drivers/net/bonding/bond_main.c

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 #include <linux/kernel.h>
  35 #include <linux/module.h>
  36 #include <linux/types.h>
  37 #include <linux/fcntl.h>
  38 #include <linux/interrupt.h>
  39 #include <linux/ptrace.h>
  40 #include <linux/ioport.h>
  41 #include <linux/in.h>
  42 #include <net/ip.h>
  43 #include <linux/ip.h>
  44 #include <linux/tcp.h>
  45 #include <linux/udp.h>
  46 #include <linux/slab.h>
  47 #include <linux/string.h>
  48 #include <linux/init.h>
  49 #include <linux/timer.h>
  50 #include <linux/socket.h>
  51 #include <linux/ctype.h>
  52 #include <linux/inet.h>
  53 #include <linux/bitops.h>
  54 #include <linux/io.h>
  55 #include <asm/dma.h>
  56 #include <linux/uaccess.h>
  57 #include <linux/errno.h>
  58 #include <linux/netdevice.h>
  59 #include <linux/inetdevice.h>
  60 #include <linux/igmp.h>
  61 #include <linux/etherdevice.h>
  62 #include <linux/skbuff.h>
  63 #include <net/sock.h>
  64 #include <linux/rtnetlink.h>
  65 #include <linux/smp.h>
  66 #include <linux/if_ether.h>
  67 #include <net/arp.h>
  68 #include <linux/mii.h>
  69 #include <linux/ethtool.h>
  70 #include <linux/if_vlan.h>
  71 #include <linux/if_bonding.h>
  72 #include <linux/jiffies.h>
  73 #include <linux/preempt.h>
  74 #include <net/route.h>
  75 #include <net/net_namespace.h>
  76 #include <net/netns/generic.h>
  77 #include <net/pkt_sched.h>
  78 #include <linux/rculist.h>
  79 #include <net/flow_dissector.h>
  80 #include <net/switchdev.h>
  81 #include <net/bonding.h>
  82 #include <net/bond_3ad.h>
  83 #include <net/bond_alb.h>
  84
  85 #include "bonding_priv.h"
  86
  87 /*---------------------------- Module parameters ----------------------------*/
  88
  89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  90
  91 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  92 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  93 static int num_peer_notif = 1;
  94 static int miimon;
  95 static int updelay;
  96 static int downdelay;
  97 static int use_carrier  = 1;
  98 static char *mode;
  99 static char *primary;
 100 static char *primary_reselect;
 101 static char *lacp_rate;
 102 static int min_links;
 103 static char *ad_select;
 104 static char *xmit_hash_policy;
 105 static int arp_interval;
 106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 107 static char *arp_validate;
 108 static char *arp_all_targets;
 109 static char *fail_over_mac;
 110 static int all_slaves_active;
 111 static struct bond_params bonding_defaults;
 112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 113 static int packets_per_slave = 1;
 114 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
 115
 116 module_param(max_bonds, int, 0);
 117 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 118 module_param(tx_queues, int, 0);
 119 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 120 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 121 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 122                                "failover event (alias of num_unsol_na)");
 123 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 124 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 125                                "failover event (alias of num_grat_arp)");
 126 module_param(miimon, int, 0);
 127 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 128 module_param(updelay, int, 0);
 129 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 130 module_param(downdelay, int, 0);
 131 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 132                             "in milliseconds");
 133 module_param(use_carrier, int, 0);
 134 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 135                               "0 for off, 1 for on (default)");
 136 module_param(mode, charp, 0);
 137 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 138                        "1 for active-backup, 2 for balance-xor, "
 139                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 140                        "6 for balance-alb");
 141 module_param(primary, charp, 0);
 142 MODULE_PARM_DESC(primary, "Primary network device to use");
 143 module_param(primary_reselect, charp, 0);
 144 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 145                                    "once it comes up; "
 146                                    "0 for always (default), "
 147                                    "1 for only if speed of primary is "
 148                                    "better, "
 149                                    "2 for only on active slave "
 150                                    "failure");
 151 module_param(lacp_rate, charp, 0);
 152 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 153                             "0 for slow, 1 for fast");
 154 module_param(ad_select, charp, 0);
 155 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
 156                             "0 for stable (default), 1 for bandwidth, "
 157                             "2 for count");
 158 module_param(min_links, int, 0);
 159 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 160
 161 module_param(xmit_hash_policy, charp, 0);
 162 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
 163                                    "0 for layer 2 (default), 1 for layer 3+4, "
 164                                    "2 for layer 2+3, 3 for encap layer 2+3, "
 165                                    "4 for encap layer 3+4");
 166 module_param(arp_interval, int, 0);
 167 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 168 module_param_array(arp_ip_target, charp, NULL, 0);
 169 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 170 module_param(arp_validate, charp, 0);
 171 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 172                                "0 for none (default), 1 for active, "
 173                                "2 for backup, 3 for all");
 174 module_param(arp_all_targets, charp, 0);
 175 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 176 module_param(fail_over_mac, charp, 0);
 177 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 178                                 "the same MAC; 0 for none (default), "
 179                                 "1 for active, 2 for follow");
 180 module_param(all_slaves_active, int, 0);
 181 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
 182                                      "by setting active flag for all slaves; "
 183                                      "0 for never (default), 1 for always.");
 184 module_param(resend_igmp, int, 0);
 185 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 186                               "link failure");
 187 module_param(packets_per_slave, int, 0);
 188 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
 189                                     "mode; 0 for a random slave, 1 packet per "
 190                                     "slave (default), >1 packets per slave.");
 191 module_param(lp_interval, uint, 0);
 192 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
 193                               "the bonding driver sends learning packets to "
 194                               "each slaves peer switch. The default is 1.");
 195
 196 /*----------------------------- Global variables ----------------------------*/
 197
 198 #ifdef CONFIG_NET_POLL_CONTROLLER
 199 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 200 #endif
 201
 202 unsigned int bond_net_id __read_mostly;
 203
 204 /*-------------------------- Forward declarations ---------------------------*/
 205
 206 static int bond_init(struct net_device *bond_dev);
 207 static void bond_uninit(struct net_device *bond_dev);
 208 static void bond_get_stats(struct net_device *bond_dev,
 209                            struct rtnl_link_stats64 *stats);
 210 static void bond_slave_arr_handler(struct work_struct *work);
 211 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
 212                                   int mod);
 213 static void bond_netdev_notify_work(struct work_struct *work);
 214
 215 /*---------------------------- General routines -----------------------------*/
 216
 217 const char *bond_mode_name(int mode)
 218 {
 219         static const char *names[] = {
 220                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 221                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 222                 [BOND_MODE_XOR] = "load balancing (xor)",
 223                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 224                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 225                 [BOND_MODE_TLB] = "transmit load balancing",
 226                 [BOND_MODE_ALB] = "adaptive load balancing",
 227         };
 228
 229         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 230                 return "unknown";
 231
 232         return names[mode];
 233 }
 234
 235 /*---------------------------------- VLAN -----------------------------------*/
 236
 237 /**
 238  * bond_dev_queue_xmit - Prepare skb for xmit.
 239  *
 240  * @bond: bond device that got this skb for tx.
 241  * @skb: hw accel VLAN tagged skb to transmit
 242  * @slave_dev: slave that is supposed to xmit this skbuff
 243  */
 244 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 245                         struct net_device *slave_dev)
 246 {
 247         skb->dev = slave_dev;
 248
 249         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 250                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 251         skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
 252
 253         if (unlikely(netpoll_tx_running(bond->dev)))
 254                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 255         else
 256                 dev_queue_xmit(skb);
 257 }
 258
 259 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 260  * We don't protect the slave list iteration with a lock because:
 261  * a. This operation is performed in IOCTL context,
 262  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 263  * c. Holding a lock with BH disabled while directly calling a base driver
 264  *    entry point is generally a BAD idea.
 265  *
 266  * The design of synchronization/protection for this operation in the 8021q
 267  * module is good for one or more VLAN devices over a single physical device
 268  * and cannot be extended for a teaming solution like bonding, so there is a
 269  * potential race condition here where a net device from the vlan group might
 270  * be referenced (either by a base driver or the 8021q code) while it is being
 271  * removed from the system. However, it turns out we're not making matters
 272  * worse, and if it works for regular VLAN usage it will work here too.
 273 */
 274
 275 /**
 276  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 277  * @bond_dev: bonding net device that got called
 278  * @vid: vlan id being added
 279  */
 280 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 281                                 __be16 proto, u16 vid)
 282 {
 283         struct bonding *bond = netdev_priv(bond_dev);
 284         struct slave *slave, *rollback_slave;
 285         struct list_head *iter;
 286         int res;
 287
 288         bond_for_each_slave(bond, slave, iter) {
 289                 res = vlan_vid_add(slave->dev, proto, vid);
 290                 if (res)
 291                         goto unwind;
 292         }
 293
 294         return 0;
 295
 296 unwind:
 297         /* unwind to the slave that failed */
 298         bond_for_each_slave(bond, rollback_slave, iter) {
 299                 if (rollback_slave == slave)
 300                         break;
 301
 302                 vlan_vid_del(rollback_slave->dev, proto, vid);
 303         }
 304
 305         return res;
 306 }
 307
 308 /**
 309  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 310  * @bond_dev: bonding net device that got called
 311  * @vid: vlan id being removed
 312  */
 313 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 314                                  __be16 proto, u16 vid)
 315 {
 316         struct bonding *bond = netdev_priv(bond_dev);
 317         struct list_head *iter;
 318         struct slave *slave;
 319
 320         bond_for_each_slave(bond, slave, iter)
 321                 vlan_vid_del(slave->dev, proto, vid);
 322
 323         if (bond_is_lb(bond))
 324                 bond_alb_clear_vlan(bond, vid);
 325
 326         return 0;
 327 }
 328
 329 /*------------------------------- Link status -------------------------------*/
 330
 331 /* Set the carrier state for the master according to the state of its
 332  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 333  * do special 802.3ad magic.
 334  *
 335  * Returns zero if carrier state does not change, nonzero if it does.
 336  */
 337 int bond_set_carrier(struct bonding *bond)
 338 {
 339         struct list_head *iter;
 340         struct slave *slave;
 341
 342         if (!bond_has_slaves(bond))
 343                 goto down;
 344
 345         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 346                 return bond_3ad_set_carrier(bond);
 347
 348         bond_for_each_slave(bond, slave, iter) {
 349                 if (slave->link == BOND_LINK_UP) {
 350                         if (!netif_carrier_ok(bond->dev)) {
 351                                 netif_carrier_on(bond->dev);
 352                                 return 1;
 353                         }
 354                         return 0;
 355                 }
 356         }
 357
 358 down:
 359         if (netif_carrier_ok(bond->dev)) {
 360                 netif_carrier_off(bond->dev);
 361                 return 1;
 362         }
 363         return 0;
 364 }
 365
 366 /* Get link speed and duplex from the slave's base driver
 367  * using ethtool. If for some reason the call fails or the
 368  * values are invalid, set speed and duplex to -1,
 369  * and return. Return 1 if speed or duplex settings are
 370  * UNKNOWN; 0 otherwise.
 371  */
 372 static int bond_update_speed_duplex(struct slave *slave)
 373 {
 374         struct net_device *slave_dev = slave->dev;
 375         struct ethtool_link_ksettings ecmd;
 376         int res;
 377
 378         slave->speed = SPEED_UNKNOWN;
 379         slave->duplex = DUPLEX_UNKNOWN;
 380
 381         res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
 382         if (res < 0)
 383                 return 1;
 384         if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
 385                 return 1;
 386         switch (ecmd.base.duplex) {
 387         case DUPLEX_FULL:
 388         case DUPLEX_HALF:
 389                 break;
 390         default:
 391                 return 1;
 392         }
 393
 394         slave->speed = ecmd.base.speed;
 395         slave->duplex = ecmd.base.duplex;
 396
 397         return 0;
 398 }
 399
 400 const char *bond_slave_link_status(s8 link)
 401 {
 402         switch (link) {
 403         case BOND_LINK_UP:
 404                 return "up";
 405         case BOND_LINK_FAIL:
 406                 return "going down";
 407         case BOND_LINK_DOWN:
 408                 return "down";
 409         case BOND_LINK_BACK:
 410                 return "going back";
 411         default:
 412                 return "unknown";
 413         }
 414 }
 415
 416 /* if <dev> supports MII link status reporting, check its link status.
 417  *
 418  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 419  * depending upon the setting of the use_carrier parameter.
 420  *
 421  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 422  * can't tell and just pretend it is), or 0, meaning that the link is
 423  * down.
