852                         dev_close(bond_dev);
1853                         ether_setup(bond_dev);
1854                         bond_dev->flags |= IFF_MASTER;
1855                         bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1856                 }
1857         }
1858
1859         return res;
1860 }
1861
1862 /* Try to release the slave device <slave> from the bond device <master>
1863  * It is legal to access curr_active_slave without a lock because all the function
1864  * is RTNL-locked. If "all" is true it means that the function is being called
1865  * while destroying a bond interface and all slaves are being released.
1866  *
1867  * The rules for slave state should be:
1868  *   for Active/Backup:
1869  *     Active stays on all backups go down
1870  *   for Bonded connections:
1871  *     The first up interface should be left on and all others downed.
1872  */
1873 static int __bond_release_one(struct net_device *bond_dev,
1874                               struct net_device *slave_dev,
1875                               bool all, bool unregister)
1876 {
1877         struct bonding *bond = netdev_priv(bond_dev);
1878         struct slave *slave, *oldcurrent;
1879         struct sockaddr_storage ss;
1880         int old_flags = bond_dev->flags;
1881         netdev_features_t old_features = bond_dev->features;
1882
1883         /* slave is not a slave or master is not master of this slave */
1884         if (!(slave_dev->flags & IFF_SLAVE) ||
1885             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1886                 netdev_dbg(bond_dev, "cannot release %s\n",
1887                            slave_dev->name);
1888                 return -EINVAL;
1889         }
1890
1891         block_netpoll_tx();
1892
1893         slave = bond_get_slave_by_dev(bond, slave_dev);
1894         if (!slave) {
1895                 /* not a slave of this bond */
1896                 netdev_info(bond_dev, "%s not enslaved\n",
1897                             slave_dev->name);
1898                 unblock_netpoll_tx();
1899                 return -EINVAL;
1900         }
1901
1902         bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1903
1904         bond_sysfs_slave_del(slave);
1905
1906         /* recompute stats just before removing the slave */
1907         bond_get_stats(bond->dev, &bond->bond_stats);
1908
1909         bond_upper_dev_unlink(bond, slave);
1910         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1911          * for this slave anymore.
1912          */
1913         netdev_rx_handler_unregister(slave_dev);
1914
1915         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1916                 bond_3ad_unbind_slave(slave);
1917
1918         if (bond_mode_can_use_xmit_hash(bond))
1919                 bond_update_slave_arr(bond, slave);
1920
1921         netdev_info(bond_dev, "Releasing %s interface %s\n",
1922                     bond_is_active_slave(slave) ? "active" : "backup",
1923                     slave_dev->name);
1924
1925         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1926
1927         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1928
1929         if (!all && (!bond->params.fail_over_mac ||
1930                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1931                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1932                     bond_has_slaves(bond))
1933                         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",
1934                                     slave_dev->name, slave->perm_hwaddr,
1935                                     bond_dev->name, slave_dev->name);
1936         }
1937
1938         if (rtnl_dereference(bond->primary_slave) == slave)
1939                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1940
1941         if (oldcurrent == slave)
1942                 bond_change_active_slave(bond, NULL);
1943
1944         if (bond_is_lb(bond)) {
1945                 /* Must be called only after the slave has been
1946                  * detached from the list and the curr_active_slave
1947                  * has been cleared (if our_slave == old_current),
1948                  * but before a new active slave is selected.
1949                  */
1950                 bond_alb_deinit_slave(bond, slave);
1951         }
1952
1953         if (all) {
1954                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1955         } else if (oldcurrent == slave) {
1956                 /* Note that we hold RTNL over this sequence, so there
1957                  * is no concern that another slave add/remove event
1958                  * will interfere.
1959                  */
1960                 bond_select_active_slave(bond);
1961         }
1962
1963         if (!bond_has_slaves(bond)) {
1964                 bond_set_carrier(bond);
1965                 eth_hw_addr_random(bond_dev);
1966         }
1967
1968         unblock_netpoll_tx();
1969         synchronize_rcu();
1970         bond->slave_cnt--;
1971
1972         if (!bond_has_slaves(bond)) {
1973                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1974                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1975         }
1976
1977         bond_compute_features(bond);
1978         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1979             (old_features & NETIF_F_VLAN_CHALLENGED))
1980                 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1981                             slave_dev->name, bond_dev->name);
1982
1983         vlan_vids_del_by_dev(slave_dev, bond_dev);
1984
1985         /* If the mode uses primary, then this case was handled above by
1986          * bond_change_active_slave(..., NULL)
1987          */
1988         if (!bond_uses_primary(bond)) {
1989                 /* unset promiscuity level from slave
1990                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1991                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1992                  * value of that flag before that change, as that was the value
1993                  * when this slave was attached, so we cache at the start of the
1994                  * function and use it here. Same goes for ALLMULTI below
1995                  */
1996                 if (old_flags & IFF_PROMISC)
1997                         dev_set_promiscuity(slave_dev, -1);
1998
1999                 /* unset allmulti level from slave */
2000                 if (old_flags & IFF_ALLMULTI)
2001                         dev_set_allmulti(slave_dev, -1);
2002
2003                 bond_hw_addr_flush(bond_dev, slave_dev);
2004         }
2005
2006         slave_disable_netpoll(slave);
2007
2008         /* close slave before restoring its mac address */
2009         dev_close(slave_dev);
2010
2011         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2012             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2013                 /* restore original ("permanent") mac address */
2014                 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2015                                   slave->dev->addr_len);
2016                 ss.ss_family = slave_dev->type;
2017                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2018         }
2019
2020         if (unregister)
2021                 __dev_set_mtu(slave_dev, slave->original_mtu);
2022         else
2023                 dev_set_mtu(slave_dev, slave->original_mtu);
2024
2025         slave_dev->priv_flags &= ~IFF_BONDING;
2026
2027         bond_free_slave(slave);
2028
2029         return 0;
2030 }
2031
2032 /* A wrapper used because of ndo_del_link */
2033 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2034 {
2035         return __bond_release_one(bond_dev, slave_dev, false, false);
2036 }
2037
2038 /* First release a slave and then destroy the bond if no more slaves are left.
2039  * Must be under rtnl_lock when this function is called.
2040  */
2041 static int  bond_release_and_destroy(struct net_device *bond_dev,
2042                                      struct net_device *slave_dev)
2043 {
2044         struct bonding *bond = netdev_priv(bond_dev);
2045         int ret;
2046
2047         ret = __bond_release_one(bond_dev, slave_dev, false, true);
2048         if (ret == 0 && !bond_has_slaves(bond)) {
2049                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2050                 netdev_info(bond_dev, "Destroying bond %s\n",
2051                             bond_dev->name);
2052                 bond_remove_proc_entry(bond);
2053                 unregister_netdevice(bond_dev);
2054         }
2055         return ret;
2056 }
2057
2058 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2059 {
2060         struct bonding *bond = netdev_priv(bond_dev);
2061         bond_fill_ifbond(bond, info);
2062 }
2063
2064 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2065 {
2066         struct bonding *bond = netdev_priv(bond_dev);
2067         struct list_head *iter;
2068         int i = 0, res = -ENODEV;
2069         struct slave *slave;
2070
2071         bond_for_each_slave(bond, slave, iter) {
2072                 if (i++ == (int)info->slave_id) {
2073                         res = 0;
2074                         bond_fill_ifslave(slave, info);
2075                         break;
2076                 }
2077         }
2078
2079         return res;
2080 }
2081
2082 /*-------------------------------- Monitoring -------------------------------*/
2083
2084 /* called with rcu_read_lock() */
2085 static int bond_miimon_inspect(struct bonding *bond)
2086 {
2087         int link_state, commit = 0;
2088         struct list_head *iter;
2089         struct slave *slave;
2090         bool ignore_updelay;
2091
2092         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2093
2094         bond_for_each_slave_rcu(bond, slave, iter) {
2095                 slave->new_link = BOND_LINK_NOCHANGE;
2096                 slave->link_new_state = slave->link;
2097
2098                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2099
2100                 switch (slave->link) {
2101                 case BOND_LINK_UP:
2102                         if (link_state)
2103                                 continue;
2104
2105                         bond_propose_link_state(slave, BOND_LINK_FAIL);
2106                         commit++;
2107                         slave->delay = bond->params.downdelay;
2108                         if (slave->delay) {
2109                                 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2110                                             (BOND_MODE(bond) ==
2111                                              BOND_MODE_ACTIVEBACKUP) ?
2112                                              (bond_is_active_slave(slave) ?
2113                                               "active " : "backup ") : "",
2114                                             slave->dev->name,
2115                                             bond->params.downdelay * bond->params.miimon);
2116                         }
2117                         /*FALLTHRU*/
2118                 case BOND_LINK_FAIL:
2119                         if (link_state) {
2120                                 /* recovered before downdelay expired */
2121                                 bond_propose_link_state(slave, BOND_LINK_UP);
2122                                 slave->last_link_up = jiffies;
2123                                 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2124                                             (bond->params.downdelay - slave->delay) *
2125                                             bond->params.miimon,
2126                                             slave->dev->name);
2127                                 commit++;
2128                                 continue;
2129                         }
2130
2131                         if (slave->delay <= 0) {
2132                                 slave->new_link = BOND_LINK_DOWN;
2133                                 commit++;
2134                                 continue;
2135                         }
2136
2137                         slave->delay--;
2138                         break;
2139
2140                 case BOND_LINK_DOWN:
2141                         if (!link_state)
2142                                 continue;
2143
2144                         bond_propose_link_state(slave, BOND_LINK_BACK);
2145                         commit++;
2146                         slave->delay = bond->params.updelay;
2147
2148                         if (slave->delay) {
2149                                 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2150                                             slave->dev->name,
2151                                             ignore_updelay ? 0 :
2152                                             bond->params.updelay *
2153                                             bond->params.miimon);
2154                         }
2155                         /*FALLTHRU*/
2156                 case BOND_LINK_BACK:
2157                         if (!link_state) {
2158                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2159                                 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2160                                             (bond->params.updelay - slave->delay) *
2161                                             bond->params.miimon,
2162                                             slave->dev->name);
2163                                 commit++;
2164                                 continue;
2165                         }
2166
2167                         if (ignore_updelay)
2168                                 slave->delay = 0;
2169
2170                         if (slave->delay <= 0) {
2171                                 slave->new_link = BOND_LINK_UP;
2172                                 commit++;
2173                                 ignore_updelay = false;
2174                                 continue;
2175                         }
2176
2177                         slave->delay--;
2178                         break;
2179                 }
2180         }
2181
2182         return commit;
2183 }
2184
2185 static void bond_miimon_link_change(struct bonding *bond,
2186                                     struct slave *slave,
2187                                     char link)
2188 {
2189         switch (BOND_MODE(bond)) {
2190         case BOND_MODE_8023AD:
2191                 bond_3ad_handle_link_change(slave, link);
2192                 break;
2193         case BOND_MODE_TLB:
2194         case BOND_MODE_ALB:
2195                 bond_alb_handle_link_change(bond, slave, link);
2196                 break;
2197         case BOND_MODE_XOR:
2198                 bond_update_slave_arr(bond, NULL);
2199                 break;
2200         }
2201 }
2202
2203 static void bond_miimon_commit(struct bonding *bond)
2204 {
2205         struct list_head *iter;
2206         struct slave *slave, *primary;
2207
2208         bond_for_each_slave(bond, slave, iter) {
2209                 switch (slave->new_link) {
2210                 case BOND_LINK_NOCHANGE:
2211                         continue;
2212
2213                 case BOND_LINK_UP:
2214                         if (bond_update_speed_duplex(slave) &&
2215                             bond_needs_speed_duplex(bond)) {
2216                                 slave->link = BOND_LINK_DOWN;
2217                                 if (net_ratelimit())
2218                                         netdev_warn(bond->dev,
2219                                                     "failed to get link speed/duplex for %s\n",
2220                                                     slave->dev->name);
2221                                 continue;
2222                         }
2223                         bond_set_slave_link_state(slave, BOND_LINK_UP,
2224                                                   BOND_SLAVE_NOTIFY_NOW);
2225                         slave->last_link_up = jiffies;
2226
2227                         primary = rtnl_dereference(bond->primary_slave);
2228                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2229                                 /* prevent it from being the active one */
2230                                 bond_set_backup_slave(slave);
2231                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2232                                 /* make it immediately active */
2233                                 bond_set_active_slave(slave);
2234                         } else if (slave != primary) {
2235                                 /* prevent it from being the active one */
2236                                 bond_set_backup_slave(slave);
2237                         }
2238
2239                         netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2240                                     slave->dev->name,
2241                                     slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2242                                     slave->duplex ? "full" : "half");
2243
2244                         bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2245
2246                         if (!bond->curr_active_slave || slave == primary)
2247                                 goto do_failover;
2248
2249                         continue;
2250
2251                 case BOND_LINK_DOWN:
2252                         if (slave->link_failure_count < UINT_MAX)
2253                                 slave->link_failure_count++;
2254
2255                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2256                                                   BOND_SLAVE_NOTIFY_NOW);
2257
2258                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2259                             BOND_MODE(bond) == BOND_MODE_8023AD)
2260                                 bond_set_slave_inactive_flags(slave,
2261                                                               BOND_SLAVE_NOTIFY_NOW);
2262
2263                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2264                                     slave->dev->name);
2265
2266                         bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2267
2268                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2269                                 goto do_failover;
2270
2271                         continue;
2272
2273                 default:
2274                         netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2275                                    slave->new_link, slave->dev->name);
2276                         slave->new_link = BOND_LINK_NOCHANGE;
2277
2278                         continue;
2279                 }
2280
2281 do_failover:
2282                 block_netpoll_tx();
2283                 bond_select_active_slave(bond);
2284                 unblock_netpoll_tx();
2285         }
2286
2287         bond_set_carrier(bond);
2288 }
2289
2290 /* bond_mii_monitor
2291  *
2292  * Really a wrapper that splits the mii monitor into two phases: an
2293  * inspection, then (if inspection indicates something needs to be done)
2294  * an acquisition of appropriate locks followed by a commit phase to
2295  * implement whatever link state changes are indicated.