 424  *
 425  * If reporting is non-zero, instead of faking link up, return -1 if
 426  * both ETHTOOL and MII ioctls fail (meaning the device does not
 427  * support them).  If use_carrier is set, return whatever it says.
 428  * It'd be nice if there was a good way to tell if a driver supports
 429  * netif_carrier, but there really isn't.
 430  */
 431 static int bond_check_dev_link(struct bonding *bond,
 432                                struct net_device *slave_dev, int reporting)
 433 {
 434         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 435         int (*ioctl)(struct net_device *, struct ifreq *, int);
 436         struct ifreq ifr;
 437         struct mii_ioctl_data *mii;
 438
 439         if (!reporting && !netif_running(slave_dev))
 440                 return 0;
 441
 442         if (bond->params.use_carrier)
 443                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 444
 445         /* Try to get link status using Ethtool first. */
 446         if (slave_dev->ethtool_ops->get_link)
 447                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 448                         BMSR_LSTATUS : 0;
 449
 450         /* Ethtool can't be used, fallback to MII ioctls. */
 451         ioctl = slave_ops->ndo_do_ioctl;
 452         if (ioctl) {
 453                 /* TODO: set pointer to correct ioctl on a per team member
 454                  *       bases to make this more efficient. that is, once
 455                  *       we determine the correct ioctl, we will always
 456                  *       call it and not the others for that team
 457                  *       member.
 458                  */
 459
 460                 /* We cannot assume that SIOCGMIIPHY will also read a
 461                  * register; not all network drivers (e.g., e100)
 462                  * support that.
 463                  */
 464
 465                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 466                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 467                 mii = if_mii(&ifr);
 468                 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 469                         mii->reg_num = MII_BMSR;
 470                         if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
 471                                 return mii->val_out & BMSR_LSTATUS;
 472                 }
 473         }
 474
 475         /* If reporting, report that either there's no dev->do_ioctl,
 476          * or both SIOCGMIIREG and get_link failed (meaning that we
 477          * cannot report link status).  If not reporting, pretend
 478          * we're ok.
 479          */
 480         return reporting ? -1 : BMSR_LSTATUS;
 481 }
 482
 483 /*----------------------------- Multicast list ------------------------------*/
 484
 485 /* Push the promiscuity flag down to appropriate slaves */
 486 static int bond_set_promiscuity(struct bonding *bond, int inc)
 487 {
 488         struct list_head *iter;
 489         int err = 0;
 490
 491         if (bond_uses_primary(bond)) {
 492                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 493
 494                 if (curr_active)
 495                         err = dev_set_promiscuity(curr_active->dev, inc);
 496         } else {
 497                 struct slave *slave;
 498
 499                 bond_for_each_slave(bond, slave, iter) {
 500                         err = dev_set_promiscuity(slave->dev, inc);
 501                         if (err)
 502                                 return err;
 503                 }
 504         }
 505         return err;
 506 }
 507
 508 /* Push the allmulti flag down to all slaves */
 509 static int bond_set_allmulti(struct bonding *bond, int inc)
 510 {
 511         struct list_head *iter;
 512         int err = 0;
 513
 514         if (bond_uses_primary(bond)) {
 515                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 516
 517                 if (curr_active)
 518                         err = dev_set_allmulti(curr_active->dev, inc);
 519         } else {
 520                 struct slave *slave;
 521
 522                 bond_for_each_slave(bond, slave, iter) {
 523                         err = dev_set_allmulti(slave->dev, inc);
 524                         if (err)
 525                                 return err;
 526                 }
 527         }
 528         return err;
 529 }
 530
 531 /* Retrieve the list of registered multicast addresses for the bonding
 532  * device and retransmit an IGMP JOIN request to the current active
 533  * slave.
 534  */
 535 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 536 {
 537         struct bonding *bond = container_of(work, struct bonding,
 538                                             mcast_work.work);
 539
 540         if (!rtnl_trylock()) {
 541                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 542                 return;
 543         }
 544         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 545
 546         if (bond->igmp_retrans > 1) {
 547                 bond->igmp_retrans--;
 548                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 549         }
 550         rtnl_unlock();
 551 }
 552
 553 /* Flush bond's hardware addresses from slave */
 554 static void bond_hw_addr_flush(struct net_device *bond_dev,
 555                                struct net_device *slave_dev)
 556 {
 557         struct bonding *bond = netdev_priv(bond_dev);
 558
 559         dev_uc_unsync(slave_dev, bond_dev);
 560         dev_mc_unsync(slave_dev, bond_dev);
 561
 562         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 563                 /* del lacpdu mc addr from mc list */
 564                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 565
 566                 dev_mc_del(slave_dev, lacpdu_multicast);
 567         }
 568 }
 569
 570 /*--------------------------- Active slave change ---------------------------*/
 571
 572 /* Update the hardware address list and promisc/allmulti for the new and
 573  * old active slaves (if any).  Modes that are not using primary keep all
 574  * slaves up date at all times; only the modes that use primary need to call
 575  * this function to swap these settings during a failover.
 576  */
 577 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 578                               struct slave *old_active)
 579 {
 580         if (old_active) {
 581                 if (bond->dev->flags & IFF_PROMISC)
 582                         dev_set_promiscuity(old_active->dev, -1);
 583
 584                 if (bond->dev->flags & IFF_ALLMULTI)
 585                         dev_set_allmulti(old_active->dev, -1);
 586
 587                 bond_hw_addr_flush(bond->dev, old_active->dev);
 588         }
 589
 590         if (new_active) {
 591                 /* FIXME: Signal errors upstream. */
 592                 if (bond->dev->flags & IFF_PROMISC)
 593                         dev_set_promiscuity(new_active->dev, 1);
 594
 595                 if (bond->dev->flags & IFF_ALLMULTI)
 596                         dev_set_allmulti(new_active->dev, 1);
 597
 598                 netif_addr_lock_bh(bond->dev);
 599                 dev_uc_sync(new_active->dev, bond->dev);
 600                 dev_mc_sync(new_active->dev, bond->dev);
 601                 netif_addr_unlock_bh(bond->dev);
 602         }
 603 }
 604
 605 /**
 606  * bond_set_dev_addr - clone slave's address to bond
 607  * @bond_dev: bond net device
 608  * @slave_dev: slave net device
 609  *
 610  * Should be called with RTNL held.
 611  */
 612 static void bond_set_dev_addr(struct net_device *bond_dev,
 613                               struct net_device *slave_dev)
 614 {
 615         netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n",
 616                    bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len);
 617         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 618         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 619         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 620 }
 621
 622 static struct slave *bond_get_old_active(struct bonding *bond,
 623                                          struct slave *new_active)
 624 {
 625         struct slave *slave;
 626         struct list_head *iter;
 627
 628         bond_for_each_slave(bond, slave, iter) {
 629                 if (slave == new_active)
 630                         continue;
 631
 632                 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
 633                         return slave;
 634         }
 635
 636         return NULL;
 637 }
 638
 639 /* bond_do_fail_over_mac
 640  *
 641  * Perform special MAC address swapping for fail_over_mac settings
 642  *
 643  * Called with RTNL
 644  */
 645 static void bond_do_fail_over_mac(struct bonding *bond,
 646                                   struct slave *new_active,
 647                                   struct slave *old_active)
 648 {
 649         u8 tmp_mac[MAX_ADDR_LEN];
 650         struct sockaddr_storage ss;
 651         int rv;
 652
 653         switch (bond->params.fail_over_mac) {
 654         case BOND_FOM_ACTIVE:
 655                 if (new_active)
 656                         bond_set_dev_addr(bond->dev, new_active->dev);
 657                 break;
 658         case BOND_FOM_FOLLOW:
 659                 /* if new_active && old_active, swap them
 660                  * if just old_active, do nothing (going to no active slave)
 661                  * if just new_active, set new_active to bond's MAC
 662                  */
 663                 if (!new_active)
 664                         return;
 665
 666                 if (!old_active)
 667                         old_active = bond_get_old_active(bond, new_active);
 668
 669                 if (old_active) {
 670                         bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
 671                                           new_active->dev->addr_len);
 672                         bond_hw_addr_copy(ss.__data,
 673                                           old_active->dev->dev_addr,
 674                                           old_active->dev->addr_len);
 675                         ss.ss_family = new_active->dev->type;
 676                 } else {
 677                         bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
 678                                           bond->dev->addr_len);
 679                         ss.ss_family = bond->dev->type;
 680                 }
 681
 682                 rv = dev_set_mac_address(new_active->dev,
 683                                          (struct sockaddr *)&ss);
 684                 if (rv) {
 685                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 686                                    -rv, new_active->dev->name);
 687                         goto out;
 688                 }
 689
 690                 if (!old_active)
 691                         goto out;
 692
 693                 bond_hw_addr_copy(ss.__data, tmp_mac,
 694                                   new_active->dev->addr_len);
 695                 ss.ss_family = old_active->dev->type;
 696
 697                 rv = dev_set_mac_address(old_active->dev,
 698                                          (struct sockaddr *)&ss);
 699                 if (rv)
 700                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 701                                    -rv, new_active->dev->name);
 702 out:
 703                 break;
 704         default:
 705                 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
 706                            bond->params.fail_over_mac);
 707                 break;
 708         }
 709
 710 }
 711
 712 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
 713 {
 714         struct slave *prim = rtnl_dereference(bond->primary_slave);
 715         struct slave *curr = rtnl_dereference(bond->curr_active_slave);
 716
 717         if (!prim || prim->link != BOND_LINK_UP) {
 718                 if (!curr || curr->link != BOND_LINK_UP)
 719                         return NULL;
 720                 return curr;
 721         }
 722
 723         if (bond->force_primary) {
 724                 bond->force_primary = false;
 725                 return prim;
 726         }
 727
 728         if (!curr || curr->link != BOND_LINK_UP)
 729                 return prim;
 730
 731         /* At this point, prim and curr are both up */
 732         switch (bond->params.primary_reselect) {
 733         case BOND_PRI_RESELECT_ALWAYS:
 734                 return prim;
 735         case BOND_PRI_RESELECT_BETTER:
 736                 if (prim->speed < curr->speed)
 737                         return curr;
 738                 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
 739                         return curr;
 740                 return prim;
 741         case BOND_PRI_RESELECT_FAILURE:
 742                 return curr;
 743         default:
 744                 netdev_err(bond->dev, "impossible primary_reselect %d\n",
 745                            bond->params.primary_reselect);
 746                 return curr;
 747         }
 748 }
 749
 750 /**
 751  * bond_find_best_slave - select the best available slave to be the active one
 752  * @bond: our bonding struct
 753  */
 754 static struct slave *bond_find_best_slave(struct bonding *bond)
 755 {
 756         struct slave *slave, *bestslave = NULL;
 757         struct list_head *iter;
 758         int mintime = bond->params.updelay;
 759
 760         slave = bond_choose_primary_or_current(bond);
 761         if (slave)
 762                 return slave;
 763
 764         bond_for_each_slave(bond, slave, iter) {
 765                 if (slave->link == BOND_LINK_UP)
 766                         return slave;
 767                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
 768                     slave->delay < mintime) {
 769                         mintime = slave->delay;
 770                         bestslave = slave;
 771                 }
 772         }
 773
 774         return bestslave;
 775 }
 776
 777 static bool bond_should_notify_peers(struct bonding *bond)
 778 {
 779         struct slave *slave;
 780
 781         rcu_read_lock();
 782         slave = rcu_dereference(bond->curr_active_slave);
 783         rcu_read_unlock();
 784
 785         if (!slave || !bond->send_peer_notif ||
 786             !netif_carrier_ok(bond->dev) ||
 787             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 788                 return false;
 789
 790         netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
 791                    slave ? slave->dev->name : "NULL");
 792
 793         return true;
 794 }
 795
 796 /**
 797  * change_active_interface - change the active slave into the specified one
 798  * @bond: our bonding struct
 799  * @new: the new slave to make the active one
 800  *
 801  * Set the new slave to the bond's settings and unset them on the old
 802  * curr_active_slave.
 803  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 804  *
 805  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 806  * because it is apparently the best available slave we have, even though its
 807  * updelay hasn't timed out yet.
 808  *
 809  * Caller must hold RTNL.
 810  */
 811 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 812 {
 813         struct slave *old_active;
 814
 815         ASSERT_RTNL();
 816
 817         old_active = rtnl_dereference(bond->curr_active_slave);
 818
 819         if (old_active == new_active)
 820                 return;
 821
 822         if (new_active) {
 823                 new_active->last_link_up = jiffies;
 824
 825                 if (new_active->link == BOND_LINK_BACK) {
 826                         if (bond_uses_primary(bond)) {
 827                                 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
 828                                             new_active->dev->name,
 829                                             (bond->params.updelay - new_active->delay) * bond->params.miimon);
 830                         }
 831
 832                         new_active->delay = 0;
 833                         bond_set_slave_link_state(new_active, BOND_LINK_UP,
 834                                                   BOND_SLAVE_NOTIFY_NOW);
 835
 836                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 837                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 838
 839                         if (bond_is_lb(bond))
 840                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 841                 } else {
 842                         if (bond_uses_primary(bond)) {
 843                                 netdev_info(bond->dev, "making interface %s the new active one\n",
 844                                             new_active->dev->name);
 845                         }
 846                 }
 847         }
 848
 849         if (bond_uses_primary(bond))
 850                 bond_hw_addr_swap(bond, new_active, old_active);
 851
 852         if (bond_is_lb(bond)) {
 853                 bond_alb_handle_active_change(bond, new_active);
 854                 if (old_active)
 855                         bond_set_slave_inactive_flags(old_active,
 856                                                       BOND_SLAVE_NOTIFY_NOW);
 857                 if (new_active)
 858                         bond_set_slave_active_flags(new_active,
 859                                                     BOND_SLAVE_NOTIFY_NOW);
 860         } else {
 861                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 862         }
 863
 864         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
 865                 if (old_active)
 866                         bond_set_slave_inactive_flags(old_active,
 867                                                       BOND_SLAVE_NOTIFY_NOW);
 868
 869                 if (new_active) {
 870                         bool should_notify_peers = false;
 871
 872                         bond_set_slave_active_flags(new_active,
 873                                                     BOND_SLAVE_NOTIFY_NOW);
 874
 875                         if (bond->params.fail_over_mac)
 876                                 bond_do_fail_over_mac(bond, new_active,
 877                                                       old_active);
 878
 879                         if (netif_running(bond->dev)) {
 880                                 bond->send_peer_notif =
 881                                         bond->params.num_peer_notif;
 882                                 should_notify_peers =
 883                                         bond_should_notify_peers(bond);
 884                         }
 885
 886                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 887                         if (should_notify_peers)
 888                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 889                                                          bond->dev);
 890                 }
 891         }
 892
 893         /* resend IGMP joins since active slave has changed or
 894          * all were sent on curr_active_slave.
 895          * resend only if bond is brought up with the affected
 896          * bonding modes and the retransmission is enabled
 897          */
 898         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 899             ((bond_uses_primary(bond) && new_active) ||
 900              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
 901                 bond->igmp_retrans = bond->params.resend_igmp;
 902                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 903         }
 904 }
 905
 906 /**
 907  * bond_select_active_slave - select a new active slave, if needed
 908  * @bond: our bonding struct
 909  *
 910  * This functions should be called when one of the following occurs:
 911  * - The old curr_active_slave has been released or lost its link.
 912  * - The primary_slave has got its link back.
 913  * - A slave has got its link back and there's no old curr_active_slave.
 914  *
 915  * Caller must hold RTNL.
 916  */
 917 void bond_select_active_slave(struct bonding *bond)
 918 {
 919         struct slave *best_slave;
 920         int rv;
 921
 922         ASSERT_RTNL();
 923
 924         best_slave = bond_find_best_slave(bond);
 925         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
 926                 bond_change_active_slave(bond, best_slave);
 927                 rv = bond_set_carrier(bond);
 928                 if (!rv)
 929                         return;
 930
 931                 if (netif_carrier_ok(bond->dev))
 932                         netdev_info(bond->dev, "first active interface up!\n");
 933                 else
 934                         netdev_info(bond->dev, "now running without any active interface!\n");
 935         }
 936 }
 937
 938 #ifdef CONFIG_NET_POLL_CONTROLLER
 939 static inline int slave_enable_netpoll(struct slave *slave)
 940 {
 941         struct netpoll *np;
 942         int err = 0;
 943
 944         np = kzalloc(sizeof(*np), GFP_KERNEL);
 945         err = -ENOMEM;
 946         if (!np)
 947                 goto out;
 948
 949         err = __netpoll_setup(np, slave->dev);
 950         if (err) {
 951                 kfree(np);
 952                 goto out;
 953         }
 954         slave->np = np;
 955 out:
 956         return err;
 957 }
 958 static inline void slave_disable_netpoll(struct slave *slave)
 959 {
 960         struct netpoll *np = slave->np;
 961
 962         if (!np)
 963                 return;
 964
 965         slave->np = NULL;
 966         __netpoll_free_async(np);
 967 }
 968
 969 static void bond_poll_controller(struct net_device *bond_dev)
 970 {
 971         struct bonding *bond = netdev_priv(bond_dev);
 972         struct slave *slave = NULL;
 973         struct list_head *iter;
 974         struct ad_info ad_info;
 975
 976         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 977                 if (bond_3ad_get_active_agg_info(bond, &ad_info))
 978                         return;
 979
 980         bond_for_each_slave_rcu(bond, slave, iter) {
 981                 if (!bond_slave_is_up(slave))
 982                         continue;
 983
 984                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 985                         struct aggregator *agg =
 986                             SLAVE_AD_INFO(slave)->port.aggregator;
 987
 988                         if (agg &&
 989                             agg->aggregator_identifier != ad_info.aggregator_id)
 990                                 continue;
 991                 }
 992
 993                 netpoll_poll_dev(slave->dev);
 994         }
 995 }
 996
 997 static void bond_netpoll_cleanup(struct net_device *bond_dev)
 998 {
 999         struct bonding *bond = netdev_priv(bond_dev);
1000         struct list_head *iter;
1001         struct slave *slave;
1002
1003         bond_for_each_slave(bond, slave, iter)
1004                 if (bond_slave_is_up(slave))
1005                         slave_disable_netpoll(slave);
1006 }
1007
1008 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1009 {
1010         struct bonding *bond = netdev_priv(dev);
1011         struct list_head *iter;
1012         struct slave *slave;
1013         int err = 0;
1014
1015         bond_for_each_slave(bond, slave, iter) {
1016                 err = slave_enable_netpoll(slave);
1017                 if (err) {
1018                         bond_netpoll_cleanup(dev);
1019                         break;
1020                 }
1021         }
1022         return err;
1023 }
1024 #else
1025 static inline int slave_enable_netpoll(struct slave *slave)
1026 {
1027         return 0;
1028 }
1029 static inline void slave_disable_netpoll(struct slave *slave)
1030 {
1031 }
1032 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1033 {
1034 }
1035 #endif
1036
1037 /*---------------------------------- IOCTL ----------------------------------*/
1038
1039 static netdev_features_t bond_fix_features(struct net_device *dev,
1040                                            netdev_features_t features)
1041 {
1042         struct bonding *bond = netdev_priv(dev);
1043         struct list_head *iter;
1044         netdev_features_t mask;
1045         struct slave *slave;
1046
1047         mask = features;
1048
1049         features &= ~NETIF_F_ONE_FOR_ALL;
1050         features |= NETIF_F_ALL_FOR_ALL;
1051
1052         bond_for_each_slave(bond, slave, iter) {
1053                 features = netdev_increment_features(features,
1054                                                      slave->dev->features,
1055                                                      mask);
1056         }
1057         features = netdev_add_tso_features(features, mask);
1058
1059         return features;
1060 }
1061
1062 #define BOND_VLAN_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1063                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1064                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1065
1066 #define BOND_ENC_FEATURES       (NETIF_F_HW_CSUM | NETIF_F_SG | \
1067                                  NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1068
1069 static void bond_compute_features(struct bonding *bond)
1070 {
1071         unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1072                                         IFF_XMIT_DST_RELEASE_PERM;
1073         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1074         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1075         struct net_device *bond_dev = bond->dev;
1076         struct list_head *iter;
1077         struct slave *slave;
1078         unsigned short max_hard_header_len = ETH_HLEN;
1079         unsigned int gso_max_size = GSO_MAX_SIZE;
1080         u16 gso_max_segs = GSO_MAX_SEGS;
1081
1082         if (!bond_has_slaves(bond))
1083                 goto done;
1084         vlan_features &= NETIF_F_ALL_FOR_ALL;
1085
1086         bond_for_each_slave(bond, slave, iter) {
1087                 vlan_features = netdev_increment_features(vlan_features,
1088                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1089
1090                 enc_features = netdev_increment_features(enc_features,
1091                                                          slave->dev->hw_enc_features,
1092                                                          BOND_ENC_FEATURES);
1093                 dst_release_flag &= slave->dev->priv_flags;
1094                 if (slave->dev->hard_header_len > max_hard_header_len)
1095                         max_hard_header_len = slave->dev->hard_header_len;
1096
1097                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1098                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1099         }
1100         bond_dev->hard_header_len = max_hard_header_len;
1101
1102 done:
1103         bond_dev->vlan_features = vlan_features;
1104         bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1105                                     NETIF_F_HW_VLAN_CTAG_TX |
1106                                     NETIF_F_HW_VLAN_STAG_TX |
1107                                     NETIF_F_GSO_UDP_L4;
1108         bond_dev->gso_max_segs = gso_max_segs;
1109         netif_set_gso_max_size(bond_dev, gso_max_size);
1110
1111         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1112         if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1113             dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1114                 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1115
1116         netdev_change_features(bond_dev);
1117 }
1118
1119 static void bond_setup_by_slave(struct net_device *bond_dev,
1120                                 struct net_device *slave_dev)
1121 {
1122         bond_dev->header_ops        = slave_dev->header_ops;
1123
1124         bond_dev->type              = slave_dev->type;
1125         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1126         bond_dev->needed_headroom   = slave_dev->needed_headroom;
1127         bond_dev->addr_len          = slave_dev->addr_len;
1128
1129         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1130                 slave_dev->addr_len);
1131
1132         if (slave_dev->flags & IFF_POINTOPOINT) {
1133                 bond_dev->flags &= ~(IFF_BROADCAST | IFF_MULTICAST);
1134                 bond_dev->flags |= (IFF_POINTOPOINT | IFF_NOARP);
1135         }
1136 }
1137
1138 /* On bonding slaves other than the currently active slave, suppress
1139  * duplicates except for alb non-mcast/bcast.
1140  */
1141 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1142                                             struct slave *slave,
1143                                             struct bonding *bond)
1144 {
1145         if (bond_is_slave_inactive(slave)) {
1146                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1147                     skb->pkt_type != PACKET_BROADCAST &&
1148                     skb->pkt_type != PACKET_MULTICAST)
1149                         return false;
1150                 return true;
1151         }
1152         return false;
1153 }
1154
1155 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1156 {
1157         struct sk_buff *skb = *pskb;
1158         struct slave *slave;
1159         struct bonding *bond;
1160         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1161                           struct slave *);
1162         int ret = RX_HANDLER_ANOTHER;
1163
1164         skb = skb_share_check(skb, GFP_ATOMIC);
1165         if (unlikely(!skb))
1166                 return RX_HANDLER_CONSUMED;
1167
1168         *pskb = skb;
1169
1170         slave = bond_slave_get_rcu(skb->dev);
1171         bond = slave->bond;
1172
1173         recv_probe = READ_ONCE(bond->recv_probe);
1174         if (recv_probe) {
1175                 ret = recv_probe(skb, bond, slave);
1176                 if (ret == RX_HANDLER_CONSUMED) {
1177                         consume_skb(skb);
1178                         return ret;
1179                 }
1180         }
1181
1182         /*
1183          * For packets determined by bond_should_deliver_exact_match() call to
1184          * be suppressed we want to make an exception for link-local packets.
1185          * This is necessary for e.g. LLDP daemons to be able to monitor
1186          * inactive slave links without being forced to bind to them
1187          * explicitly.
1188          *
1189          * At the same time, packets that are passed to the bonding master
1190          * (including link-local ones) can have their originating interface
1191          * determined via PACKET_ORIGDEV socket option.