2296  */
2297 static void bond_mii_monitor(struct work_struct *work)
2298 {
2299         struct bonding *bond = container_of(work, struct bonding,
2300                                             mii_work.work);
2301         bool should_notify_peers = false;
2302         unsigned long delay;
2303         struct slave *slave;
2304         struct list_head *iter;
2305
2306         delay = msecs_to_jiffies(bond->params.miimon);
2307
2308         if (!bond_has_slaves(bond))
2309                 goto re_arm;
2310
2311         rcu_read_lock();
2312
2313         should_notify_peers = bond_should_notify_peers(bond);
2314
2315         if (bond_miimon_inspect(bond)) {
2316                 rcu_read_unlock();
2317
2318                 /* Race avoidance with bond_close cancel of workqueue */
2319                 if (!rtnl_trylock()) {
2320                         delay = 1;
2321                         should_notify_peers = false;
2322                         goto re_arm;
2323                 }
2324
2325                 bond_for_each_slave(bond, slave, iter) {
2326                         bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2327                 }
2328                 bond_miimon_commit(bond);
2329
2330                 rtnl_unlock();  /* might sleep, hold no other locks */
2331         } else
2332                 rcu_read_unlock();
2333
2334 re_arm:
2335         if (bond->params.miimon)
2336                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2337
2338         if (should_notify_peers) {
2339                 if (!rtnl_trylock())
2340                         return;
2341                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2342                 rtnl_unlock();
2343         }
2344 }
2345
2346 static int bond_upper_dev_walk(struct net_device *upper, void *data)
2347 {
2348         __be32 ip = *((__be32 *)data);
2349
2350         return ip == bond_confirm_addr(upper, 0, ip);
2351 }
2352
2353 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2354 {
2355         bool ret = false;
2356
2357         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2358                 return true;
2359
2360         rcu_read_lock();
2361         if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip))
2362                 ret = true;
2363         rcu_read_unlock();
2364
2365         return ret;
2366 }
2367
2368 /* We go to the (large) trouble of VLAN tagging ARP frames because
2369  * switches in VLAN mode (especially if ports are configured as
2370  * "native" to a VLAN) might not pass non-tagged frames.
2371  */
2372 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2373                           __be32 dest_ip, __be32 src_ip,
2374                           struct bond_vlan_tag *tags)
2375 {
2376         struct sk_buff *skb;
2377         struct bond_vlan_tag *outer_tag = tags;
2378
2379         netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2380                    arp_op, slave_dev->name, &dest_ip, &src_ip);
2381
2382         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2383                          NULL, slave_dev->dev_addr, NULL);
2384
2385         if (!skb) {
2386                 net_err_ratelimited("ARP packet allocation failed\n");
2387                 return;
2388         }
2389
2390         if (!tags || tags->vlan_proto == VLAN_N_VID)
2391                 goto xmit;
2392
2393         tags++;
2394
2395         /* Go through all the tags backwards and add them to the packet */
2396         while (tags->vlan_proto != VLAN_N_VID) {
2397                 if (!tags->vlan_id) {
2398                         tags++;
2399                         continue;
2400                 }
2401
2402                 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2403                            ntohs(outer_tag->vlan_proto), tags->vlan_id);
2404                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2405                                                 tags->vlan_id);
2406                 if (!skb) {
2407                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2408                         return;
2409                 }
2410
2411                 tags++;
2412         }
2413         /* Set the outer tag */
2414         if (outer_tag->vlan_id) {
2415                 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2416                            ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2417                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2418                                        outer_tag->vlan_id);
2419         }
2420
2421 xmit:
2422         arp_xmit(skb);
2423 }
2424
2425 /* Validate the device path between the @start_dev and the @end_dev.
2426  * The path is valid if the @end_dev is reachable through device
2427  * stacking.
2428  * When the path is validated, collect any vlan information in the
2429  * path.
2430  */
2431 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2432                                               struct net_device *end_dev,
2433                                               int level)
2434 {
2435         struct bond_vlan_tag *tags;
2436         struct net_device *upper;
2437         struct list_head  *iter;
2438
2439         if (start_dev == end_dev) {
2440                 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2441                 if (!tags)
2442                         return ERR_PTR(-ENOMEM);
2443                 tags[level].vlan_proto = VLAN_N_VID;
2444                 return tags;
2445         }
2446
2447         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2448                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2449                 if (IS_ERR_OR_NULL(tags)) {
2450                         if (IS_ERR(tags))
2451                                 return tags;
2452                         continue;
2453                 }
2454                 if (is_vlan_dev(upper)) {
2455                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2456                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2457                 }
2458
2459                 return tags;
2460         }
2461
2462         return NULL;
2463 }
2464
2465 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2466 {
2467         struct rtable *rt;
2468         struct bond_vlan_tag *tags;
2469         __be32 *targets = bond->params.arp_targets, addr;
2470         int i;
2471
2472         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2473                 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2474                 tags = NULL;
2475
2476                 /* Find out through which dev should the packet go */
2477                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2478                                      RTO_ONLINK, 0);
2479                 if (IS_ERR(rt)) {
2480                         /* there's no route to target - try to send arp
2481                          * probe to generate any traffic (arp_validate=0)
2482                          */
2483                         if (bond->params.arp_validate)
2484                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2485                                                      bond->dev->name,
2486                                                      &targets[i]);
2487                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2488                                       0, tags);
2489                         continue;
2490                 }
2491
2492                 /* bond device itself */
2493                 if (rt->dst.dev == bond->dev)
2494                         goto found;
2495
2496                 rcu_read_lock();
2497                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2498                 rcu_read_unlock();
2499
2500                 if (!IS_ERR_OR_NULL(tags))
2501                         goto found;
2502
2503                 /* Not our device - skip */
2504                 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2505                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2506
2507                 ip_rt_put(rt);
2508                 continue;
2509
2510 found:
2511                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2512                 ip_rt_put(rt);
2513                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2514                               addr, tags);
2515                 kfree(tags);
2516         }
2517 }
2518
2519 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2520 {
2521         int i;
2522
2523         if (!sip || !bond_has_this_ip(bond, tip)) {
2524                 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2525                            &sip, &tip);
2526                 return;
2527         }
2528
2529         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2530         if (i == -1) {
2531                 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2532                            &sip);
2533                 return;
2534         }
2535         slave->last_rx = jiffies;
2536         slave->target_last_arp_rx[i] = jiffies;
2537 }
2538
2539 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2540                  struct slave *slave)
2541 {
2542         struct arphdr *arp = (struct arphdr *)skb->data;
2543         struct slave *curr_active_slave, *curr_arp_slave;
2544         unsigned char *arp_ptr;
2545         __be32 sip, tip;
2546         int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2547         unsigned int alen;
2548
2549         if (!slave_do_arp_validate(bond, slave)) {
2550                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2551                     !slave_do_arp_validate_only(bond))
2552                         slave->last_rx = jiffies;
2553                 return RX_HANDLER_ANOTHER;
2554         } else if (!is_arp) {
2555                 return RX_HANDLER_ANOTHER;
2556         }
2557
2558         alen = arp_hdr_len(bond->dev);
2559
2560         netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2561                    skb->dev->name);
2562
2563         if (alen > skb_headlen(skb)) {
2564                 arp = kmalloc(alen, GFP_ATOMIC);
2565                 if (!arp)
2566                         goto out_unlock;
2567                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2568                         goto out_unlock;
2569         }
2570
2571         if (arp->ar_hln != bond->dev->addr_len ||
2572             skb->pkt_type == PACKET_OTHERHOST ||
2573             skb->pkt_type == PACKET_LOOPBACK ||
2574             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2575             arp->ar_pro != htons(ETH_P_IP) ||
2576             arp->ar_pln != 4)
2577                 goto out_unlock;
2578
2579         arp_ptr = (unsigned char *)(arp + 1);
2580         arp_ptr += bond->dev->addr_len;
2581         memcpy(&sip, arp_ptr, 4);
2582         arp_ptr += 4 + bond->dev->addr_len;
2583         memcpy(&tip, arp_ptr, 4);
2584
2585         netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2586                    slave->dev->name, bond_slave_state(slave),
2587                      bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2588                      &sip, &tip);
2589
2590         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2591         curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2592
2593         /* We 'trust' the received ARP enough to validate it if:
2594          *
2595          * (a) the slave receiving the ARP is active (which includes the
2596          * current ARP slave, if any), or
2597          *
2598          * (b) the receiving slave isn't active, but there is a currently
2599          * active slave and it received valid arp reply(s) after it became
2600          * the currently active slave, or
2601          *
2602          * (c) there is an ARP slave that sent an ARP during the prior ARP
2603          * interval, and we receive an ARP reply on any slave.  We accept
2604          * these because switch FDB update delays may deliver the ARP
2605          * reply to a slave other than the sender of the ARP request.
2606          *
2607          * Note: for (b), backup slaves are receiving the broadcast ARP
2608          * request, not a reply.  This request passes from the sending
2609          * slave through the L2 switch(es) to the receiving slave.  Since
2610          * this is checking the request, sip/tip are swapped for
2611          * validation.
2612          *
2613          * This is done to avoid endless looping when we can't reach the
2614          * arp_ip_target and fool ourselves with our own arp requests.
2615          */
2616         if (bond_is_active_slave(slave))
2617                 bond_validate_arp(bond, slave, sip, tip);
2618         else if (curr_active_slave &&
2619                  time_after(slave_last_rx(bond, curr_active_slave),
2620                             curr_active_slave->last_link_up))
2621                 bond_validate_arp(bond, slave, tip, sip);
2622         else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2623                  bond_time_in_interval(bond,
2624                                        dev_trans_start(curr_arp_slave->dev), 1))
2625                 bond_validate_arp(bond, slave, sip, tip);
2626
2627 out_unlock:
2628         if (arp != (struct arphdr *)skb->data)
2629                 kfree(arp);
2630         return RX_HANDLER_ANOTHER;
2631 }
2632
2633 /* function to verify if we're in the arp_interval timeslice, returns true if
2634  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2635  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2636  */
2637 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2638                                   int mod)
2639 {
2640         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2641
2642         return time_in_range(jiffies,
2643                              last_act - delta_in_ticks,
2644                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2645 }
2646
2647 /* This function is called regularly to monitor each slave's link
2648  * ensuring that traffic is being sent and received when arp monitoring
2649  * is used in load-balancing mode. if the adapter has been dormant, then an
2650  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2651  * arp monitoring in active backup mode.