1192          */
1193         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1194                 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1195                         return RX_HANDLER_PASS;
1196                 return RX_HANDLER_EXACT;
1197         }
1198
1199         skb->dev = bond->dev;
1200
1201         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1202             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1203             skb->pkt_type == PACKET_HOST) {
1204
1205                 if (unlikely(skb_cow_head(skb,
1206                                           skb->data - skb_mac_header(skb)))) {
1207                         kfree_skb(skb);
1208                         return RX_HANDLER_CONSUMED;
1209                 }
1210                 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1211                                   bond->dev->addr_len);
1212         }
1213
1214         return ret;
1215 }
1216
1217 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1218 {
1219         switch (BOND_MODE(bond)) {
1220         case BOND_MODE_ROUNDROBIN:
1221                 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1222         case BOND_MODE_ACTIVEBACKUP:
1223                 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1224         case BOND_MODE_BROADCAST:
1225                 return NETDEV_LAG_TX_TYPE_BROADCAST;
1226         case BOND_MODE_XOR:
1227         case BOND_MODE_8023AD:
1228                 return NETDEV_LAG_TX_TYPE_HASH;
1229         default:
1230                 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1231         }
1232 }
1233
1234 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1235                                                enum netdev_lag_tx_type type)
1236 {
1237         if (type != NETDEV_LAG_TX_TYPE_HASH)
1238                 return NETDEV_LAG_HASH_NONE;
1239
1240         switch (bond->params.xmit_policy) {
1241         case BOND_XMIT_POLICY_LAYER2:
1242                 return NETDEV_LAG_HASH_L2;
1243         case BOND_XMIT_POLICY_LAYER34:
1244                 return NETDEV_LAG_HASH_L34;
1245         case BOND_XMIT_POLICY_LAYER23:
1246                 return NETDEV_LAG_HASH_L23;
1247         case BOND_XMIT_POLICY_ENCAP23:
1248                 return NETDEV_LAG_HASH_E23;
1249         case BOND_XMIT_POLICY_ENCAP34:
1250                 return NETDEV_LAG_HASH_E34;
1251         default:
1252                 return NETDEV_LAG_HASH_UNKNOWN;
1253         }
1254 }
1255
1256 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1257                                       struct netlink_ext_ack *extack)
1258 {
1259         struct netdev_lag_upper_info lag_upper_info;
1260         enum netdev_lag_tx_type type;
1261
1262         type = bond_lag_tx_type(bond);
1263         lag_upper_info.tx_type = type;
1264         lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1265
1266         return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1267                                             &lag_upper_info, extack);
1268 }
1269
1270 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1271 {
1272         netdev_upper_dev_unlink(slave->dev, bond->dev);
1273         slave->dev->flags &= ~IFF_SLAVE;
1274 }
1275
1276 static void slave_kobj_release(struct kobject *kobj)
1277 {
1278         struct slave *slave = to_slave(kobj);
1279         struct bonding *bond = bond_get_bond_by_slave(slave);
1280
1281         cancel_delayed_work_sync(&slave->notify_work);
1282         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1283                 kfree(SLAVE_AD_INFO(slave));
1284
1285         kfree(slave);
1286 }
1287
1288 static struct kobj_type slave_ktype = {
1289         .release = slave_kobj_release,
1290 #ifdef CONFIG_SYSFS
1291         .sysfs_ops = &slave_sysfs_ops,
1292 #endif
1293 };
1294
1295 static int bond_kobj_init(struct slave *slave)
1296 {
1297         int err;
1298
1299         err = kobject_init_and_add(&slave->kobj, &slave_ktype,
1300                                    &(slave->dev->dev.kobj), "bonding_slave");
1301         if (err)
1302                 kobject_put(&slave->kobj);
1303
1304         return err;
1305 }
1306
1307 static struct slave *bond_alloc_slave(struct bonding *bond,
1308                                       struct net_device *slave_dev)
1309 {
1310         struct slave *slave = NULL;
1311
1312         slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1313         if (!slave)
1314                 return NULL;
1315
1316         slave->bond = bond;
1317         slave->dev = slave_dev;
1318         INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1319
1320         if (bond_kobj_init(slave))
1321                 return NULL;
1322
1323         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1324                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1325                                                GFP_KERNEL);
1326                 if (!SLAVE_AD_INFO(slave)) {
1327                         kobject_put(&slave->kobj);
1328                         return NULL;
1329                 }
1330         }
1331
1332         return slave;
1333 }
1334
1335 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1336 {
1337         info->bond_mode = BOND_MODE(bond);
1338         info->miimon = bond->params.miimon;
1339         info->num_slaves = bond->slave_cnt;
1340 }
1341
1342 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1343 {
1344         strcpy(info->slave_name, slave->dev->name);
1345         info->link = slave->link;
1346         info->state = bond_slave_state(slave);
1347         info->link_failure_count = slave->link_failure_count;
1348 }
1349
1350 static void bond_netdev_notify_work(struct work_struct *_work)
1351 {
1352         struct slave *slave = container_of(_work, struct slave,
1353                                            notify_work.work);
1354
1355         if (rtnl_trylock()) {
1356                 struct netdev_bonding_info binfo;
1357
1358                 bond_fill_ifslave(slave, &binfo.slave);
1359                 bond_fill_ifbond(slave->bond, &binfo.master);
1360                 netdev_bonding_info_change(slave->dev, &binfo);
1361                 rtnl_unlock();
1362         } else {
1363                 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1364         }
1365 }
1366
1367 void bond_queue_slave_event(struct slave *slave)
1368 {
1369         queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1370 }
1371
1372 void bond_lower_state_changed(struct slave *slave)
1373 {
1374         struct netdev_lag_lower_state_info info;
1375
1376         info.link_up = slave->link == BOND_LINK_UP ||
1377                        slave->link == BOND_LINK_FAIL;
1378         info.tx_enabled = bond_is_active_slave(slave);
1379         netdev_lower_state_changed(slave->dev, &info);
1380 }
1381
1382 /* enslave device <slave> to bond device <master> */
1383 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1384                  struct netlink_ext_ack *extack)
1385 {
1386         struct bonding *bond = netdev_priv(bond_dev);
1387         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1388         struct slave *new_slave = NULL, *prev_slave;
1389         struct sockaddr_storage ss;
1390         int link_reporting;
1391         int res = 0, i;
1392
1393         if (!bond->params.use_carrier &&
1394             slave_dev->ethtool_ops->get_link == NULL &&
1395             slave_ops->ndo_do_ioctl == NULL) {
1396                 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1397                             slave_dev->name);
1398         }
1399
1400         /* already in-use? */
1401         if (netdev_is_rx_handler_busy(slave_dev)) {
1402                 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1403                 netdev_err(bond_dev,
1404                            "Error: Device is in use and cannot be enslaved\n");
1405                 return -EBUSY;
1406         }
1407
1408         if (bond_dev == slave_dev) {
1409                 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1410                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1411                 return -EPERM;
1412         }
1413
1414         /* vlan challenged mutual exclusion */
1415         /* no need to lock since we're protected by rtnl_lock */
1416         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1417                 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1418                            slave_dev->name);
1419                 if (vlan_uses_dev(bond_dev)) {
1420                         NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1421                         netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1422                                    slave_dev->name, bond_dev->name);
1423                         return -EPERM;
1424                 } else {
1425                         netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1426                                     slave_dev->name, slave_dev->name,
1427                                     bond_dev->name);
1428                 }
1429         } else {
1430                 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1431                            slave_dev->name);
1432         }
1433
1434         /* Old ifenslave binaries are no longer supported.  These can
1435          * be identified with moderate accuracy by the state of the slave:
1436          * the current ifenslave will set the interface down prior to
1437          * enslaving it; the old ifenslave will not.
1438          */
1439         if (slave_dev->flags & IFF_UP) {
1440                 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
1441                 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1442                            slave_dev->name);
1443                 return -EPERM;
1444         }
1445
1446         /* set bonding device ether type by slave - bonding netdevices are
1447          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1448          * there is a need to override some of the type dependent attribs/funcs.
1449          *
1450          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1451          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1452          */
1453         if (!bond_has_slaves(bond)) {
1454                 if (bond_dev->type != slave_dev->type) {
1455                         netdev_dbg(bond_dev, "change device type from %d to %d\n",
1456                                    bond_dev->type, slave_dev->type);
1457
1458                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1459                                                        bond_dev);
1460                         res = notifier_to_errno(res);
1461                         if (res) {
1462                                 netdev_err(bond_dev, "refused to change device type\n");
1463                                 return -EBUSY;
1464                         }
1465
1466                         /* Flush unicast and multicast addresses */
1467                         dev_uc_flush(bond_dev);
1468                         dev_mc_flush(bond_dev);
1469
1470                         if (slave_dev->type != ARPHRD_ETHER)
1471                                 bond_setup_by_slave(bond_dev, slave_dev);
1472                         else {
1473                                 ether_setup(bond_dev);
1474                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1475                         }
1476
1477                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1478                                                  bond_dev);
1479                 }
1480         } else if (bond_dev->type != slave_dev->type) {
1481                 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1482                 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1483                            slave_dev->name, slave_dev->type, bond_dev->type);
1484                 return -EINVAL;
1485         }
1486
1487         if (slave_dev->type == ARPHRD_INFINIBAND &&
1488             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1489                 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1490                 netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n",
1491                             slave_dev->type);
1492                 res = -EOPNOTSUPP;
1493                 goto err_undo_flags;
1494         }
1495
1496         if (!slave_ops->ndo_set_mac_address ||
1497             slave_dev->type == ARPHRD_INFINIBAND) {
1498                 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1499                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1500                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1501                         if (!bond_has_slaves(bond)) {
1502                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1503                                 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1504                         } else {
1505                                 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1506                                 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");
1507                                 res = -EOPNOTSUPP;
1508                                 goto err_undo_flags;
1509                         }
1510                 }
1511         }
1512
1513         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1514
1515         /* If this is the first slave, then we need to set the master's hardware
1516          * address to be the same as the slave's.
1517          */
1518         if (!bond_has_slaves(bond) &&
1519             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1520                 bond_set_dev_addr(bond->dev, slave_dev);
1521
1522         new_slave = bond_alloc_slave(bond, slave_dev);
1523         if (!new_slave) {
1524                 res = -ENOMEM;
1525                 goto err_undo_flags;
1526         }
1527
1528         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1529          * is set via sysfs or module option if desired.
1530          */
1531         new_slave->queue_id = 0;
1532
1533         /* Save slave's original mtu and then set it to match the bond */
1534         new_slave->original_mtu = slave_dev->mtu;
1535         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1536         if (res) {
1537                 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1538                 goto err_free;
1539         }
1540
1541         /* Save slave's original ("permanent") mac address for modes
1542          * that need it, and for restoring it upon release, and then
1543          * set it to the master's address
1544          */
1545         bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1546                           slave_dev->addr_len);
1547
1548         if (!bond->params.fail_over_mac ||
1549             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1550                 /* Set slave to master's mac address.  The application already
1551                  * set the master's mac address to that of the first slave
1552                  */
1553                 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1554                 ss.ss_family = slave_dev->type;
1555                 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1556                 if (res) {
1557                         netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1558                         goto err_restore_mtu;
1559                 }
1560         }
1561
1562         /* set slave flag before open to prevent IPv6 addrconf */
1563         slave_dev->flags |= IFF_SLAVE;
1564
1565         /* open the slave since the application closed it */
1566         res = dev_open(slave_dev);
1567         if (res) {
1568                 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1569                 goto err_restore_mac;
1570         }
1571
1572         slave_dev->priv_flags |= IFF_BONDING;
1573         /* initialize slave stats */
1574         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1575
1576         if (bond_is_lb(bond)) {
1577                 /* bond_alb_init_slave() must be called before all other stages since
1578                  * it might fail and we do not want to have to undo everything
1579                  */
1580                 res = bond_alb_init_slave(bond, new_slave);
1581                 if (res)
1582                         goto err_close;
1583         }
1584
1585         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1586         if (res) {
1587                 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1588                            slave_dev->name);
1589                 goto err_close;
1590         }
1591
1592         prev_slave = bond_last_slave(bond);
1593
1594         new_slave->delay = 0;
1595         new_slave->link_failure_count = 0;
1596
1597         if (bond_update_speed_duplex(new_slave) &&
1598             bond_needs_speed_duplex(bond))
1599                 new_slave->link = BOND_LINK_DOWN;
1600
1601         new_slave->last_rx = jiffies -
1602                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1603         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1604                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1605
1606         if (bond->params.miimon && !bond->params.use_carrier) {
1607                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1608
1609                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1610                         /* miimon is set but a bonded network driver
1611                          * does not support ETHTOOL/MII and
1612                          * arp_interval is not set.  Note: if
1613                          * use_carrier is enabled, we will never go
1614                          * here (because netif_carrier is always
1615                          * supported); thus, we don't need to change
1616                          * the messages for netif_carrier.
1617                          */
1618                         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",
1619                                     slave_dev->name);
1620                 } else if (link_reporting == -1) {
1621                         /* unable get link status using mii/ethtool */
1622                         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",
1623                                     slave_dev->name);
1624                 }
1625         }
1626
1627         /* check for initial state */
1628         new_slave->link = BOND_LINK_NOCHANGE;
1629         if (bond->params.miimon) {
1630                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1631                         if (bond->params.updelay) {
1632                                 bond_set_slave_link_state(new_slave,
1633                                                           BOND_LINK_BACK,
1634                                                           BOND_SLAVE_NOTIFY_NOW);
1635                                 new_slave->delay = bond->params.updelay;
1636                         } else {
1637                                 bond_set_slave_link_state(new_slave,
1638                                                           BOND_LINK_UP,
1639                                                           BOND_SLAVE_NOTIFY_NOW);
1640                         }
1641                 } else {
1642                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1643                                                   BOND_SLAVE_NOTIFY_NOW);
1644                 }
1645         } else if (bond->params.arp_interval) {
1646                 bond_set_slave_link_state(new_slave,
1647                                           (netif_carrier_ok(slave_dev) ?