2652  */
2653 static void bond_loadbalance_arp_mon(struct bonding *bond)
2654 {
2655         struct slave *slave, *oldcurrent;
2656         struct list_head *iter;
2657         int do_failover = 0, slave_state_changed = 0;
2658
2659         if (!bond_has_slaves(bond))
2660                 goto re_arm;
2661
2662         rcu_read_lock();
2663
2664         oldcurrent = rcu_dereference(bond->curr_active_slave);
2665         /* see if any of the previous devices are up now (i.e. they have
2666          * xmt and rcv traffic). the curr_active_slave does not come into
2667          * the picture unless it is null. also, slave->last_link_up is not
2668          * needed here because we send an arp on each slave and give a slave
2669          * as long as it needs to get the tx/rx within the delta.
2670          * TODO: what about up/down delay in arp mode? it wasn't here before
2671          *       so it can wait
2672          */
2673         bond_for_each_slave_rcu(bond, slave, iter) {
2674                 unsigned long trans_start = dev_trans_start(slave->dev);
2675
2676                 slave->new_link = BOND_LINK_NOCHANGE;
2677
2678                 if (slave->link != BOND_LINK_UP) {
2679                         if (bond_time_in_interval(bond, trans_start, 1) &&
2680                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2681
2682                                 slave->new_link = BOND_LINK_UP;
2683                                 slave_state_changed = 1;
2684
2685                                 /* primary_slave has no meaning in round-robin
2686                                  * mode. the window of a slave being up and
2687                                  * curr_active_slave being null after enslaving
2688                                  * is closed.
2689                                  */
2690                                 if (!oldcurrent) {
2691                                         netdev_info(bond->dev, "link status definitely up for interface %s\n",
2692                                                     slave->dev->name);
2693                                         do_failover = 1;
2694                                 } else {
2695                                         netdev_info(bond->dev, "interface %s is now up\n",
2696                                                     slave->dev->name);
2697                                 }
2698                         }
2699                 } else {
2700                         /* slave->link == BOND_LINK_UP */
2701
2702                         /* not all switches will respond to an arp request
2703                          * when the source ip is 0, so don't take the link down
2704                          * if we don't know our ip yet
2705                          */
2706                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2707                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2708
2709                                 slave->new_link = BOND_LINK_DOWN;
2710                                 slave_state_changed = 1;
2711
2712                                 if (slave->link_failure_count < UINT_MAX)
2713                                         slave->link_failure_count++;
2714
2715                                 netdev_info(bond->dev, "interface %s is now down\n",
2716                                             slave->dev->name);
2717
2718                                 if (slave == oldcurrent)
2719                                         do_failover = 1;
2720                         }
2721                 }
2722
2723                 /* note: if switch is in round-robin mode, all links
2724                  * must tx arp to ensure all links rx an arp - otherwise
2725                  * links may oscillate or not come up at all; if switch is
2726                  * in something like xor mode, there is nothing we can
2727                  * do - all replies will be rx'ed on same link causing slaves
2728                  * to be unstable during low/no traffic periods
2729                  */
2730                 if (bond_slave_is_up(slave))
2731                         bond_arp_send_all(bond, slave);
2732         }
2733
2734         rcu_read_unlock();
2735
2736         if (do_failover || slave_state_changed) {
2737                 if (!rtnl_trylock())
2738                         goto re_arm;
2739
2740                 bond_for_each_slave(bond, slave, iter) {
2741                         if (slave->new_link != BOND_LINK_NOCHANGE)
2742                                 slave->link = slave->new_link;
2743                 }
2744
2745                 if (slave_state_changed) {
2746                         bond_slave_state_change(bond);
2747                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2748                                 bond_update_slave_arr(bond, NULL);
2749                 }
2750                 if (do_failover) {
2751                         block_netpoll_tx();
2752                         bond_select_active_slave(bond);
2753                         unblock_netpoll_tx();
2754                 }
2755                 rtnl_unlock();
2756         }
2757
2758 re_arm:
2759         if (bond->params.arp_interval)
2760                 queue_delayed_work(bond->wq, &bond->arp_work,
2761                                    msecs_to_jiffies(bond->params.arp_interval));
2762 }
2763
2764 /* Called to inspect slaves for active-backup mode ARP monitor link state
2765  * changes.  Sets new_link in slaves to specify what action should take
2766  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2767  * to link states must be committed.
2768  *
2769  * Called with rcu_read_lock held.
2770  */
2771 static int bond_ab_arp_inspect(struct bonding *bond)
2772 {
2773         unsigned long trans_start, last_rx;
2774         struct list_head *iter;
2775         struct slave *slave;
2776         int commit = 0;
2777
2778         bond_for_each_slave_rcu(bond, slave, iter) {
2779                 slave->new_link = BOND_LINK_NOCHANGE;
2780                 last_rx = slave_last_rx(bond, slave);
2781
2782                 if (slave->link != BOND_LINK_UP) {
2783                         if (bond_time_in_interval(bond, last_rx, 1)) {
2784                                 slave->new_link = BOND_LINK_UP;
2785                                 commit++;
2786                         }
2787                         continue;
2788                 }
2789
2790                 /* Give slaves 2*delta after being enslaved or made
2791                  * active.  This avoids bouncing, as the last receive
2792                  * times need a full ARP monitor cycle to be updated.
2793                  */
2794                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2795                         continue;
2796
2797                 /* Backup slave is down if:
2798                  * - No current_arp_slave AND
2799                  * - more than 3*delta since last receive AND
2800                  * - the bond has an IP address
2801                  *
2802                  * Note: a non-null current_arp_slave indicates
2803                  * the curr_active_slave went down and we are
2804                  * searching for a new one; under this condition
2805                  * we only take the curr_active_slave down - this
2806                  * gives each slave a chance to tx/rx traffic
2807                  * before being taken out
2808                  */
2809                 if (!bond_is_active_slave(slave) &&
2810                     !rcu_access_pointer(bond->current_arp_slave) &&
2811                     !bond_time_in_interval(bond, last_rx, 3)) {
2812                         slave->new_link = BOND_LINK_DOWN;
2813                         commit++;
2814                 }
2815
2816                 /* Active slave is down if:
2817                  * - more than 2*delta since transmitting OR
2818                  * - (more than 2*delta since receive AND
2819                  *    the bond has an IP address)
2820                  */
2821                 trans_start = dev_trans_start(slave->dev);
2822                 if (bond_is_active_slave(slave) &&
2823                     (!bond_time_in_interval(bond, trans_start, 2) ||
2824                      !bond_time_in_interval(bond, last_rx, 2))) {
2825                         slave->new_link = BOND_LINK_DOWN;
2826                         commit++;
2827                 }
2828         }
2829
2830         return commit;
2831 }
2832
2833 /* Called to commit link state changes noted by inspection step of
2834  * active-backup mode ARP monitor.
2835  *
2836  * Called with RTNL hold.
2837  */
2838 static void bond_ab_arp_commit(struct bonding *bond)
2839 {
2840         unsigned long trans_start;
2841         struct list_head *iter;
2842         struct slave *slave;
2843
2844         bond_for_each_slave(bond, slave, iter) {
2845                 switch (slave->new_link) {
2846                 case BOND_LINK_NOCHANGE:
2847                         continue;
2848
2849                 case BOND_LINK_UP:
2850                         trans_start = dev_trans_start(slave->dev);
2851                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
2852                             (!rtnl_dereference(bond->curr_active_slave) &&
2853                              bond_time_in_interval(bond, trans_start, 1))) {
2854                                 struct slave *current_arp_slave;
2855
2856                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2857                                 bond_set_slave_link_state(slave, BOND_LINK_UP,
2858                                                           BOND_SLAVE_NOTIFY_NOW);
2859                                 if (current_arp_slave) {
2860                                         bond_set_slave_inactive_flags(
2861                                                 current_arp_slave,
2862                                                 BOND_SLAVE_NOTIFY_NOW);
2863                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2864                                 }
2865
2866                                 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2867                                             slave->dev->name);
2868
2869                                 if (!rtnl_dereference(bond->curr_active_slave) ||
2870                                     slave == rtnl_dereference(bond->primary_slave))
2871                                         goto do_failover;
2872
2873                         }
2874
2875                         continue;
2876
2877                 case BOND_LINK_DOWN:
2878                         if (slave->link_failure_count < UINT_MAX)
2879                                 slave->link_failure_count++;
2880
2881                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2882                                                   BOND_SLAVE_NOTIFY_NOW);
2883                         bond_set_slave_inactive_flags(slave,
2884                                                       BOND_SLAVE_NOTIFY_NOW);
2885
2886                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2887                                     slave->dev->name);
2888
2889                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
2890                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2891                                 goto do_failover;
2892                         }
2893
2894                         continue;
2895
2896                 default:
2897                         netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2898                                    slave->new_link, slave->dev->name);
2899                         continue;
2900                 }
2901
2902 do_failover:
2903                 block_netpoll_tx();
2904                 bond_select_active_slave(bond);
2905                 unblock_netpoll_tx();
2906         }
2907
2908         bond_set_carrier(bond);
2909 }
2910
2911 /* Send ARP probes for active-backup mode ARP monitor.
2912  *
2913  * Called with rcu_read_lock held.