1648                                           BOND_LINK_UP : BOND_LINK_DOWN),
1649                                           BOND_SLAVE_NOTIFY_NOW);
1650         } else {
1651                 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1652                                           BOND_SLAVE_NOTIFY_NOW);
1653         }
1654
1655         if (new_slave->link != BOND_LINK_DOWN)
1656                 new_slave->last_link_up = jiffies;
1657         netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1658                    new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1659                    (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1660
1661         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1662                 /* if there is a primary slave, remember it */
1663                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1664                         rcu_assign_pointer(bond->primary_slave, new_slave);
1665                         bond->force_primary = true;
1666                 }
1667         }
1668
1669         switch (BOND_MODE(bond)) {
1670         case BOND_MODE_ACTIVEBACKUP:
1671                 bond_set_slave_inactive_flags(new_slave,
1672                                               BOND_SLAVE_NOTIFY_NOW);
1673                 break;
1674         case BOND_MODE_8023AD:
1675                 /* in 802.3ad mode, the internal mechanism
1676                  * will activate the slaves in the selected
1677                  * aggregator
1678                  */
1679                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1680                 /* if this is the first slave */
1681                 if (!prev_slave) {
1682                         SLAVE_AD_INFO(new_slave)->id = 1;
1683                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1684                          * can be called only after the mac address of the bond is set
1685                          */
1686                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1687                 } else {
1688                         SLAVE_AD_INFO(new_slave)->id =
1689                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1690                 }
1691
1692                 bond_3ad_bind_slave(new_slave);
1693                 break;
1694         case BOND_MODE_TLB:
1695         case BOND_MODE_ALB:
1696                 bond_set_active_slave(new_slave);
1697                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1698                 break;
1699         default:
1700                 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1701
1702                 /* always active in trunk mode */
1703                 bond_set_active_slave(new_slave);
1704
1705                 /* In trunking mode there is little meaning to curr_active_slave
1706                  * anyway (it holds no special properties of the bond device),
1707                  * so we can change it without calling change_active_interface()
1708                  */
1709                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1710                     new_slave->link == BOND_LINK_UP)
1711                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1712
1713                 break;
1714         } /* switch(bond_mode) */
1715
1716 #ifdef CONFIG_NET_POLL_CONTROLLER
1717         if (bond->dev->npinfo) {
1718                 if (slave_enable_netpoll(new_slave)) {
1719                         netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1720                         res = -EBUSY;
1721                         goto err_detach;
1722                 }
1723         }
1724 #endif
1725
1726         if (!(bond_dev->features & NETIF_F_LRO))
1727                 dev_disable_lro(slave_dev);
1728
1729         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1730                                          new_slave);
1731         if (res) {
1732                 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1733                 goto err_detach;
1734         }
1735
1736         res = bond_master_upper_dev_link(bond, new_slave, extack);
1737         if (res) {
1738                 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1739                 goto err_unregister;
1740         }
1741
1742         res = bond_sysfs_slave_add(new_slave);
1743         if (res) {
1744                 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1745                 goto err_upper_unlink;
1746         }
1747
1748         bond->nest_level = dev_get_nest_level(bond_dev) + 1;
1749
1750         /* If the mode uses primary, then the following is handled by
1751          * bond_change_active_slave().
1752          */
1753         if (!bond_uses_primary(bond)) {
1754                 /* set promiscuity level to new slave */
1755                 if (bond_dev->flags & IFF_PROMISC) {
1756                         res = dev_set_promiscuity(slave_dev, 1);
1757                         if (res)
1758                                 goto err_sysfs_del;
1759                 }
1760
1761                 /* set allmulti level to new slave */
1762                 if (bond_dev->flags & IFF_ALLMULTI) {
1763                         res = dev_set_allmulti(slave_dev, 1);
1764                         if (res) {
1765                                 if (bond_dev->flags & IFF_PROMISC)
1766                                         dev_set_promiscuity(slave_dev, -1);
1767                                 goto err_sysfs_del;
1768                         }
1769                 }
1770
1771                 netif_addr_lock_bh(bond_dev);
1772                 dev_mc_sync_multiple(slave_dev, bond_dev);
1773                 dev_uc_sync_multiple(slave_dev, bond_dev);
1774                 netif_addr_unlock_bh(bond_dev);
1775
1776                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1777                         /* add lacpdu mc addr to mc list */
1778                         u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1779
1780                         dev_mc_add(slave_dev, lacpdu_multicast);
1781                 }
1782         }
1783
1784         bond->slave_cnt++;
1785         bond_compute_features(bond);
1786         bond_set_carrier(bond);
1787
1788         if (bond_uses_primary(bond)) {
1789                 block_netpoll_tx();
1790                 bond_select_active_slave(bond);
1791                 unblock_netpoll_tx();
1792         }
1793
1794         if (bond_mode_can_use_xmit_hash(bond))
1795                 bond_update_slave_arr(bond, NULL);
1796
1797
1798         netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1799                     slave_dev->name,
1800                     bond_is_active_slave(new_slave) ? "an active" : "a backup",
1801                     new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1802
1803         /* enslave is successful */
1804         bond_queue_slave_event(new_slave);
1805         return 0;
1806
1807 /* Undo stages on error */
1808 err_sysfs_del:
1809         bond_sysfs_slave_del(new_slave);
1810
1811 err_upper_unlink:
1812         bond_upper_dev_unlink(bond, new_slave);
1813
1814 err_unregister:
1815         netdev_rx_handler_unregister(slave_dev);
1816
1817 err_detach:
1818         vlan_vids_del_by_dev(slave_dev, bond_dev);
1819         if (rcu_access_pointer(bond->primary_slave) == new_slave)
1820                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1821         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1822                 block_netpoll_tx();
1823                 bond_change_active_slave(bond, NULL);
1824                 bond_select_active_slave(bond);
1825                 unblock_netpoll_tx();
1826         }
1827         /* either primary_slave or curr_active_slave might've changed */
1828         synchronize_rcu();
1829         slave_disable_netpoll(new_slave);
1830
1831 err_close:
1832         if (!netif_is_bond_master(slave_dev))
1833                 slave_dev->priv_flags &= ~IFF_BONDING;
1834         dev_close(slave_dev);
1835
1836 err_restore_mac:
1837         slave_dev->flags &= ~IFF_SLAVE;
1838         if (!bond->params.fail_over_mac ||
1839             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1840                 /* XXX TODO - fom follow mode needs to change master's
1841                  * MAC if this slave's MAC is in use by the bond, or at
1842                  * least print a warning.
1843                  */
1844                 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
1845                                   new_slave->dev->addr_len);
1846                 ss.ss_family = slave_dev->type;
1847                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1848         }
1849
1850 err_restore_mtu:
1851         dev_set_mtu(slave_dev, new_slave->original_mtu);
1852
1853 err_free:
1854         kobject_put(&new_slave->kobj);
1855
1856 err_undo_flags:
1857         /* Enslave of first slave has failed and we need to fix master's mac */
1858         if (!bond_has_slaves(bond)) {
1859                 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1860                                             slave_dev->dev_addr))
1861                         eth_hw_addr_random(bond_dev);
1862                 if (bond_dev->type != ARPHRD_ETHER) {
1863                         dev_close(bond_dev);
1864                         ether_setup(bond_dev);
1865                         bond_dev->flags |= IFF_MASTER;
1866                         bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1867                 }
1868         }
1869
1870         return res;
1871 }
1872
1873 /* Try to release the slave device <slave> from the bond device <master>
1874  * It is legal to access curr_active_slave without a lock because all the function
1875  * is RTNL-locked. If "all" is true it means that the function is being called
1876  * while destroying a bond interface and all slaves are being released.
1877  *
1878  * The rules for slave state should be:
1879  *   for Active/Backup:
1880  *     Active stays on all backups go down
1881  *   for Bonded connections:
1882  *     The first up interface should be left on and all others downed.
1883  */
1884 static int __bond_release_one(struct net_device *bond_dev,
1885                               struct net_device *slave_dev,
1886                               bool all, bool unregister)
1887 {
1888         struct bonding *bond = netdev_priv(bond_dev);
1889         struct slave *slave, *oldcurrent;
1890         struct sockaddr_storage ss;
1891         int old_flags = bond_dev->flags;
1892         netdev_features_t old_features = bond_dev->features;
1893
1894         /* slave is not a slave or master is not master of this slave */
1895         if (!(slave_dev->flags & IFF_SLAVE) ||
1896             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1897                 netdev_dbg(bond_dev, "cannot release %s\n",
1898                            slave_dev->name);
1899                 return -EINVAL;
1900         }
1901
1902         block_netpoll_tx();
1903
1904         slave = bond_get_slave_by_dev(bond, slave_dev);
1905         if (!slave) {
1906                 /* not a slave of this bond */
1907                 netdev_info(bond_dev, "%s not enslaved\n",
1908                             slave_dev->name);
1909                 unblock_netpoll_tx();
1910                 return -EINVAL;
1911         }
1912
1913         bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1914
1915         bond_sysfs_slave_del(slave);
1916
1917         /* recompute stats just before removing the slave */
1918         bond_get_stats(bond->dev, &bond->bond_stats);
1919
1920         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1921          * for this slave anymore.
1922          */
1923         netdev_rx_handler_unregister(slave_dev);
1924
1925         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1926                 bond_3ad_unbind_slave(slave);
1927
1928         bond_upper_dev_unlink(bond, slave);
1929
1930         if (bond_mode_can_use_xmit_hash(bond))
1931                 bond_update_slave_arr(bond, slave);
1932
1933         netdev_info(bond_dev, "Releasing %s interface %s\n",
1934                     bond_is_active_slave(slave) ? "active" : "backup",
1935                     slave_dev->name);
1936
1937         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1938
1939         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1940
1941         if (!all && (!bond->params.fail_over_mac ||
1942                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1943                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1944                     bond_has_slaves(bond))
1945                         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",
1946                                     slave_dev->name, slave->perm_hwaddr,
1947                                     bond_dev->name, slave_dev->name);
1948         }
1949
1950         if (rtnl_dereference(bond->primary_slave) == slave)
1951                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1952
1953         if (oldcurrent == slave)
1954                 bond_change_active_slave(bond, NULL);
1955
1956         if (bond_is_lb(bond)) {
1957                 /* Must be called only after the slave has been
1958                  * detached from the list and the curr_active_slave
1959                  * has been cleared (if our_slave == old_current),
1960                  * but before a new active slave is selected.
1961                  */
1962                 bond_alb_deinit_slave(bond, slave);
1963         }
1964
1965         if (all) {
1966                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1967         } else if (oldcurrent == slave) {
1968                 /* Note that we hold RTNL over this sequence, so there
1969                  * is no concern that another slave add/remove event
1970                  * will interfere.
1971                  */
1972                 bond_select_active_slave(bond);
1973         }
1974
1975         if (!bond_has_slaves(bond)) {
1976                 bond_set_carrier(bond);
1977                 eth_hw_addr_random(bond_dev);
1978                 bond->nest_level = SINGLE_DEPTH_NESTING;
1979         } else {
1980                 bond->nest_level = dev_get_nest_level(bond_dev) + 1;
1981         }
1982
1983         unblock_netpoll_tx();
1984         synchronize_rcu();
1985         bond->slave_cnt--;
1986
1987         if (!bond_has_slaves(bond)) {
1988                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1989                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1990         }
1991
1992         bond_compute_features(bond);
1993         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1994             (old_features & NETIF_F_VLAN_CHALLENGED))
1995                 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1996                             slave_dev->name, bond_dev->name);
1997
1998         vlan_vids_del_by_dev(slave_dev, bond_dev);
1999
2000         /* If the mode uses primary, then this case was handled above by
2001          * bond_change_active_slave(..., NULL)
2002          */
2003         if (!bond_uses_primary(bond)) {
2004                 /* unset promiscuity level from slave
2005                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
2006                  * of the IFF_PROMISC flag in the bond_dev, but we need the
2007                  * value of that flag before that change, as that was the value
2008                  * when this slave was attached, so we cache at the start of the
2009                  * function and use it here. Same goes for ALLMULTI below
2010                  */
2011                 if (old_flags & IFF_PROMISC)
2012                         dev_set_promiscuity(slave_dev, -1);
2013
2014                 /* unset allmulti level from slave */
2015                 if (old_flags & IFF_ALLMULTI)
2016                         dev_set_allmulti(slave_dev, -1);
2017
2018                 bond_hw_addr_flush(bond_dev, slave_dev);
2019         }
2020
2021         slave_disable_netpoll(slave);
2022
2023         /* close slave before restoring its mac address */
2024         dev_close(slave_dev);
2025
2026         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2027             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2028                 /* restore original ("permanent") mac address */
2029                 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2030                                   slave->dev->addr_len);
2031                 ss.ss_family = slave_dev->type;
2032                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
2033         }
2034
2035         if (unregister)
2036                 __dev_set_mtu(slave_dev, slave->original_mtu);
2037         else
2038                 dev_set_mtu(slave_dev, slave->original_mtu);
2039
2040         if (!netif_is_bond_master(slave_dev))
2041                 slave_dev->priv_flags &= ~IFF_BONDING;
2042
2043         kobject_put(&slave->kobj);
2044
2045         return 0;
2046 }
2047
2048 /* A wrapper used because of ndo_del_link */
2049 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2050 {
2051         return __bond_release_one(bond_dev, slave_dev, false, false);
2052 }
2053
2054 /* First release a slave and then destroy the bond if no more slaves are left.