2914  */
2915 static bool bond_ab_arp_probe(struct bonding *bond)
2916 {
2917         struct slave *slave, *before = NULL, *new_slave = NULL,
2918                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2919                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2920         struct list_head *iter;
2921         bool found = false;
2922         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2923
2924         if (curr_arp_slave && curr_active_slave)
2925                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2926                             curr_arp_slave->dev->name,
2927                             curr_active_slave->dev->name);
2928
2929         if (curr_active_slave) {
2930                 bond_arp_send_all(bond, curr_active_slave);
2931                 return should_notify_rtnl;
2932         }
2933
2934         /* if we don't have a curr_active_slave, search for the next available
2935          * backup slave from the current_arp_slave and make it the candidate
2936          * for becoming the curr_active_slave
2937          */
2938
2939         if (!curr_arp_slave) {
2940                 curr_arp_slave = bond_first_slave_rcu(bond);
2941                 if (!curr_arp_slave)
2942                         return should_notify_rtnl;
2943         }
2944
2945         bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2946
2947         bond_for_each_slave_rcu(bond, slave, iter) {
2948                 if (!found && !before && bond_slave_is_up(slave))
2949                         before = slave;
2950
2951                 if (found && !new_slave && bond_slave_is_up(slave))
2952                         new_slave = slave;
2953                 /* if the link state is up at this point, we
2954                  * mark it down - this can happen if we have
2955                  * simultaneous link failures and
2956                  * reselect_active_interface doesn't make this
2957                  * one the current slave so it is still marked
2958                  * up when it is actually down
2959                  */
2960                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2961                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2962                                                   BOND_SLAVE_NOTIFY_LATER);
2963                         if (slave->link_failure_count < UINT_MAX)
2964                                 slave->link_failure_count++;
2965
2966                         bond_set_slave_inactive_flags(slave,
2967                                                       BOND_SLAVE_NOTIFY_LATER);
2968
2969                         netdev_info(bond->dev, "backup interface %s is now down\n",
2970                                     slave->dev->name);
2971                 }
2972                 if (slave == curr_arp_slave)
2973                         found = true;
2974         }
2975
2976         if (!new_slave && before)
2977                 new_slave = before;
2978
2979         if (!new_slave)
2980                 goto check_state;
2981
2982         bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2983                                   BOND_SLAVE_NOTIFY_LATER);
2984         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2985         bond_arp_send_all(bond, new_slave);
2986         new_slave->last_link_up = jiffies;
2987         rcu_assign_pointer(bond->current_arp_slave, new_slave);
2988
2989 check_state:
2990         bond_for_each_slave_rcu(bond, slave, iter) {
2991                 if (slave->should_notify || slave->should_notify_link) {
2992                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2993                         break;
2994                 }
2995         }
2996         return should_notify_rtnl;
2997 }
2998
2999 static void bond_activebackup_arp_mon(struct bonding *bond)
3000 {
3001         bool should_notify_peers = false;
3002         bool should_notify_rtnl = false;
3003         int delta_in_ticks;
3004
3005         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3006
3007         if (!bond_has_slaves(bond))
3008                 goto re_arm;
3009
3010         rcu_read_lock();
3011
3012         should_notify_peers = bond_should_notify_peers(bond);
3013
3014         if (bond_ab_arp_inspect(bond)) {
3015                 rcu_read_unlock();
3016
3017                 /* Race avoidance with bond_close flush of workqueue */
3018                 if (!rtnl_trylock()) {
3019                         delta_in_ticks = 1;
3020                         should_notify_peers = false;
3021                         goto re_arm;
3022                 }
3023
3024                 bond_ab_arp_commit(bond);
3025
3026                 rtnl_unlock();
3027                 rcu_read_lock();
3028         }
3029
3030         should_notify_rtnl = bond_ab_arp_probe(bond);
3031         rcu_read_unlock();
3032
3033 re_arm:
3034         if (bond->params.arp_interval)
3035                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3036
3037         if (should_notify_peers || should_notify_rtnl) {
3038                 if (!rtnl_trylock())
3039                         return;
3040
3041                 if (should_notify_peers)
3042                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3043                                                  bond->dev);
3044                 if (should_notify_rtnl) {
3045                         bond_slave_state_notify(bond);
3046                         bond_slave_link_notify(bond);
3047                 }
3048
3049                 rtnl_unlock();
3050         }
3051 }
3052
3053 static void bond_arp_monitor(struct work_struct *work)
3054 {
3055         struct bonding *bond = container_of(work, struct bonding,
3056                                             arp_work.work);
3057
3058         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3059                 bond_activebackup_arp_mon(bond);
3060         else
3061                 bond_loadbalance_arp_mon(bond);
3062 }
3063
3064 /*-------------------------- netdev event handling --------------------------*/
3065
3066 /* Change device name */
3067 static int bond_event_changename(struct bonding *bond)
3068 {
3069         bond_remove_proc_entry(bond);
3070         bond_create_proc_entry(bond);
3071
3072         bond_debug_reregister(bond);
3073
3074         return NOTIFY_DONE;
3075 }
3076
3077 static int bond_master_netdev_event(unsigned long event,
3078                                     struct net_device *bond_dev)
3079 {
3080         struct bonding *event_bond = netdev_priv(bond_dev);
3081
3082         switch (event) {
3083         case NETDEV_CHANGENAME:
3084                 return bond_event_changename(event_bond);
3085         case NETDEV_UNREGISTER:
3086                 bond_remove_proc_entry(event_bond);
3087                 break;
3088         case NETDEV_REGISTER:
3089                 bond_create_proc_entry(event_bond);
3090                 break;
3091         case NETDEV_NOTIFY_PEERS:
3092                 if (event_bond->send_peer_notif)
3093                         event_bond->send_peer_notif--;
3094                 break;
3095         default:
3096                 break;
3097         }
3098
3099         return NOTIFY_DONE;
3100 }
3101
3102 static int bond_slave_netdev_event(unsigned long event,
3103                                    struct net_device *slave_dev)
3104 {
3105         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3106         struct bonding *bond;
3107         struct net_device *bond_dev;
3108
3109         /* A netdev event can be generated while enslaving a device
3110          * before netdev_rx_handler_register is called in which case
3111          * slave will be NULL
3112          */
3113         if (!slave)
3114                 return NOTIFY_DONE;
3115         bond_dev = slave->bond->dev;
3116         bond = slave->bond;
3117         primary = rtnl_dereference(bond->primary_slave);
3118
3119         switch (event) {
3120         case NETDEV_UNREGISTER:
3121                 if (bond_dev->type != ARPHRD_ETHER)
3122                         bond_release_and_destroy(bond_dev, slave_dev);
3123                 else
3124                         __bond_release_one(bond_dev, slave_dev, false, true);
3125                 break;
3126         case NETDEV_UP:
3127         case NETDEV_CHANGE:
3128                 /* For 802.3ad mode only:
3129                  * Getting invalid Speed/Duplex values here will put slave
3130                  * in weird state. So mark it as link-fail for the time
3131                  * being and let link-monitoring (miimon) set it right when
3132                  * correct speeds/duplex are available.
3133                  */
3134                 if (bond_update_speed_duplex(slave) &&
3135                     BOND_MODE(bond) == BOND_MODE_8023AD)
3136                         slave->link = BOND_LINK_FAIL;
3137
3138                 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3139                         bond_3ad_adapter_speed_duplex_changed(slave);
3140                 /* Fallthrough */
3141         case NETDEV_DOWN:
3142                 /* Refresh slave-array if applicable!
3143                  * If the setup does not use miimon or arpmon (mode-specific!),
3144                  * then these events will not cause the slave-array to be
3145                  * refreshed. This will cause xmit to use a slave that is not
3146                  * usable. Avoid such situation by refeshing the array at these
3147                  * events. If these (miimon/arpmon) parameters are configured
3148                  * then array gets refreshed twice and that should be fine!
3149                  */
3150                 if (bond_mode_can_use_xmit_hash(bond))
3151                         bond_update_slave_arr(bond, NULL);
3152                 break;
3153         case NETDEV_CHANGEMTU:
3154                 /* TODO: Should slaves be allowed to
3155                  * independently alter their MTU?  For
3156                  * an active-backup bond, slaves need
3157                  * not be the same type of device, so
3158                  * MTUs may vary.  For other modes,
3159                  * slaves arguably should have the
3160                  * same MTUs. To do this, we'd need to
3161                  * take over the slave's change_mtu
3162                  * function for the duration of their
3163                  * servitude.
3164                  */
3165                 break;
3166         case NETDEV_CHANGENAME:
3167                 /* we don't care if we don't have primary set */
3168                 if (!bond_uses_primary(bond) ||
3169                     !bond->params.primary[0])
3170                         break;
3171
3172                 if (slave == primary) {
3173                         /* slave's name changed - he's no longer primary */
3174                         RCU_INIT_POINTER(bond->primary_slave, NULL);
3175                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3176                         /* we have a new primary slave */
3177                         rcu_assign_pointer(bond->primary_slave, slave);
3178                 } else { /* we didn't change primary - exit */
3179                         break;
3180                 }
3181
3182                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3183                             primary ? slave_dev->name : "none");
3184
3185                 block_netpoll_tx();
3186                 bond_select_active_slave(bond);
3187                 unblock_netpoll_tx();
3188                 break;
3189         case NETDEV_FEAT_CHANGE:
3190                 bond_compute_features(bond);
3191                 break;
3192         case NETDEV_RESEND_IGMP:
3193                 /* Propagate to master device */
3194                 call_netdevice_notifiers(event, slave->bond->dev);
3195                 break;
3196         default:
3197                 break;
3198         }
3199
3200         return NOTIFY_DONE;
3201 }
3202
3203 /* bond_netdev_event: handle netdev notifier chain events.
3204  *
3205  * This function receives events for the netdev chain.  The caller (an
3206  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3207  * locks for us to safely manipulate the slave devices (RTNL lock,
3208  * dev_probe_lock).
3209  */
3210 static int bond_netdev_event(struct notifier_block *this,
3211                              unsigned long event, void *ptr)
3212 {
3213         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3214
3215         netdev_dbg(event_dev, "event: %lx\n", event);
3216
3217         if (!(event_dev->priv_flags & IFF_BONDING))
3218                 return NOTIFY_DONE;
3219
3220         if (event_dev->flags & IFF_MASTER) {
3221                 netdev_dbg(event_dev, "IFF_MASTER\n");
3222                 return bond_master_netdev_event(event, event_dev);
3223         }
3224
3225         if (event_dev->flags & IFF_SLAVE) {
3226                 netdev_dbg(event_dev, "IFF_SLAVE\n");
3227                 return bond_slave_netdev_event(event, event_dev);
3228         }
3229
3230         return NOTIFY_DONE;
3231 }
3232
3233 static struct notifier_block bond_netdev_notifier = {
3234         .notifier_call = bond_netdev_event,
3235 };
3236
3237 /*---------------------------- Hashing Policies -----------------------------*/
3238
3239 /* L2 hash helper */
3240 static inline u32 bond_eth_hash(struct sk_buff *skb)
3241 {
3242         struct ethhdr *ep, hdr_tmp;
3243
3244         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3245         if (ep)
3246                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3247         return 0;
3248 }
3249
3250 /* Extract the appropriate headers based on bond's xmit policy */
3251 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3252                               struct flow_keys *fk)
3253 {
3254         const struct ipv6hdr *iph6;
3255         const struct iphdr *iph;
3256         int noff, proto = -1;
3257
3258         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3259                 return skb_flow_dissect_flow_keys(skb, fk, 0);
3260
3261         fk->ports.ports = 0;
3262         noff = skb_network_offset(skb);
3263         if (skb->protocol == htons(ETH_P_IP)) {
3264                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3265                         return false;
3266                 iph = ip_hdr(skb);
3267                 iph_to_flow_copy_v4addrs(fk, iph);
3268                 noff += iph->ihl << 2;
3269                 if (!ip_is_fragment(iph))
3270                         proto = iph->protocol;
3271         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3272                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3273                         return false;
3274                 iph6 = ipv6_hdr(skb);
3275                 iph_to_flow_copy_v6addrs(fk, iph6);
3276                 noff += sizeof(*iph6);
3277                 proto = iph6->nexthdr;
3278         } else {
3279                 return false;
3280         }
3281         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3282                 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3283
3284         return true;
3285 }
3286
3287 /**
3288  * bond_xmit_hash - generate a hash value based on the xmit policy
3289  * @bond: bonding device
3290  * @skb: buffer to use for headers
3291  *
3292  * This function will extract the necessary headers from the skb buffer and use
3293  * them to generate a hash based on the xmit_policy set in the bonding device
3294  */
3295 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3296 {
3297         struct flow_keys flow;
3298         u32 hash;
3299
3300         if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3301             skb->l4_hash)
3302                 return skb->hash;
3303
3304         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3305             !bond_flow_dissect(bond, skb, &flow))
3306                 return bond_eth_hash(skb);
3307
3308         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3309             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3310                 hash = bond_eth_hash(skb);
3311         else
3312                 hash = (__force u32)flow.ports.ports;
3313         hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3314                 (__force u32)flow_get_u32_src(&flow);
3315         hash ^= (hash >> 16);
3316         hash ^= (hash >> 8);
3317
3318         return hash >> 1;
3319 }
3320
3321 /*-------------------------- Device entry points ----------------------------*/
3322
3323 void bond_work_init_all(struct bonding *bond)
3324 {
3325         INIT_DELAYED_WORK(&bond->mcast_work,
3326                           bond_resend_igmp_join_requests_delayed);
3327         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3328         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3329         INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3330         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3331         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3332 }
3333
3334 static void bond_work_cancel_all(struct bonding *bond)
3335 {
3336         cancel_delayed_work_sync(&bond->mii_work);
3337         cancel_delayed_work_sync(&bond->arp_work);
3338         cancel_delayed_work_sync(&bond->alb_work);
3339         cancel_delayed_work_sync(&bond->ad_work);
3340         cancel_delayed_work_sync(&bond->mcast_work);
3341         cancel_delayed_work_sync(&bond->slave_arr_work);
3342 }
3343
3344 static int bond_open(struct net_device *bond_dev)
3345 {
3346         struct bonding *bond = netdev_priv(bond_dev);
3347         struct list_head *iter;
3348         struct slave *slave;
3349
3350         /* reset slave->backup and slave->inactive */
3351         if (bond_has_slaves(bond)) {
3352                 bond_for_each_slave(bond, slave, iter) {
3353                         if (bond_uses_primary(bond) &&
3354                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3355                                 bond_set_slave_inactive_flags(slave,
3356                                                               BOND_SLAVE_NOTIFY_NOW);
3357                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3358                                 bond_set_slave_active_flags(slave,
3359                                                             BOND_SLAVE_NOTIFY_NOW);
3360                         }
3361                 }
3362         }
3363
3364         if (bond_is_lb(bond)) {
3365                 /* bond_alb_initialize must be called before the timer
3366                  * is started.
3367                  */
3368                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3369                         return -ENOMEM;
3370                 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
3371                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3372         }
3373
3374         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3375                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3376
3377         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3378                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3379                 bond->recv_probe = bond_arp_rcv;
3380         }
3381
3382         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3383                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3384                 /* register to receive LACPDUs */
3385                 bond->recv_probe = bond_3ad_lacpdu_recv;
3386                 bond_3ad_initiate_agg_selection(bond, 1);
3387         }
3388
3389         if (bond_mode_can_use_xmit_hash(bond))
3390                 bond_update_slave_arr(bond, NULL);
3391
3392         return 0;
3393 }
3394
3395 static int bond_close(struct net_device *bond_dev)
3396 {
3397         struct bonding *bond = netdev_priv(bond_dev);
3398
3399         bond_work_cancel_all(bond);
3400         bond->send_peer_notif = 0;
3401         if (bond_is_lb(bond))
3402                 bond_alb_deinitialize(bond);
3403         bond->recv_probe = NULL;
3404
3405         return 0;
3406 }
3407
3408 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3409  * that some drivers can provide 32bit values only.
3410  */
3411 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3412                             const struct rtnl_link_stats64 *_new,
3413                             const struct rtnl_link_stats64 *_old)
3414 {
3415         const u64 *new = (const u64 *)_new;
3416         const u64 *old = (const u64 *)_old;
3417         u64 *res = (u64 *)_res;
3418         int i;
3419
3420         for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3421                 u64 nv = new[i];
3422                 u64 ov = old[i];
3423                 s64 delta = nv - ov;
3424
3425                 /* detects if this particular field is 32bit only */
3426                 if (((nv | ov) >> 32) == 0)
3427                         delta = (s64)(s32)((u32)nv - (u32)ov);
3428
3429                 /* filter anomalies, some drivers reset their stats
3430                  * at down/up events.
3431                  */
3432                 if (delta > 0)
3433                         res[i] += delta;
3434         }
3435 }
3436
3437 static int bond_get_nest_level(struct net_device *bond_dev)
3438 {
3439         struct bonding *bond = netdev_priv(bond_dev);
3440
3441         return bond->nest_level;
3442 }
3443
3444 static void bond_get_stats(struct net_device *bond_dev,
3445                            struct rtnl_link_stats64 *stats)
3446 {
3447         struct bonding *bond = netdev_priv(bond_dev);
3448         struct rtnl_link_stats64 temp;
3449         struct list_head *iter;
3450         struct slave *slave;
3451
3452         spin_lock_nested(&bond->stats_lock, bond_get_nest_level(bond_dev));
3453         memcpy(stats, &bond->bond_stats, sizeof(*stats));
3454
3455         rcu_read_lock();
3456         bond_for_each_slave_rcu(bond, slave, iter) {
3457                 const struct rtnl_link_stats64 *new =
3458                         dev_get_stats(slave->dev, &temp);
3459
3460                 bond_fold_stats(stats, new, &slave->slave_stats);
3461
3462                 /* save off the slave stats for the next run */
3463                 memcpy(&slave->slave_stats, new, sizeof(*new));
3464         }
3465         rcu_read_unlock();
3466
3467         memcpy(&bond->bond_stats, stats, sizeof(*stats));
3468         spin_unlock(&bond->stats_lock);
3469 }
3470
3471 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3472 {
3473         struct bonding *bond = netdev_priv(bond_dev);
3474         struct net_device *slave_dev = NULL;
3475         struct ifbond k_binfo;
3476         struct ifbond __user *u_binfo = NULL;
3477         struct ifslave k_sinfo;
3478         struct ifslave __user *u_sinfo = NULL;
3479         struct mii_ioctl_data *mii = NULL;
3480         struct bond_opt_value newval;
3481         struct net *net;
3482         int res = 0;
3483
3484         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3485
3486         switch (cmd) {
3487         case SIOCGMIIPHY:
3488                 mii = if_mii(ifr);
3489                 if (!mii)
3490                         return -EINVAL;
3491
3492                 mii->phy_id = 0;
3493                 /* Fall Through */
3494         case SIOCGMIIREG:
3495                 /* We do this again just in case we were called by SIOCGMIIREG
3496                  * instead of SIOCGMIIPHY.
3497                  */
3498                 mii = if_mii(ifr);
3499                 if (!mii)
3500                         return -EINVAL;
3501
3502                 if (mii->reg_num == 1) {
3503                         mii->val_out = 0;
3504                         if (netif_carrier_ok(bond->dev))
3505                                 mii->val_out = BMSR_LSTATUS;
3506                 }
3507
3508                 return 0;
3509         case BOND_INFO_QUERY_OLD:
3510         case SIOCBONDINFOQUERY:
3511                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3512
3513                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3514                         return -EFAULT;
3515
3516                 bond_info_query(bond_dev, &k_binfo);
3517                 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3518                         return -EFAULT;
3519
3520                 return 0;
3521         case BOND_SLAVE_INFO_QUERY_OLD:
3522         case SIOCBONDSLAVEINFOQUERY:
3523                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3524
3525                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3526                         return -EFAULT;
3527
3528                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3529                 if (res == 0 &&
3530                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3531                         return -EFAULT;
3532
3533                 return res;
3534         default:
3535                 break;
3536         }
3537
3538         net = dev_net(bond_dev);
3539
3540         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3541                 return -EPERM;
3542
3543         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3544
3545         netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3546
3547         if (!slave_dev)
3548                 return -ENODEV;
3549
3550         netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3551         switch (cmd) {
3552         case BOND_ENSLAVE_OLD:
3553         case SIOCBONDENSLAVE:
3554                 res = bond_enslave(bond_dev, slave_dev, NULL);
3555                 break;
3556         case BOND_RELEASE_OLD:
3557         case SIOCBONDRELEASE:
3558                 res = bond_release(bond_dev, slave_dev);
3559                 break;
3560         case BOND_SETHWADDR_OLD:
3561         case SIOCBONDSETHWADDR:
3562                 res = bond_set_dev_addr(bond_dev, slave_dev);
3563                 break;
3564         case BOND_CHANGE_ACTIVE_OLD:
3565         case SIOCBONDCHANGEACTIVE:
3566                 bond_opt_initstr(&newval, slave_dev->name);
3567                 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
3568                                             &newval);
3569                 break;
3570         default:
3571                 res = -EOPNOTSUPP;
3572         }
3573
3574         return res;
3575 }
3576
3577 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3578 {
3579         struct bonding *bond = netdev_priv(bond_dev);
3580
3581         if (change & IFF_PROMISC)
3582                 bond_set_promiscuity(bond,
3583                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3584
3585         if (change & IFF_ALLMULTI)
3586                 bond_set_allmulti(bond,
3587                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3588 }
3589
3590 static void bond_set_rx_mode(struct net_device *bond_dev)
3591 {
3592         struct bonding *bond = netdev_priv(bond_dev);
3593         struct list_head *iter;
3594         struct slave *slave;
3595
3596         rcu_read_lock();
3597         if (bond_uses_primary(bond)) {
3598                 slave = rcu_dereference(bond->curr_active_slave);
3599                 if (slave) {
3600                         dev_uc_sync(slave->dev, bond_dev);
3601                         dev_mc_sync(slave->dev, bond_dev);
3602                 }
3603         } else {
3604                 bond_for_each_slave_rcu(bond, slave, iter) {
3605                         dev_uc_sync_multiple(slave->dev, bond_dev);
3606                         dev_mc_sync_multiple(slave->dev, bond_dev);
3607                 }
3608         }
3609         rcu_read_unlock();
3610 }
3611
3612 static int bond_neigh_init(struct neighbour *n)
3613 {
3614         struct bonding *bond = netdev_priv(n->dev);
3615         const struct net_device_ops *slave_ops;
3616         struct neigh_parms parms;
3617         struct slave *slave;
3618         int ret;
3619
3620         slave = bond_first_slave(bond);
3621         if (!slave)
3622                 return 0;
3623         slave_ops = slave->dev->netdev_ops;
3624         if (!slave_ops->ndo_neigh_setup)
3625                 return 0;
3626
3627         parms.neigh_setup = NULL;
3628         parms.neigh_cleanup = NULL;
3629         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3630         if (ret)
3631                 return ret;
3632
3633         /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3634          * after the last slave has been detached.  Assumes that all slaves
3635          * utilize the same neigh_cleanup (true at this writing as only user
3636          * is ipoib).
3637          */
3638         n->parms->neigh_cleanup = parms.neigh_cleanup;
3639
3640         if (!parms.neigh_setup)
3641                 return 0;
3642
3643         return parms.neigh_setup(n);
3644 }
3645
3646 /* The bonding ndo_neigh_setup is called at init time beofre any
3647  * slave exists. So we must declare proxy setup function which will
3648  * be used at run time to resolve the actual slave neigh param setup.
3649  *
3650  * It's also called by master devices (such as vlans) to setup their
3651  * underlying devices. In that case - do nothing, we're already set up from
3652  * our init.
3653  */
3654 static int bond_neigh_setup(struct net_device *dev,
3655                             struct neigh_parms *parms)
3656 {
3657         /* modify only our neigh_parms */
3658         if (parms->dev == dev)
3659                 parms->neigh_setup = bond_neigh_init;
3660
3661         return 0;
3662 }
3663
3664 /* Change the MTU of all of a master's slaves to match the master */
3665 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3666 {
3667         struct bonding *bond = netdev_priv(bond_dev);
3668         struct slave *slave, *rollback_slave;
3669         struct list_head *iter;
3670         int res = 0;
3671
3672         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3673
3674         bond_for_each_slave(bond, slave, iter) {
3675                 netdev_dbg(bond_dev, "s %p c_m %p\n",
3676                            slave, slave->dev->netdev_ops->ndo_change_mtu);
3677
3678                 res = dev_set_mtu(slave->dev, new_mtu);
3679
3680                 if (res) {
3681                         /* If we failed to set the slave's mtu to the new value
3682                          * we must abort the operation even in ACTIVE_BACKUP
3683                          * mode, because if we allow the backup slaves to have
3684                          * different mtu values than the active slave we'll
3685                          * need to change their mtu when doing a failover. That
3686                          * means changing their mtu from timer context, which
3687                          * is probably not a good idea.
3688                          */
3689                         netdev_dbg(bond_dev, "err %d %s\n", res,
3690                                    slave->dev->name);
3691                         goto unwind;
3692                 }
3693         }
3694
3695         bond_dev->mtu = new_mtu;
3696
3697         return 0;
3698
3699 unwind:
3700         /* unwind from head to the slave that failed */
3701         bond_for_each_slave(bond, rollback_slave, iter) {
3702                 int tmp_res;
3703
3704                 if (rollback_slave == slave)
3705                         break;
3706
3707                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3708                 if (tmp_res) {
3709                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3710                                    tmp_res, rollback_slave->dev->name);
3711                 }
3712         }
3713
3714         return res;
3715 }
3716
3717 /* Change HW address
3718  *
3719  * Note that many devices must be down to change the HW address, and
3720  * downing the master releases all slaves.  We can make bonds full of
3721  * bonding devices to test this, however.
3722  */
3723 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3724 {
3725         struct bonding *bond = netdev_priv(bond_dev);
3726         struct slave *slave, *rollback_slave;
3727         struct sockaddr_storage *ss = addr, tmp_ss;
3728         struct list_head *iter;
3729         int res = 0;
3730
3731         if (BOND_MODE(bond) == BOND_MODE_ALB)
3732                 return bond_alb_set_mac_address(bond_dev, addr);
3733
3734
3735         netdev_dbg(bond_dev, "bond=%p\n", bond);
3736
3737         /* If fail_over_mac is enabled, do nothing and return success.
3738          * Returning an error causes ifenslave to fail.
3739          */
3740         if (bond->params.fail_over_mac &&
3741             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3742                 return 0;
3743
3744         if (!is_valid_ether_addr(ss->__data))
3745                 return -EADDRNOTAVAIL;
3746
3747         bond_for_each_slave(bond, slave, iter) {
3748                 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3749                 res = dev_set_mac_address(slave->dev, addr, NULL);
3750                 if (res) {
3751                         /* TODO: consider downing the slave
3752                          * and retry ?
3753                          * User should expect communications
3754                          * breakage anyway until ARP finish
3755                          * updating, so...