2055  * Must be under rtnl_lock when this function is called.
2056  */
2057 static int  bond_release_and_destroy(struct net_device *bond_dev,
2058                                      struct net_device *slave_dev)
2059 {
2060         struct bonding *bond = netdev_priv(bond_dev);
2061         int ret;
2062
2063         ret = __bond_release_one(bond_dev, slave_dev, false, true);
2064         if (ret == 0 && !bond_has_slaves(bond) &&
2065             bond_dev->reg_state != NETREG_UNREGISTERING) {
2066                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2067                 netdev_info(bond_dev, "Destroying bond %s\n",
2068                             bond_dev->name);
2069                 bond_remove_proc_entry(bond);
2070                 unregister_netdevice(bond_dev);
2071         }
2072         return ret;
2073 }
2074
2075 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2076 {
2077         struct bonding *bond = netdev_priv(bond_dev);
2078         bond_fill_ifbond(bond, info);
2079 }
2080
2081 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2082 {
2083         struct bonding *bond = netdev_priv(bond_dev);
2084         struct list_head *iter;
2085         int i = 0, res = -ENODEV;
2086         struct slave *slave;
2087
2088         bond_for_each_slave(bond, slave, iter) {
2089                 if (i++ == (int)info->slave_id) {
2090                         res = 0;
2091                         bond_fill_ifslave(slave, info);
2092                         break;
2093                 }
2094         }
2095
2096         return res;
2097 }
2098
2099 /*-------------------------------- Monitoring -------------------------------*/
2100
2101 /* called with rcu_read_lock() */
2102 static int bond_miimon_inspect(struct bonding *bond)
2103 {
2104         bool ignore_updelay = false;
2105         int link_state, commit = 0;
2106         struct list_head *iter;
2107         struct slave *slave;
2108
2109         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2110
2111         bond_for_each_slave_rcu(bond, slave, iter) {
2112                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2113
2114                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2115
2116                 switch (slave->link) {
2117                 case BOND_LINK_UP:
2118                         if (link_state)
2119                                 continue;
2120
2121                         bond_propose_link_state(slave, BOND_LINK_FAIL);
2122                         commit++;
2123                         slave->delay = bond->params.downdelay;
2124                         if (slave->delay) {
2125                                 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2126                                             (BOND_MODE(bond) ==
2127                                              BOND_MODE_ACTIVEBACKUP) ?
2128                                              (bond_is_active_slave(slave) ?
2129                                               "active " : "backup ") : "",
2130                                             slave->dev->name,
2131                                             bond->params.downdelay * bond->params.miimon);
2132                         }
2133                         /*FALLTHRU*/
2134                 case BOND_LINK_FAIL:
2135                         if (link_state) {
2136                                 /* recovered before downdelay expired */
2137                                 bond_propose_link_state(slave, BOND_LINK_UP);
2138                                 slave->last_link_up = jiffies;
2139                                 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2140                                             (bond->params.downdelay - slave->delay) *
2141                                             bond->params.miimon,
2142                                             slave->dev->name);
2143                                 commit++;
2144                                 continue;
2145                         }
2146
2147                         if (slave->delay <= 0) {
2148                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2149                                 commit++;
2150                                 continue;
2151                         }
2152
2153                         slave->delay--;
2154                         break;
2155
2156                 case BOND_LINK_DOWN:
2157                         if (!link_state)
2158                                 continue;
2159
2160                         bond_propose_link_state(slave, BOND_LINK_BACK);
2161                         commit++;
2162                         slave->delay = bond->params.updelay;
2163
2164                         if (slave->delay) {
2165                                 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2166                                             slave->dev->name,
2167                                             ignore_updelay ? 0 :
2168                                             bond->params.updelay *
2169                                             bond->params.miimon);
2170                         }
2171                         /*FALLTHRU*/
2172                 case BOND_LINK_BACK:
2173                         if (!link_state) {
2174                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2175                                 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2176                                             (bond->params.updelay - slave->delay) *
2177                                             bond->params.miimon,
2178                                             slave->dev->name);
2179                                 commit++;
2180                                 continue;
2181                         }
2182
2183                         if (ignore_updelay)
2184                                 slave->delay = 0;
2185
2186                         if (slave->delay <= 0) {
2187                                 bond_propose_link_state(slave, BOND_LINK_UP);
2188                                 commit++;
2189                                 ignore_updelay = false;
2190                                 continue;
2191                         }
2192
2193                         slave->delay--;
2194                         break;
2195                 }
2196         }
2197
2198         return commit;
2199 }
2200
2201 static void bond_miimon_link_change(struct bonding *bond,
2202                                     struct slave *slave,
2203                                     char link)
2204 {
2205         switch (BOND_MODE(bond)) {
2206         case BOND_MODE_8023AD:
2207                 bond_3ad_handle_link_change(slave, link);
2208                 break;
2209         case BOND_MODE_TLB:
2210         case BOND_MODE_ALB:
2211                 bond_alb_handle_link_change(bond, slave, link);
2212                 break;
2213         case BOND_MODE_XOR:
2214                 bond_update_slave_arr(bond, NULL);
2215                 break;
2216         }
2217 }
2218
2219 static void bond_miimon_commit(struct bonding *bond)
2220 {
2221         struct list_head *iter;
2222         struct slave *slave, *primary;
2223
2224         bond_for_each_slave(bond, slave, iter) {
2225                 switch (slave->link_new_state) {
2226                 case BOND_LINK_NOCHANGE:
2227                         /* For 802.3ad mode, check current slave speed and
2228                          * duplex again in case its port was disabled after
2229                          * invalid speed/duplex reporting but recovered before
2230                          * link monitoring could make a decision on the actual
2231                          * link status
2232                          */
2233                         if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2234                             slave->link == BOND_LINK_UP)
2235                                 bond_3ad_adapter_speed_duplex_changed(slave);
2236                         continue;
2237
2238                 case BOND_LINK_UP:
2239                         if (bond_update_speed_duplex(slave) &&
2240                             bond_needs_speed_duplex(bond)) {
2241                                 slave->link = BOND_LINK_DOWN;
2242                                 if (net_ratelimit())
2243                                         netdev_warn(bond->dev,
2244                                                     "failed to get link speed/duplex for %s\n",
2245                                                     slave->dev->name);
2246                                 continue;
2247                         }
2248                         bond_set_slave_link_state(slave, BOND_LINK_UP,
2249                                                   BOND_SLAVE_NOTIFY_NOW);
2250                         slave->last_link_up = jiffies;
2251
2252                         primary = rtnl_dereference(bond->primary_slave);
2253                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2254                                 /* prevent it from being the active one */
2255                                 bond_set_backup_slave(slave);
2256                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2257                                 /* make it immediately active */
2258                                 bond_set_active_slave(slave);
2259                         }
2260
2261                         netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2262                                     slave->dev->name,
2263                                     slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2264                                     slave->duplex ? "full" : "half");
2265
2266                         bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2267
2268                         if (!bond->curr_active_slave || slave == primary)
2269                                 goto do_failover;
2270
2271                         continue;
2272
2273                 case BOND_LINK_DOWN:
2274                         if (slave->link_failure_count < UINT_MAX)
2275                                 slave->link_failure_count++;
2276
2277                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2278                                                   BOND_SLAVE_NOTIFY_NOW);
2279
2280                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2281                             BOND_MODE(bond) == BOND_MODE_8023AD)
2282                                 bond_set_slave_inactive_flags(slave,
2283                                                               BOND_SLAVE_NOTIFY_NOW);
2284
2285                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2286                                     slave->dev->name);
2287
2288                         bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2289
2290                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2291                                 goto do_failover;
2292
2293                         continue;
2294
2295                 default:
2296                         netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2297                                    slave->link_new_state, slave->dev->name);
2298                         bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2299
2300                         continue;
2301                 }
2302
2303 do_failover:
2304                 block_netpoll_tx();
2305                 bond_select_active_slave(bond);
2306                 unblock_netpoll_tx();
2307         }
2308
2309         bond_set_carrier(bond);
2310 }
2311
2312 /* bond_mii_monitor
2313  *
2314  * Really a wrapper that splits the mii monitor into two phases: an
2315  * inspection, then (if inspection indicates something needs to be done)
2316  * an acquisition of appropriate locks followed by a commit phase to
2317  * implement whatever link state changes are indicated.
2318  */
2319 static void bond_mii_monitor(struct work_struct *work)
2320 {
2321         struct bonding *bond = container_of(work, struct bonding,
2322                                             mii_work.work);
2323         bool should_notify_peers = false;
2324         unsigned long delay;
2325         struct slave *slave;
2326         struct list_head *iter;
2327
2328         delay = msecs_to_jiffies(bond->params.miimon);
2329
2330         if (!bond_has_slaves(bond))
2331                 goto re_arm;
2332
2333         rcu_read_lock();
2334
2335         should_notify_peers = bond_should_notify_peers(bond);
2336
2337         if (bond_miimon_inspect(bond)) {
2338                 rcu_read_unlock();
2339
2340                 /* Race avoidance with bond_close cancel of workqueue */
2341                 if (!rtnl_trylock()) {
2342                         delay = 1;
2343                         should_notify_peers = false;
2344                         goto re_arm;
2345                 }
2346
2347                 bond_for_each_slave(bond, slave, iter) {
2348                         bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2349                 }
2350                 bond_miimon_commit(bond);
2351
2352                 rtnl_unlock();  /* might sleep, hold no other locks */
2353         } else
2354                 rcu_read_unlock();
2355
2356 re_arm:
2357         if (bond->params.miimon)
2358                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2359
2360         if (should_notify_peers) {
2361                 if (!rtnl_trylock())
2362                         return;
2363                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2364                 rtnl_unlock();
2365         }
2366 }
2367
2368 static int bond_upper_dev_walk(struct net_device *upper, void *data)
2369 {
2370         __be32 ip = *((__be32 *)data);
2371
2372         return ip == bond_confirm_addr(upper, 0, ip);
2373 }
2374
2375 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2376 {
2377         bool ret = false;
2378
2379         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2380                 return true;
2381
2382         rcu_read_lock();
2383         if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip))
2384                 ret = true;
2385         rcu_read_unlock();
2386
2387         return ret;
2388 }
2389
2390 /* We go to the (large) trouble of VLAN tagging ARP frames because
2391  * switches in VLAN mode (especially if ports are configured as
2392  * "native" to a VLAN) might not pass non-tagged frames.
2393  */
2394 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2395                           __be32 dest_ip, __be32 src_ip,
2396                           struct bond_vlan_tag *tags)
2397 {
2398         struct sk_buff *skb;
2399         struct bond_vlan_tag *outer_tag = tags;
2400
2401         netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2402                    arp_op, slave_dev->name, &dest_ip, &src_ip);
2403
2404         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2405                          NULL, slave_dev->dev_addr, NULL);
2406
2407         if (!skb) {
2408                 net_err_ratelimited("ARP packet allocation failed\n");
2409                 return;
2410         }
2411
2412         if (!tags || tags->vlan_proto == VLAN_N_VID)
2413                 goto xmit;
2414
2415         tags++;
2416
2417         /* Go through all the tags backwards and add them to the packet */
2418         while (tags->vlan_proto != VLAN_N_VID) {
2419                 if (!tags->vlan_id) {
2420                         tags++;
2421                         continue;
2422                 }
2423
2424                 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2425                            ntohs(outer_tag->vlan_proto), tags->vlan_id);
2426                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2427                                                 tags->vlan_id);
2428                 if (!skb) {
2429                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2430                         return;
2431                 }
2432
2433                 tags++;
2434         }
2435         /* Set the outer tag */
2436         if (outer_tag->vlan_id) {
2437                 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2438                            ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2439                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2440                                        outer_tag->vlan_id);
2441         }
2442
2443 xmit:
2444         arp_xmit(skb);
2445 }
2446
2447 /* Validate the device path between the @start_dev and the @end_dev.