3756                          */
3757                         netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3758                         goto unwind;
3759                 }
3760         }
3761
3762         /* success */
3763         memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
3764         return 0;
3765
3766 unwind:
3767         memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
3768         tmp_ss.ss_family = bond_dev->type;
3769
3770         /* unwind from head to the slave that failed */
3771         bond_for_each_slave(bond, rollback_slave, iter) {
3772                 int tmp_res;
3773
3774                 if (rollback_slave == slave)
3775                         break;
3776
3777                 tmp_res = dev_set_mac_address(rollback_slave->dev,
3778                                               (struct sockaddr *)&tmp_ss, NULL);
3779                 if (tmp_res) {
3780                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3781                                    tmp_res, rollback_slave->dev->name);
3782                 }
3783         }
3784
3785         return res;
3786 }
3787
3788 /**
3789  * bond_xmit_slave_id - transmit skb through slave with slave_id
3790  * @bond: bonding device that is transmitting
3791  * @skb: buffer to transmit
3792  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3793  *
3794  * This function tries to transmit through slave with slave_id but in case
3795  * it fails, it tries to find the first available slave for transmission.
3796  * The skb is consumed in all cases, thus the function is void.
3797  */
3798 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3799 {
3800         struct list_head *iter;
3801         struct slave *slave;
3802         int i = slave_id;
3803
3804         /* Here we start from the slave with slave_id */
3805         bond_for_each_slave_rcu(bond, slave, iter) {
3806                 if (--i < 0) {
3807                         if (bond_slave_can_tx(slave)) {
3808                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3809                                 return;
3810                         }
3811                 }
3812         }
3813
3814         /* Here we start from the first slave up to slave_id */
3815         i = slave_id;
3816         bond_for_each_slave_rcu(bond, slave, iter) {
3817                 if (--i < 0)
3818                         break;
3819                 if (bond_slave_can_tx(slave)) {
3820                         bond_dev_queue_xmit(bond, skb, slave->dev);
3821                         return;
3822                 }
3823         }
3824         /* no slave that can tx has been found */
3825         bond_tx_drop(bond->dev, skb);
3826 }
3827
3828 /**
3829  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3830  * @bond: bonding device to use
3831  *
3832  * Based on the value of the bonding device's packets_per_slave parameter
3833  * this function generates a slave id, which is usually used as the next
3834  * slave to transmit through.
3835  */
3836 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3837 {
3838         u32 slave_id;
3839         struct reciprocal_value reciprocal_packets_per_slave;
3840         int packets_per_slave = bond->params.packets_per_slave;
3841
3842         switch (packets_per_slave) {
3843         case 0:
3844                 slave_id = prandom_u32();
3845                 break;
3846         case 1:
3847                 slave_id = bond->rr_tx_counter;
3848                 break;
3849         default:
3850                 reciprocal_packets_per_slave =
3851                         bond->params.reciprocal_packets_per_slave;
3852                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3853                                              reciprocal_packets_per_slave);
3854                 break;
3855         }
3856         bond->rr_tx_counter++;
3857
3858         return slave_id;
3859 }
3860
3861 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
3862                                         struct net_device *bond_dev)
3863 {
3864         struct bonding *bond = netdev_priv(bond_dev);
3865         struct iphdr *iph = ip_hdr(skb);
3866         struct slave *slave;
3867         u32 slave_id;
3868
3869         /* Start with the curr_active_slave that joined the bond as the
3870          * default for sending IGMP traffic.  For failover purposes one
3871          * needs to maintain some consistency for the interface that will
3872          * send the join/membership reports.  The curr_active_slave found
3873          * will send all of this type of traffic.
3874          */
3875         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3876                 slave = rcu_dereference(bond->curr_active_slave);
3877                 if (slave)
3878                         bond_dev_queue_xmit(bond, skb, slave->dev);
3879                 else
3880                         bond_xmit_slave_id(bond, skb, 0);
3881         } else {
3882                 int slave_cnt = READ_ONCE(bond->slave_cnt);
3883
3884                 if (likely(slave_cnt)) {
3885                         slave_id = bond_rr_gen_slave_id(bond);
3886                         bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3887                 } else {
3888                         bond_tx_drop(bond_dev, skb);
3889                 }
3890         }
3891
3892         return NETDEV_TX_OK;
3893 }
3894
3895 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3896  * the bond has a usable interface.
3897  */
3898 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
3899                                           struct net_device *bond_dev)
3900 {
3901         struct bonding *bond = netdev_priv(bond_dev);
3902         struct slave *slave;
3903
3904         slave = rcu_dereference(bond->curr_active_slave);
3905         if (slave)
3906                 bond_dev_queue_xmit(bond, skb, slave->dev);
3907         else
3908                 bond_tx_drop(bond_dev, skb);
3909
3910         return NETDEV_TX_OK;
3911 }
3912
3913 /* Use this to update slave_array when (a) it's not appropriate to update
3914  * slave_array right away (note that update_slave_array() may sleep)
3915  * and / or (b) RTNL is not held.
3916  */
3917 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3918 {
3919         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3920 }
3921
3922 /* Slave array work handler. Holds only RTNL */
3923 static void bond_slave_arr_handler(struct work_struct *work)
3924 {
3925         struct bonding *bond = container_of(work, struct bonding,
3926                                             slave_arr_work.work);
3927         int ret;
3928
3929         if (!rtnl_trylock())
3930                 goto err;
3931
3932         ret = bond_update_slave_arr(bond, NULL);
3933         rtnl_unlock();
3934         if (ret) {
3935                 pr_warn_ratelimited("Failed to update slave array from WT\n");
3936                 goto err;
3937         }
3938         return;
3939
3940 err:
3941         bond_slave_arr_work_rearm(bond, 1);
3942 }
3943
3944 /* Build the usable slaves array in control path for modes that use xmit-hash
3945  * to determine the slave interface -
3946  * (a) BOND_MODE_8023AD
3947  * (b) BOND_MODE_XOR
3948  * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
3949  *
3950  * The caller is expected to hold RTNL only and NO other lock!
3951  */
3952 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3953 {
3954         struct slave *slave;
3955         struct list_head *iter;
3956         struct bond_up_slave *new_arr, *old_arr;
3957         int agg_id = 0;
3958         int ret = 0;
3959
3960 #ifdef CONFIG_LOCKDEP
3961         WARN_ON(lockdep_is_held(&bond->mode_lock));
3962 #endif
3963
3964         new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3965                           GFP_KERNEL);
3966         if (!new_arr) {
3967                 ret = -ENOMEM;
3968                 pr_err("Failed to build slave-array.\n");
3969                 goto out;
3970         }
3971         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3972                 struct ad_info ad_info;
3973
3974                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3975                         pr_debug("bond_3ad_get_active_agg_info failed\n");
3976                         kfree_rcu(new_arr, rcu);
3977                         /* No active aggragator means it's not safe to use
3978                          * the previous array.
3979                          */
3980                         old_arr = rtnl_dereference(bond->slave_arr);
3981                         if (old_arr) {
3982                                 RCU_INIT_POINTER(bond->slave_arr, NULL);
3983                                 kfree_rcu(old_arr, rcu);
3984                         }
3985                         goto out;
3986                 }
3987                 agg_id = ad_info.aggregator_id;
3988         }
3989         bond_for_each_slave(bond, slave, iter) {
3990                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3991                         struct aggregator *agg;
3992
3993                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
3994                         if (!agg || agg->aggregator_identifier != agg_id)
3995                                 continue;
3996                 }
3997                 if (!bond_slave_can_tx(slave))
3998                         continue;
3999                 if (skipslave == slave)
4000                         continue;
4001
4002                 netdev_dbg(bond->dev,
4003                            "Adding slave dev %s to tx hash array[%d]\n",
4004                            slave->dev->name, new_arr->count);
4005
4006                 new_arr->arr[new_arr->count++] = slave;
4007         }
4008
4009         old_arr = rtnl_dereference(bond->slave_arr);
4010         rcu_assign_pointer(bond->slave_arr, new_arr);
4011         if (old_arr)
4012                 kfree_rcu(old_arr, rcu);
4013 out:
4014         if (ret != 0 && skipslave) {
4015                 int idx;
4016
4017                 /* Rare situation where caller has asked to skip a specific
4018                  * slave but allocation failed (most likely!). BTW this is
4019                  * only possible when the call is initiated from
4020                  * __bond_release_one(). In this situation; overwrite the
4021                  * skipslave entry in the array with the last entry from the
4022                  * array to avoid a situation where the xmit path may choose
4023                  * this to-be-skipped slave to send a packet out.
4024                  */
4025                 old_arr = rtnl_dereference(bond->slave_arr);
4026                 for (idx = 0; idx < old_arr->count; idx++) {
4027                         if (skipslave == old_arr->arr[idx]) {
4028                                 old_arr->arr[idx] =
4029                                     old_arr->arr[old_arr->count-1];
4030                                 old_arr->count--;
4031                                 break;
4032                         }
4033                 }
4034         }
4035         return ret;
4036 }
4037
4038 /* Use this Xmit function for 3AD as well as XOR modes. The current
4039  * usable slave array is formed in the control path. The xmit function
4040  * just calculates hash and sends the packet out.
4041  */
4042 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
4043                                      struct net_device *dev)
4044 {
4045         struct bonding *bond = netdev_priv(dev);
4046         struct slave *slave;
4047         struct bond_up_slave *slaves;
4048         unsigned int count;
4049
4050         slaves = rcu_dereference(bond->slave_arr);
4051         count = slaves ? READ_ONCE(slaves->count) : 0;
4052         if (likely(count)) {
4053                 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
4054                 bond_dev_queue_xmit(bond, skb, slave->dev);
4055         } else {
4056                 bond_tx_drop(dev, skb);
4057         }
4058
4059         return NETDEV_TX_OK;
4060 }
4061
4062 /* in broadcast mode, we send everything to all usable interfaces. */
4063 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
4064                                        struct net_device *bond_dev)
4065 {
4066         struct bonding *bond = netdev_priv(bond_dev);
4067         struct slave *slave = NULL;
4068         struct list_head *iter;
4069
4070         bond_for_each_slave_rcu(bond, slave, iter) {
4071                 if (bond_is_last_slave(bond, slave))
4072                         break;
4073                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4074                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4075
4076                         if (!skb2) {
4077                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4078                                                     bond_dev->name, __func__);
4079                                 continue;
4080                         }
4081                         bond_dev_queue_xmit(bond, skb2, slave->dev);
4082                 }
4083         }
4084         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4085                 bond_dev_queue_xmit(bond, skb, slave->dev);
4086         else
4087                 bond_tx_drop(bond_dev, skb);
4088
4089         return NETDEV_TX_OK;
4090 }
4091
4092 /*------------------------- Device initialization ---------------------------*/
4093
4094 /* Lookup the slave that corresponds to a qid */
4095 static inline int bond_slave_override(struct bonding *bond,
4096                                       struct sk_buff *skb)
4097 {
4098         struct slave *slave = NULL;
4099         struct list_head *iter;
4100
4101         if (!skb_rx_queue_recorded(skb))
4102                 return 1;
4103
4104         /* Find out if any slaves have the same mapping as this skb. */
4105         bond_for_each_slave_rcu(bond, slave, iter) {
4106                 if (slave->queue_id == skb_get_queue_mapping(skb)) {
4107                         if (bond_slave_is_up(slave) &&
4108                             slave->link == BOND_LINK_UP) {
4109                                 bond_dev_queue_xmit(bond, skb, slave->dev);
4110                                 return 0;
4111                         }
4112                         /* If the slave isn't UP, use default transmit policy. */
4113                         break;
4114                 }
4115         }
4116
4117         return 1;
4118 }
4119
4120
4121 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4122                              struct net_device *sb_dev,
4123                              select_queue_fallback_t fallback)
4124 {
4125         /* This helper function exists to help dev_pick_tx get the correct
4126          * destination queue.  Using a helper function skips a call to
4127          * skb_tx_hash and will put the skbs in the queue we expect on their
4128          * way down to the bonding driver.
4129          */
4130         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4131
4132         /* Save the original txq to restore before passing to the driver */
4133         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
4134
4135         if (unlikely(txq >= dev->real_num_tx_queues)) {
4136                 do {
4137                         txq -= dev->real_num_tx_queues;
4138                 } while (txq >= dev->real_num_tx_queues);
4139         }
4140         return txq;
4141 }
4142
4143 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4144 {
4145         struct bonding *bond = netdev_priv(dev);
4146
4147         if (bond_should_override_tx_queue(bond) &&
4148             !bond_slave_override(bond, skb))
4149                 return NETDEV_TX_OK;
4150
4151         switch (BOND_MODE(bond)) {
4152         case BOND_MODE_ROUNDROBIN:
4153                 return bond_xmit_roundrobin(skb, dev);
4154         case BOND_MODE_ACTIVEBACKUP:
4155                 return bond_xmit_activebackup(skb, dev);
4156         case BOND_MODE_8023AD:
4157         case BOND_MODE_XOR:
4158                 return bond_3ad_xor_xmit(skb, dev);
4159         case BOND_MODE_BROADCAST:
4160                 return bond_xmit_broadcast(skb, dev);
4161         case BOND_MODE_ALB:
4162                 return bond_alb_xmit(skb, dev);
4163         case BOND_MODE_TLB:
4164                 return bond_tlb_xmit(skb, dev);
4165         default:
4166                 /* Should never happen, mode already checked */
4167                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4168                 WARN_ON_ONCE(1);
4169                 bond_tx_drop(dev, skb);
4170                 return NETDEV_TX_OK;
4171         }
4172 }
4173
4174 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4175 {
4176         struct bonding *bond = netdev_priv(dev);
4177         netdev_tx_t ret = NETDEV_TX_OK;
4178
4179         /* If we risk deadlock from transmitting this in the
4180          * netpoll path, tell netpoll to queue the frame for later tx
4181          */
4182         if (unlikely(is_netpoll_tx_blocked(dev)))
4183                 return NETDEV_TX_BUSY;
4184
4185         rcu_read_lock();
4186         if (bond_has_slaves(bond))
4187                 ret = __bond_start_xmit(skb, dev);
4188         else
4189                 bond_tx_drop(dev, skb);
4190         rcu_read_unlock();
4191
4192         return ret;
4193 }
4194
4195 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4196                                            struct ethtool_link_ksettings *cmd)
4197 {
4198         struct bonding *bond = netdev_priv(bond_dev);
4199         unsigned long speed = 0;
4200         struct list_head *iter;
4201         struct slave *slave;
4202
4203         cmd->base.duplex = DUPLEX_UNKNOWN;
4204         cmd->base.port = PORT_OTHER;
4205
4206         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4207          * do not need to check mode.  Though link speed might not represent
4208          * the true receive or transmit bandwidth (not all modes are symmetric)
4209          * this is an accurate maximum.
4210          */
4211         bond_for_each_slave(bond, slave, iter) {
4212                 if (bond_slave_can_tx(slave)) {
4213                         if (slave->speed != SPEED_UNKNOWN)
4214                                 speed += slave->speed;
4215                         if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4216                             slave->duplex != DUPLEX_UNKNOWN)
4217                                 cmd->base.duplex = slave->duplex;
4218                 }
4219         }
4220         cmd->base.speed = speed ? : SPEED_UNKNOWN;
4221
4222         return 0;
4223 }
4224
4225 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4226                                      struct ethtool_drvinfo *drvinfo)
4227 {
4228         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4229         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4230         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4231                  BOND_ABI_VERSION);
4232 }
4233
4234 static const struct ethtool_ops bond_ethtool_ops = {
4235         .get_drvinfo            = bond_ethtool_get_drvinfo,
4236         .get_link               = ethtool_op_get_link,
4237         .get_link_ksettings     = bond_ethtool_get_link_ksettings,
4238 };
4239
4240 static const struct net_device_ops bond_netdev_ops = {
4241         .ndo_init               = bond_init,
4242         .ndo_uninit             = bond_uninit,
4243         .ndo_open               = bond_open,
4244         .ndo_stop               = bond_close,
4245         .ndo_start_xmit         = bond_start_xmit,
4246         .ndo_select_queue       = bond_select_queue,
4247         .ndo_get_stats64        = bond_get_stats,
4248         .ndo_do_ioctl           = bond_do_ioctl,
4249         .ndo_change_rx_flags    = bond_change_rx_flags,
4250         .ndo_set_rx_mode        = bond_set_rx_mode,
4251         .ndo_change_mtu         = bond_change_mtu,
4252         .ndo_set_mac_address    = bond_set_mac_address,
4253         .ndo_neigh_setup        = bond_neigh_setup,
4254         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4255         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4256         .ndo_get_lock_subclass  = bond_get_nest_level,
4257 #ifdef CONFIG_NET_POLL_CONTROLLER
4258         .ndo_netpoll_setup      = bond_netpoll_setup,
4259         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4260         .ndo_poll_controller    = bond_poll_controller,
4261 #endif
4262         .ndo_add_slave          = bond_enslave,
4263         .ndo_del_slave          = bond_release,
4264         .ndo_fix_features       = bond_fix_features,
4265         .ndo_features_check     = passthru_features_check,
4266 };
4267
4268 static const struct device_type bond_type = {
4269         .name = "bond",
4270 };
4271
4272 static void bond_destructor(struct net_device *bond_dev)
4273 {
4274         struct bonding *bond = netdev_priv(bond_dev);
4275         if (bond->wq)
4276                 destroy_workqueue(bond->wq);
4277 }
4278
4279 void bond_setup(struct net_device *bond_dev)
4280 {
4281         struct bonding *bond = netdev_priv(bond_dev);
4282
4283         spin_lock_init(&bond->mode_lock);
4284         spin_lock_init(&bond->stats_lock);
4285         bond->params = bonding_defaults;
4286
4287         /* Initialize pointers */
4288         bond->dev = bond_dev;
4289
4290         /* Initialize the device entry points */
4291         ether_setup(bond_dev);
4292         bond_dev->max_mtu = ETH_MAX_MTU;
4293         bond_dev->netdev_ops = &bond_netdev_ops;
4294         bond_dev->ethtool_ops = &bond_ethtool_ops;
4295
4296         bond_dev->needs_free_netdev = true;
4297         bond_dev->priv_destructor = bond_destructor;
4298
4299         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4300
4301         /* Initialize the device options */
4302         bond_dev->flags |= IFF_MASTER;
4303         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4304         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4305
4306         /* don't acquire bond device's netif_tx_lock when transmitting */
4307         bond_dev->features |= NETIF_F_LLTX;
4308
4309         /* By default, we declare the bond to be fully
4310          * VLAN hardware accelerated capable. Special
4311          * care is taken in the various xmit functions
4312          * when there are slaves that are not hw accel
4313          * capable
4314          */
4315
4316         /* Don't allow bond devices to change network namespaces. */
4317         bond_dev->features |= NETIF_F_NETNS_LOCAL;
4318
4319         bond_dev->hw_features = BOND_VLAN_FEATURES |
4320                                 NETIF_F_HW_VLAN_CTAG_TX |
4321                                 NETIF_F_HW_VLAN_CTAG_RX |
4322                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4323
4324         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4;
4325         bond_dev->features |= bond_dev->hw_features;
4326 }
4327
4328 /* Destroy a bonding device.
4329  * Must be under rtnl_lock when this function is called.
4330  */
4331 static void bond_uninit(struct net_device *bond_dev)
4332 {
4333         struct bonding *bond = netdev_priv(bond_dev);
4334         struct list_head *iter;
4335         struct slave *slave;
4336         struct bond_up_slave *arr;
4337
4338         bond_netpoll_cleanup(bond_dev);
4339
4340         /* Release the bonded slaves */
4341         bond_for_each_slave(bond, slave, iter)
4342                 __bond_release_one(bond_dev, slave->dev, true, true);
4343         netdev_info(bond_dev, "Released all slaves\n");
4344
4345         arr = rtnl_dereference(bond->slave_arr);
4346         if (arr) {
4347                 RCU_INIT_POINTER(bond->slave_arr, NULL);
4348                 kfree_rcu(arr, rcu);
4349         }
4350
4351         list_del(&bond->bond_list);
4352
4353         bond_debug_unregister(bond);
4354 }
4355
4356 /*------------------------- Module initialization ---------------------------*/
4357
4358 static int bond_check_params(struct bond_params *params)
4359 {
4360         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4361         struct bond_opt_value newval;
4362         const struct bond_opt_value *valptr;
4363         int arp_all_targets_value = 0;
4364         u16 ad_actor_sys_prio = 0;
4365         u16 ad_user_port_key = 0;
4366         __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
4367         int arp_ip_count;
4368         int bond_mode   = BOND_MODE_ROUNDROBIN;
4369         int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4370         int lacp_fast = 0;
4371         int tlb_dynamic_lb;
4372
4373         /* Convert string parameters. */
4374         if (mode) {
4375                 bond_opt_initstr(&newval, mode);
4376                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4377                 if (!valptr) {
4378                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4379                         return -EINVAL;
4380                 }
4381                 bond_mode = valptr->value;
4382         }
4383
4384         if (xmit_hash_policy) {
4385                 if (bond_mode == BOND_MODE_ROUNDROBIN ||
4386                     bond_mode == BOND_MODE_ACTIVEBACKUP ||
4387                     bond_mode == BOND_MODE_BROADCAST) {
4388                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4389                                 bond_mode_name(bond_mode));
4390                 } else {
4391                         bond_opt_initstr(&newval, xmit_hash_policy);
4392                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4393                                                 &newval);
4394                         if (!valptr) {
4395                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4396                                        xmit_hash_policy);
4397                                 return -EINVAL;
4398                         }
4399                         xmit_hashtype = valptr->value;
4400                 }
4401         }
4402
4403         if (lacp_rate) {
4404                 if (bond_mode != BOND_MODE_8023AD) {
4405                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4406                                 bond_mode_name(bond_mode));
4407                 } else {
4408                         bond_opt_initstr(&newval, lacp_rate);
4409                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4410                                                 &newval);
4411                         if (!valptr) {
4412                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4413                                        lacp_rate);
4414                                 return -EINVAL;
4415                         }
4416                         lacp_fast = valptr->value;
4417                 }
4418         }
4419
4420         if (ad_select) {
4421                 bond_opt_initstr(&newval, ad_select);
4422                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4423                                         &newval);
4424                 if (!valptr) {
4425                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4426                         return -EINVAL;
4427                 }
4428                 params->ad_select = valptr->value;
4429                 if (bond_mode != BOND_MODE_8023AD)
4430                         pr_warn("ad_select param only affects 802.3ad mode\n");
4431         } else {
4432                 params->ad_select = BOND_AD_STABLE;
4433         }
4434
4435         if (max_bonds < 0) {
4436                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4437                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4438                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4439         }
4440
4441         if (miimon < 0) {
4442                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4443                         miimon, INT_MAX);
4444                 miimon = 0;
4445         }
4446
4447         if (updelay < 0) {
4448                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4449                         updelay, INT_MAX);
4450                 updelay = 0;
4451         }
4452
4453         if (downdelay < 0) {
4454                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4455                         downdelay, INT_MAX);
4456                 downdelay = 0;
4457         }
4458
4459         if ((use_carrier != 0) && (use_carrier != 1)) {
4460                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4461                         use_carrier);
4462                 use_carrier = 1;
4463         }
4464
4465         if (num_peer_notif < 0 || num_peer_notif > 255) {
4466                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4467                         num_peer_notif);
4468                 num_peer_notif = 1;
4469         }
4470
4471         /* reset values for 802.3ad/TLB/ALB */
4472         if (!bond_mode_uses_arp(bond_mode)) {
4473                 if (!miimon) {
4474                         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");
4475                         pr_warn("Forcing miimon to 100msec\n");
4476                         miimon = BOND_DEFAULT_MIIMON;
4477                 }
4478         }
4479
4480         if (tx_queues < 1 || tx_queues > 255) {
4481                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4482                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4483                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4484         }
4485
4486         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4487                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4488                         all_slaves_active);
4489                 all_slaves_active = 0;
4490         }
4491
4492         if (resend_igmp < 0 || resend_igmp > 255) {
4493                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4494                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4495                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4496         }
4497
4498         bond_opt_initval(&newval, packets_per_slave);
4499         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4500                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4501                         packets_per_slave, USHRT_MAX);
4502                 packets_per_slave = 1;
4503         }
4504
4505         if (bond_mode == BOND_MODE_ALB) {
4506                 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",
4507                           updelay);
4508         }
4509
4510         if (!miimon) {
4511                 if (updelay || downdelay) {
4512                         /* just warn the user the up/down delay will have
4513                          * no effect since miimon is zero...