2448  * The path is valid if the @end_dev is reachable through device
2449  * stacking.
2450  * When the path is validated, collect any vlan information in the
2451  * path.
2452  */
2453 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2454                                               struct net_device *end_dev,
2455                                               int level)
2456 {
2457         struct bond_vlan_tag *tags;
2458         struct net_device *upper;
2459         struct list_head  *iter;
2460
2461         if (start_dev == end_dev) {
2462                 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2463                 if (!tags)
2464                         return ERR_PTR(-ENOMEM);
2465                 tags[level].vlan_proto = VLAN_N_VID;
2466                 return tags;
2467         }
2468
2469         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2470                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2471                 if (IS_ERR_OR_NULL(tags)) {
2472                         if (IS_ERR(tags))
2473                                 return tags;
2474                         continue;
2475                 }
2476                 if (is_vlan_dev(upper)) {
2477                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2478                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2479                 }
2480
2481                 return tags;
2482         }
2483
2484         return NULL;
2485 }
2486
2487 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2488 {
2489         struct rtable *rt;
2490         struct bond_vlan_tag *tags;
2491         __be32 *targets = bond->params.arp_targets, addr;
2492         int i;
2493
2494         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2495                 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2496                 tags = NULL;
2497
2498                 /* Find out through which dev should the packet go */
2499                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2500                                      RTO_ONLINK, 0);
2501                 if (IS_ERR(rt)) {
2502                         /* there's no route to target - try to send arp
2503                          * probe to generate any traffic (arp_validate=0)
2504                          */
2505                         if (bond->params.arp_validate)
2506                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2507                                                      bond->dev->name,
2508                                                      &targets[i]);
2509                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2510                                       0, tags);
2511                         continue;
2512                 }
2513
2514                 /* bond device itself */
2515                 if (rt->dst.dev == bond->dev)
2516                         goto found;
2517
2518                 rcu_read_lock();
2519                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2520                 rcu_read_unlock();
2521
2522                 if (!IS_ERR_OR_NULL(tags))
2523                         goto found;
2524
2525                 /* Not our device - skip */
2526                 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2527                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2528
2529                 ip_rt_put(rt);
2530                 continue;
2531
2532 found:
2533                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2534                 ip_rt_put(rt);
2535                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2536                               addr, tags);
2537                 kfree(tags);
2538         }
2539 }
2540
2541 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2542 {
2543         int i;
2544
2545         if (!sip || !bond_has_this_ip(bond, tip)) {
2546                 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2547                            &sip, &tip);
2548                 return;
2549         }
2550
2551         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2552         if (i == -1) {
2553                 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2554                            &sip);
2555                 return;
2556         }
2557         slave->last_rx = jiffies;
2558         slave->target_last_arp_rx[i] = jiffies;
2559 }
2560
2561 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2562                  struct slave *slave)
2563 {
2564         struct arphdr *arp = (struct arphdr *)skb->data;
2565         struct slave *curr_active_slave, *curr_arp_slave;
2566         unsigned char *arp_ptr;
2567         __be32 sip, tip;
2568         int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2569         unsigned int alen;
2570
2571         if (!slave_do_arp_validate(bond, slave)) {
2572                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2573                     !slave_do_arp_validate_only(bond))
2574                         slave->last_rx = jiffies;
2575                 return RX_HANDLER_ANOTHER;
2576         } else if (!is_arp) {
2577                 return RX_HANDLER_ANOTHER;
2578         }
2579
2580         alen = arp_hdr_len(bond->dev);
2581
2582         netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2583                    skb->dev->name);
2584
2585         if (alen > skb_headlen(skb)) {
2586                 arp = kmalloc(alen, GFP_ATOMIC);
2587                 if (!arp)
2588                         goto out_unlock;
2589                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2590                         goto out_unlock;
2591         }
2592
2593         if (arp->ar_hln != bond->dev->addr_len ||
2594             skb->pkt_type == PACKET_OTHERHOST ||
2595             skb->pkt_type == PACKET_LOOPBACK ||
2596             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2597             arp->ar_pro != htons(ETH_P_IP) ||
2598             arp->ar_pln != 4)
2599                 goto out_unlock;
2600
2601         arp_ptr = (unsigned char *)(arp + 1);
2602         arp_ptr += bond->dev->addr_len;
2603         memcpy(&sip, arp_ptr, 4);
2604         arp_ptr += 4 + bond->dev->addr_len;
2605         memcpy(&tip, arp_ptr, 4);
2606
2607         netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2608                    slave->dev->name, bond_slave_state(slave),
2609                      bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2610                      &sip, &tip);
2611
2612         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2613         curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2614
2615         /* We 'trust' the received ARP enough to validate it if:
2616          *
2617          * (a) the slave receiving the ARP is active (which includes the
2618          * current ARP slave, if any), or
2619          *
2620          * (b) the receiving slave isn't active, but there is a currently
2621          * active slave and it received valid arp reply(s) after it became
2622          * the currently active slave, or
2623          *
2624          * (c) there is an ARP slave that sent an ARP during the prior ARP
2625          * interval, and we receive an ARP reply on any slave.  We accept
2626          * these because switch FDB update delays may deliver the ARP
2627          * reply to a slave other than the sender of the ARP request.
2628          *
2629          * Note: for (b), backup slaves are receiving the broadcast ARP
2630          * request, not a reply.  This request passes from the sending
2631          * slave through the L2 switch(es) to the receiving slave.  Since
2632          * this is checking the request, sip/tip are swapped for
2633          * validation.
2634          *
2635          * This is done to avoid endless looping when we can't reach the
2636          * arp_ip_target and fool ourselves with our own arp requests.
2637          */
2638         if (bond_is_active_slave(slave))
2639                 bond_validate_arp(bond, slave, sip, tip);
2640         else if (curr_active_slave &&
2641                  time_after(slave_last_rx(bond, curr_active_slave),
2642                             curr_active_slave->last_link_up))
2643                 bond_validate_arp(bond, slave, tip, sip);
2644         else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2645                  bond_time_in_interval(bond,
2646                                        dev_trans_start(curr_arp_slave->dev), 1))
2647                 bond_validate_arp(bond, slave, sip, tip);
2648
2649 out_unlock:
2650         if (arp != (struct arphdr *)skb->data)
2651                 kfree(arp);
2652         return RX_HANDLER_ANOTHER;
2653 }
2654
2655 /* function to verify if we're in the arp_interval timeslice, returns true if
2656  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2657  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2658  */
2659 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2660                                   int mod)
2661 {
2662         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2663
2664         return time_in_range(jiffies,
2665                              last_act - delta_in_ticks,
2666                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2667 }
2668
2669 /* This function is called regularly to monitor each slave's link
2670  * ensuring that traffic is being sent and received when arp monitoring
2671  * is used in load-balancing mode. if the adapter has been dormant, then an
2672  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2673  * arp monitoring in active backup mode.
2674  */
2675 static void bond_loadbalance_arp_mon(struct bonding *bond)
2676 {
2677         struct slave *slave, *oldcurrent;
2678         struct list_head *iter;
2679         int do_failover = 0, slave_state_changed = 0;
2680
2681         if (!bond_has_slaves(bond))
2682                 goto re_arm;
2683
2684         rcu_read_lock();
2685
2686         oldcurrent = rcu_dereference(bond->curr_active_slave);
2687         /* see if any of the previous devices are up now (i.e. they have
2688          * xmt and rcv traffic). the curr_active_slave does not come into
2689          * the picture unless it is null. also, slave->last_link_up is not
2690          * needed here because we send an arp on each slave and give a slave
2691          * as long as it needs to get the tx/rx within the delta.
2692          * TODO: what about up/down delay in arp mode? it wasn't here before
2693          *       so it can wait
2694          */
2695         bond_for_each_slave_rcu(bond, slave, iter) {
2696                 unsigned long trans_start = dev_trans_start(slave->dev);
2697
2698                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2699
2700                 if (slave->link != BOND_LINK_UP) {
2701                         if (bond_time_in_interval(bond, trans_start, 1) &&
2702                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2703
2704                                 bond_propose_link_state(slave, BOND_LINK_UP);
2705                                 slave_state_changed = 1;
2706
2707                                 /* primary_slave has no meaning in round-robin
2708                                  * mode. the window of a slave being up and
2709                                  * curr_active_slave being null after enslaving
2710                                  * is closed.
2711                                  */
2712                                 if (!oldcurrent) {
2713                                         netdev_info(bond->dev, "link status definitely up for interface %s\n",
2714                                                     slave->dev->name);
2715                                         do_failover = 1;
2716                                 } else {
2717                                         netdev_info(bond->dev, "interface %s is now up\n",
2718                                                     slave->dev->name);
2719                                 }
2720                         }
2721                 } else {
2722                         /* slave->link == BOND_LINK_UP */
2723
2724                         /* not all switches will respond to an arp request
2725                          * when the source ip is 0, so don't take the link down
2726                          * if we don't know our ip yet
2727                          */
2728                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2729                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2730
2731                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2732                                 slave_state_changed = 1;
2733
2734                                 if (slave->link_failure_count < UINT_MAX)
2735                                         slave->link_failure_count++;
2736
2737                                 netdev_info(bond->dev, "interface %s is now down\n",
2738                                             slave->dev->name);
2739
2740                                 if (slave == oldcurrent)
2741                                         do_failover = 1;
2742                         }
2743                 }
2744
2745                 /* note: if switch is in round-robin mode, all links
2746                  * must tx arp to ensure all links rx an arp - otherwise
2747                  * links may oscillate or not come up at all; if switch is
2748                  * in something like xor mode, there is nothing we can
2749                  * do - all replies will be rx'ed on same link causing slaves
2750                  * to be unstable during low/no traffic periods
2751                  */
2752                 if (bond_slave_is_up(slave))
2753                         bond_arp_send_all(bond, slave);
2754         }
2755
2756         rcu_read_unlock();
2757
2758         if (do_failover || slave_state_changed) {
2759                 if (!rtnl_trylock())
2760                         goto re_arm;
2761
2762                 bond_for_each_slave(bond, slave, iter) {
2763                         if (slave->link_new_state != BOND_LINK_NOCHANGE)
2764                                 slave->link = slave->link_new_state;
2765                 }
2766
2767                 if (slave_state_changed) {
2768                         bond_slave_state_change(bond);
2769                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2770                                 bond_update_slave_arr(bond, NULL);
2771                 }
2772                 if (do_failover) {
2773                         block_netpoll_tx();
2774                         bond_select_active_slave(bond);
2775                         unblock_netpoll_tx();
2776                 }
2777                 rtnl_unlock();
2778         }
2779
2780 re_arm:
2781         if (bond->params.arp_interval)
2782                 queue_delayed_work(bond->wq, &bond->arp_work,
2783                                    msecs_to_jiffies(bond->params.arp_interval));
2784 }
2785
2786 /* Called to inspect slaves for active-backup mode ARP monitor link state
2787  * changes.  Sets proposed link state in slaves to specify what action
2788  * should take place for the slave.  Returns 0 if no changes are found, >0
2789  * if changes to link states must be committed.
2790  *
2791  * Called with rcu_read_lock held.
2792  */
2793 static int bond_ab_arp_inspect(struct bonding *bond)
2794 {
2795         unsigned long trans_start, last_rx;
2796         struct list_head *iter;
2797         struct slave *slave;
2798         int commit = 0;
2799
2800         bond_for_each_slave_rcu(bond, slave, iter) {
2801                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2802                 last_rx = slave_last_rx(bond, slave);
2803
2804                 if (slave->link != BOND_LINK_UP) {
2805                         if (bond_time_in_interval(bond, last_rx, 1)) {
2806                                 bond_propose_link_state(slave, BOND_LINK_UP);
2807                                 commit++;
2808                         } else if (slave->link == BOND_LINK_BACK) {
2809                                 bond_propose_link_state(slave, BOND_LINK_FAIL);
2810                                 commit++;
2811                         }
2812                         continue;
2813                 }
2814
2815                 /* Give slaves 2*delta after being enslaved or made
2816                  * active.  This avoids bouncing, as the last receive
2817                  * times need a full ARP monitor cycle to be updated.