4514                          */
4515                         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",
4516                                 updelay, downdelay);
4517                 }
4518         } else {
4519                 /* don't allow arp monitoring */
4520                 if (arp_interval) {
4521                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4522                                 miimon, arp_interval);
4523                         arp_interval = 0;
4524                 }
4525
4526                 if ((updelay % miimon) != 0) {
4527                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4528                                 updelay, miimon, (updelay / miimon) * miimon);
4529                 }
4530
4531                 updelay /= miimon;
4532
4533                 if ((downdelay % miimon) != 0) {
4534                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4535                                 downdelay, miimon,
4536                                 (downdelay / miimon) * miimon);
4537                 }
4538
4539                 downdelay /= miimon;
4540         }
4541
4542         if (arp_interval < 0) {
4543                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4544                         arp_interval, INT_MAX);
4545                 arp_interval = 0;
4546         }
4547
4548         for (arp_ip_count = 0, i = 0;
4549              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4550                 __be32 ip;
4551
4552                 /* not a complete check, but good enough to catch mistakes */
4553                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4554                     !bond_is_ip_target_ok(ip)) {
4555                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4556                                 arp_ip_target[i]);
4557                         arp_interval = 0;
4558                 } else {
4559                         if (bond_get_targets_ip(arp_target, ip) == -1)
4560                                 arp_target[arp_ip_count++] = ip;
4561                         else
4562                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4563                                         &ip);
4564                 }
4565         }
4566
4567         if (arp_interval && !arp_ip_count) {
4568                 /* don't allow arping if no arp_ip_target given... */
4569                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4570                         arp_interval);
4571                 arp_interval = 0;
4572         }
4573
4574         if (arp_validate) {
4575                 if (!arp_interval) {
4576                         pr_err("arp_validate requires arp_interval\n");
4577                         return -EINVAL;
4578                 }
4579
4580                 bond_opt_initstr(&newval, arp_validate);
4581                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4582                                         &newval);
4583                 if (!valptr) {
4584                         pr_err("Error: invalid arp_validate \"%s\"\n",
4585                                arp_validate);
4586                         return -EINVAL;
4587                 }
4588                 arp_validate_value = valptr->value;
4589         } else {
4590                 arp_validate_value = 0;
4591         }
4592
4593         if (arp_all_targets) {
4594                 bond_opt_initstr(&newval, arp_all_targets);
4595                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4596                                         &newval);
4597                 if (!valptr) {
4598                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4599                                arp_all_targets);
4600                         arp_all_targets_value = 0;
4601                 } else {
4602                         arp_all_targets_value = valptr->value;
4603                 }
4604         }
4605
4606         if (miimon) {
4607                 pr_info("MII link monitoring set to %d ms\n", miimon);
4608         } else if (arp_interval) {
4609                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4610                                           arp_validate_value);
4611                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4612                         arp_interval, valptr->string, arp_ip_count);
4613
4614                 for (i = 0; i < arp_ip_count; i++)
4615                         pr_cont(" %s", arp_ip_target[i]);
4616
4617                 pr_cont("\n");
4618
4619         } else if (max_bonds) {
4620                 /* miimon and arp_interval not set, we need one so things
4621                  * work as expected, see bonding.txt for details
4622                  */
4623                 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");
4624         }
4625
4626         if (primary && !bond_mode_uses_primary(bond_mode)) {
4627                 /* currently, using a primary only makes sense
4628                  * in active backup, TLB or ALB modes
4629                  */
4630                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4631                         primary, bond_mode_name(bond_mode));
4632                 primary = NULL;
4633         }
4634
4635         if (primary && primary_reselect) {
4636                 bond_opt_initstr(&newval, primary_reselect);
4637                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4638                                         &newval);
4639                 if (!valptr) {
4640                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4641                                primary_reselect);
4642                         return -EINVAL;
4643                 }
4644                 primary_reselect_value = valptr->value;
4645         } else {
4646                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4647         }
4648
4649         if (fail_over_mac) {
4650                 bond_opt_initstr(&newval, fail_over_mac);
4651                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4652                                         &newval);
4653                 if (!valptr) {
4654                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4655                                fail_over_mac);
4656                         return -EINVAL;
4657                 }
4658                 fail_over_mac_value = valptr->value;
4659                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4660                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4661         } else {
4662                 fail_over_mac_value = BOND_FOM_NONE;
4663         }
4664
4665         bond_opt_initstr(&newval, "default");
4666         valptr = bond_opt_parse(
4667                         bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4668                                      &newval);
4669         if (!valptr) {
4670                 pr_err("Error: No ad_actor_sys_prio default value");
4671                 return -EINVAL;
4672         }
4673         ad_actor_sys_prio = valptr->value;
4674
4675         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4676                                 &newval);
4677         if (!valptr) {
4678                 pr_err("Error: No ad_user_port_key default value");
4679                 return -EINVAL;
4680         }
4681         ad_user_port_key = valptr->value;
4682
4683         bond_opt_initstr(&newval, "default");
4684         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
4685         if (!valptr) {
4686                 pr_err("Error: No tlb_dynamic_lb default value");
4687                 return -EINVAL;
4688         }
4689         tlb_dynamic_lb = valptr->value;
4690
4691         if (lp_interval == 0) {
4692                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4693                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4694                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4695         }
4696
4697         /* fill params struct with the proper values */
4698         params->mode = bond_mode;
4699         params->xmit_policy = xmit_hashtype;
4700         params->miimon = miimon;
4701         params->num_peer_notif = num_peer_notif;
4702         params->arp_interval = arp_interval;
4703         params->arp_validate = arp_validate_value;
4704         params->arp_all_targets = arp_all_targets_value;
4705         params->updelay = updelay;
4706         params->downdelay = downdelay;
4707         params->use_carrier = use_carrier;
4708         params->lacp_fast = lacp_fast;
4709         params->primary[0] = 0;
4710         params->primary_reselect = primary_reselect_value;
4711         params->fail_over_mac = fail_over_mac_value;
4712         params->tx_queues = tx_queues;
4713         params->all_slaves_active = all_slaves_active;
4714         params->resend_igmp = resend_igmp;
4715         params->min_links = min_links;
4716         params->lp_interval = lp_interval;
4717         params->packets_per_slave = packets_per_slave;
4718         params->tlb_dynamic_lb = tlb_dynamic_lb;
4719         params->ad_actor_sys_prio = ad_actor_sys_prio;
4720         eth_zero_addr(params->ad_actor_system);
4721         params->ad_user_port_key = ad_user_port_key;
4722         if (packets_per_slave > 0) {
4723                 params->reciprocal_packets_per_slave =
4724                         reciprocal_value(packets_per_slave);
4725         } else {
4726                 /* reciprocal_packets_per_slave is unused if
4727                  * packets_per_slave is 0 or 1, just initialize it
4728                  */
4729                 params->reciprocal_packets_per_slave =
4730                         (struct reciprocal_value) { 0 };
4731         }
4732
4733         if (primary) {
4734                 strncpy(params->primary, primary, IFNAMSIZ);
4735                 params->primary[IFNAMSIZ - 1] = 0;
4736         }
4737
4738         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4739
4740         return 0;
4741 }
4742
4743 /* Called from registration process */
4744 static int bond_init(struct net_device *bond_dev)
4745 {
4746         struct bonding *bond = netdev_priv(bond_dev);
4747         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4748
4749         netdev_dbg(bond_dev, "Begin bond_init\n");
4750
4751         bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
4752         if (!bond->wq)
4753                 return -ENOMEM;
4754
4755         bond->nest_level = SINGLE_DEPTH_NESTING;
4756         netdev_lockdep_set_classes(bond_dev);
4757
4758         list_add_tail(&bond->bond_list, &bn->dev_list);
4759
4760         bond_prepare_sysfs_group(bond);
4761
4762         bond_debug_register(bond);
4763
4764         /* Ensure valid dev_addr */
4765         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4766             bond_dev->addr_assign_type == NET_ADDR_PERM)
4767                 eth_hw_addr_random(bond_dev);
4768
4769         return 0;
4770 }
4771
4772 unsigned int bond_get_num_tx_queues(void)
4773 {
4774         return tx_queues;
4775 }
4776
4777 /* Create a new bond based on the specified name and bonding parameters.
4778  * If name is NULL, obtain a suitable "bond%d" name for us.
4779  * Caller must NOT hold rtnl_lock; we need to release it here before we
4780  * set up our sysfs entries.
4781  */
4782 int bond_create(struct net *net, const char *name)
4783 {
4784         struct net_device *bond_dev;
4785         struct bonding *bond;
4786         struct alb_bond_info *bond_info;
4787         int res;
4788
4789         rtnl_lock();
4790
4791         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4792                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
4793                                    bond_setup, tx_queues);
4794         if (!bond_dev) {
4795                 pr_err("%s: eek! can't alloc netdev!\n", name);
4796                 rtnl_unlock();
4797                 return -ENOMEM;
4798         }
4799
4800         /*
4801          * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4802          * It is set to 0 by default which is wrong.
4803          */
4804         bond = netdev_priv(bond_dev);
4805         bond_info = &(BOND_ALB_INFO(bond));
4806         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4807
4808         dev_net_set(bond_dev, net);
4809         bond_dev->rtnl_link_ops = &bond_link_ops;
4810
4811         res = register_netdevice(bond_dev);
4812
4813         netif_carrier_off(bond_dev);
4814
4815         bond_work_init_all(bond);
4816
4817         rtnl_unlock();
4818         if (res < 0)
4819                 free_netdev(bond_dev);
4820         return res;
4821 }
4822
4823 static int __net_init bond_net_init(struct net *net)
4824 {
4825         struct bond_net *bn = net_generic(net, bond_net_id);
4826
4827         bn->net = net;
4828         INIT_LIST_HEAD(&bn->dev_list);
4829
4830         bond_create_proc_dir(bn);
4831         bond_create_sysfs(bn);
4832
4833         return 0;
4834 }
4835
4836 static void __net_exit bond_net_exit(struct net *net)
4837 {
4838         struct bond_net *bn = net_generic(net, bond_net_id);
4839         struct bonding *bond, *tmp_bond;
4840         LIST_HEAD(list);
4841
4842         bond_destroy_sysfs(bn);
4843
4844         /* Kill off any bonds created after unregistering bond rtnl ops */
4845         rtnl_lock();
4846         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4847                 unregister_netdevice_queue(bond->dev, &list);
4848         unregister_netdevice_many(&list);
4849         rtnl_unlock();
4850
4851         bond_destroy_proc_dir(bn);
4852 }
4853
4854 static struct pernet_operations bond_net_ops = {
4855         .init = bond_net_init,
4856         .exit = bond_net_exit,
4857         .id   = &bond_net_id,
4858         .size = sizeof(struct bond_net),
4859 };
4860
4861 static int __init bonding_init(void)
4862 {
4863         int i;
4864         int res;
4865
4866         pr_info("%s", bond_version);
4867
4868         res = bond_check_params(&bonding_defaults);
4869         if (res)
4870                 goto out;
4871
4872         res = register_pernet_subsys(&bond_net_ops);
4873         if (res)
4874                 goto out;
4875
4876         res = bond_netlink_init();
4877         if (res)
4878                 goto err_link;
4879
4880         bond_create_debugfs();
4881
4882         for (i = 0; i < max_bonds; i++) {
4883                 res = bond_create(&init_net, NULL);
4884                 if (res)
4885                         goto err;
4886         }
4887
4888         register_netdevice_notifier(&bond_netdev_notifier);
4889 out:
4890         return res;
4891 err:
4892         bond_destroy_debugfs();
4893         bond_netlink_fini();
4894 err_link:
4895         unregister_pernet_subsys(&bond_net_ops);
4896         goto out;
4897
4898 }
4899
4900 static void __exit bonding_exit(void)
4901 {
4902         unregister_netdevice_notifier(&bond_netdev_notifier);
4903
4904         bond_destroy_debugfs();
4905
4906         bond_netlink_fini();
4907         unregister_pernet_subsys(&bond_net_ops);
4908
4909 #ifdef CONFIG_NET_POLL_CONTROLLER
4910         /* Make sure we don't have an imbalance on our netpoll blocking */
4911         WARN_ON(atomic_read(&netpoll_block_tx));
4912 #endif
4913 }
4914
4915 module_init(bonding_init);
4916 module_exit(bonding_exit);
4917 MODULE_LICENSE("GPL");
4918 MODULE_VERSION(DRV_VERSION);
4919 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4920 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");