2818                  */
2819                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2820                         continue;
2821
2822                 /* Backup slave is down if:
2823                  * - No current_arp_slave AND
2824                  * - more than 3*delta since last receive AND
2825                  * - the bond has an IP address
2826                  *
2827                  * Note: a non-null current_arp_slave indicates
2828                  * the curr_active_slave went down and we are
2829                  * searching for a new one; under this condition
2830                  * we only take the curr_active_slave down - this
2831                  * gives each slave a chance to tx/rx traffic
2832                  * before being taken out
2833                  */
2834                 if (!bond_is_active_slave(slave) &&
2835                     !rcu_access_pointer(bond->current_arp_slave) &&
2836                     !bond_time_in_interval(bond, last_rx, 3)) {
2837                         bond_propose_link_state(slave, BOND_LINK_DOWN);
2838                         commit++;
2839                 }
2840
2841                 /* Active slave is down if:
2842                  * - more than 2*delta since transmitting OR
2843                  * - (more than 2*delta since receive AND
2844                  *    the bond has an IP address)
2845                  */
2846                 trans_start = dev_trans_start(slave->dev);
2847                 if (bond_is_active_slave(slave) &&
2848                     (!bond_time_in_interval(bond, trans_start, 2) ||
2849                      !bond_time_in_interval(bond, last_rx, 2))) {
2850                         bond_propose_link_state(slave, BOND_LINK_DOWN);
2851                         commit++;
2852                 }
2853         }
2854
2855         return commit;
2856 }
2857
2858 /* Called to commit link state changes noted by inspection step of
2859  * active-backup mode ARP monitor.
2860  *
2861  * Called with RTNL hold.
2862  */
2863 static void bond_ab_arp_commit(struct bonding *bond)
2864 {
2865         unsigned long trans_start;
2866         struct list_head *iter;
2867         struct slave *slave;
2868
2869         bond_for_each_slave(bond, slave, iter) {
2870                 switch (slave->link_new_state) {
2871                 case BOND_LINK_NOCHANGE:
2872                         continue;
2873
2874                 case BOND_LINK_UP:
2875                         trans_start = dev_trans_start(slave->dev);
2876                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
2877                             (!rtnl_dereference(bond->curr_active_slave) &&
2878                              bond_time_in_interval(bond, trans_start, 1))) {
2879                                 struct slave *current_arp_slave;
2880
2881                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2882                                 bond_set_slave_link_state(slave, BOND_LINK_UP,
2883                                                           BOND_SLAVE_NOTIFY_NOW);
2884                                 if (current_arp_slave) {
2885                                         bond_set_slave_inactive_flags(
2886                                                 current_arp_slave,
2887                                                 BOND_SLAVE_NOTIFY_NOW);
2888                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2889                                 }
2890
2891                                 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2892                                             slave->dev->name);
2893
2894                                 if (!rtnl_dereference(bond->curr_active_slave) ||
2895                                     slave == rtnl_dereference(bond->primary_slave))
2896                                         goto do_failover;
2897
2898                         }
2899
2900                         continue;
2901
2902                 case BOND_LINK_DOWN:
2903                         if (slave->link_failure_count < UINT_MAX)
2904                                 slave->link_failure_count++;
2905
2906                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2907                                                   BOND_SLAVE_NOTIFY_NOW);
2908                         bond_set_slave_inactive_flags(slave,
2909                                                       BOND_SLAVE_NOTIFY_NOW);
2910
2911                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2912                                     slave->dev->name);
2913
2914                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
2915                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2916                                 goto do_failover;
2917                         }
2918
2919                         continue;
2920
2921                 case BOND_LINK_FAIL:
2922                         bond_set_slave_link_state(slave, BOND_LINK_FAIL,
2923                                                   BOND_SLAVE_NOTIFY_NOW);
2924                         bond_set_slave_inactive_flags(slave,
2925                                                       BOND_SLAVE_NOTIFY_NOW);
2926
2927                         /* A slave has just been enslaved and has become
2928                          * the current active slave.
2929                          */
2930                         if (rtnl_dereference(bond->curr_active_slave))
2931                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2932                         continue;
2933
2934                 default:
2935                         netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2936                                    slave->link_new_state, slave->dev->name);
2937                         continue;
2938                 }
2939
2940 do_failover:
2941                 block_netpoll_tx();
2942                 bond_select_active_slave(bond);
2943                 unblock_netpoll_tx();
2944         }
2945
2946         bond_set_carrier(bond);
2947 }
2948
2949 /* Send ARP probes for active-backup mode ARP monitor.
2950  *
2951  * Called with rcu_read_lock held.
2952  */
2953 static bool bond_ab_arp_probe(struct bonding *bond)
2954 {
2955         struct slave *slave, *before = NULL, *new_slave = NULL,
2956                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2957                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2958         struct list_head *iter;
2959         bool found = false;
2960         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2961
2962         if (curr_arp_slave && curr_active_slave)
2963                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2964                             curr_arp_slave->dev->name,
2965                             curr_active_slave->dev->name);
2966
2967         if (curr_active_slave) {
2968                 bond_arp_send_all(bond, curr_active_slave);
2969                 return should_notify_rtnl;
2970         }
2971
2972         /* if we don't have a curr_active_slave, search for the next available
2973          * backup slave from the current_arp_slave and make it the candidate
2974          * for becoming the curr_active_slave
2975          */
2976
2977         if (!curr_arp_slave) {
2978                 curr_arp_slave = bond_first_slave_rcu(bond);
2979                 if (!curr_arp_slave)
2980                         return should_notify_rtnl;
2981         }
2982
2983         bond_for_each_slave_rcu(bond, slave, iter) {
2984                 if (!found && !before && bond_slave_is_up(slave))
2985                         before = slave;
2986
2987                 if (found && !new_slave && bond_slave_is_up(slave))
2988                         new_slave = slave;
2989                 /* if the link state is up at this point, we
2990                  * mark it down - this can happen if we have
2991                  * simultaneous link failures and
2992                  * reselect_active_interface doesn't make this
2993                  * one the current slave so it is still marked
2994                  * up when it is actually down
2995                  */
2996                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2997                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2998                                                   BOND_SLAVE_NOTIFY_LATER);
2999                         if (slave->link_failure_count < UINT_MAX)
3000                                 slave->link_failure_count++;
3001
3002                         bond_set_slave_inactive_flags(slave,
3003                                                       BOND_SLAVE_NOTIFY_LATER);
3004
3005                         netdev_info(bond->dev, "backup interface %s is now down\n",
3006                                     slave->dev->name);
3007                 }
3008                 if (slave == curr_arp_slave)
3009                         found = true;
3010         }
3011
3012         if (!new_slave && before)
3013                 new_slave = before;
3014
3015         if (!new_slave)
3016                 goto check_state;
3017
3018         bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
3019                                   BOND_SLAVE_NOTIFY_LATER);
3020         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
3021         bond_arp_send_all(bond, new_slave);
3022         new_slave->last_link_up = jiffies;
3023         rcu_assign_pointer(bond->current_arp_slave, new_slave);
3024
3025 check_state:
3026         bond_for_each_slave_rcu(bond, slave, iter) {
3027                 if (slave->should_notify || slave->should_notify_link) {
3028                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3029                         break;
3030                 }
3031         }
3032         return should_notify_rtnl;
3033 }
3034
3035 static void bond_activebackup_arp_mon(struct bonding *bond)
3036 {
3037         bool should_notify_peers = false;
3038         bool should_notify_rtnl = false;
3039         int delta_in_ticks;
3040
3041         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3042
3043         if (!bond_has_slaves(bond))
3044                 goto re_arm;
3045
3046         rcu_read_lock();
3047
3048         should_notify_peers = bond_should_notify_peers(bond);
3049
3050         if (bond_ab_arp_inspect(bond)) {
3051                 rcu_read_unlock();
3052
3053                 /* Race avoidance with bond_close flush of workqueue */
3054                 if (!rtnl_trylock()) {
3055                         delta_in_ticks = 1;
3056                         should_notify_peers = false;
3057                         goto re_arm;
3058                 }
3059
3060                 bond_ab_arp_commit(bond);
3061
3062                 rtnl_unlock();
3063                 rcu_read_lock();
3064         }
3065
3066         should_notify_rtnl = bond_ab_arp_probe(bond);
3067         rcu_read_unlock();
3068
3069 re_arm:
3070         if (bond->params.arp_interval)
3071                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3072
3073         if (should_notify_peers || should_notify_rtnl) {
3074                 if (!rtnl_trylock())
3075                         return;
3076
3077                 if (should_notify_peers) {
3078                         bond->send_peer_notif--;
3079                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3080                                                  bond->dev);
3081                 }
3082                 if (should_notify_rtnl) {
3083                         bond_slave_state_notify(bond);
3084                         bond_slave_link_notify(bond);
3085                 }
3086
3087                 rtnl_unlock();
3088         }
3089 }
3090
3091 static void bond_arp_monitor(struct work_struct *work)
3092 {
3093         struct bonding *bond = container_of(work, struct bonding,
3094                                             arp_work.work);
3095
3096         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3097                 bond_activebackup_arp_mon(bond);
3098         else
3099                 bond_loadbalance_arp_mon(bond);
3100 }
3101
3102 /*-------------------------- netdev event handling --------------------------*/
3103
3104 /* Change device name */
3105 static int bond_event_changename(struct bonding *bond)
3106 {
3107         bond_remove_proc_entry(bond);
3108         bond_create_proc_entry(bond);
3109
3110         bond_debug_reregister(bond);
3111
3112         return NOTIFY_DONE;
3113 }
3114
3115 static int bond_master_netdev_event(unsigned long event,
3116                                     struct net_device *bond_dev)
3117 {
3118         struct bonding *event_bond = netdev_priv(bond_dev);
3119
3120         switch (event) {
3121         case NETDEV_CHANGENAME:
3122                 return bond_event_changename(event_bond);
3123         case NETDEV_UNREGISTER:
3124                 bond_remove_proc_entry(event_bond);
3125                 break;
3126         case NETDEV_REGISTER:
3127                 bond_create_proc_entry(event_bond);
3128                 break;
3129         case NETDEV_NOTIFY_PEERS:
3130                 if (event_bond->send_peer_notif)
3131                         event_bond->send_peer_notif--;
3132                 break;
3133         default:
3134                 break;
3135         }
3136
3137         return NOTIFY_DONE;
3138 }
3139
3140 static int bond_slave_netdev_event(unsigned long event,
3141                                    struct net_device *slave_dev)
3142 {
3143         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3144         struct bonding *bond;
3145         struct net_device *bond_dev;
3146
3147         /* A netdev event can be generated while enslaving a device
3148          * before netdev_rx_handler_register is called in which case
3149          * slave will be NULL
3150          */
3151         if (!slave)
3152                 return NOTIFY_DONE;
3153         bond_dev = slave->bond->dev;
3154         bond = slave->bond;
3155         primary = rtnl_dereference(bond->primary_slave);
3156
3157         switch (event) {
3158         case NETDEV_UNREGISTER:
3159                 if (bond_dev->type != ARPHRD_ETHER)
3160                         bond_release_and_destroy(bond_dev, slave_dev);
3161                 else
3162                         __bond_release_one(bond_dev, slave_dev, false, true);
3163                 break;
3164         case NETDEV_UP:
3165         case NETDEV_CHANGE:
3166                 /* For 802.3ad mode only:
3167                  * Getting invalid Speed/Duplex values here will put slave
3168                  * in weird state. Mark it as link-fail if the link was
3169                  * previously up or link-down if it hasn't yet come up, and
3170                  * let link-monitoring (miimon) set it right when correct
3171                  * speeds/duplex are available.
3172                  */
3173                 if (bond_update_speed_duplex(slave) &&
3174                     BOND_MODE(bond) == BOND_MODE_8023AD) {
3175                         if (slave->last_link_up)
3176                                 slave->link = BOND_LINK_FAIL;
3177                         else
3178                                 slave->link = BOND_LINK_DOWN;
3179                 }
3180
3181                 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3182                         bond_3ad_adapter_speed_duplex_changed(slave);
3183                 /* Fallthrough */
3184         case NETDEV_DOWN:
3185                 /* Refresh slave-array if applicable!
3186                  * If the setup does not use miimon or arpmon (mode-specific!),
3187                  * then these events will not cause the slave-array to be
3188                  * refreshed. This will cause xmit to use a slave that is not
3189                  * usable. Avoid such situation by refeshing the array at these
3190                  * events. If these (miimon/arpmon) parameters are configured
3191                  * then array gets refreshed twice and that should be fine!
3192                  */
3193                 if (bond_mode_can_use_xmit_hash(bond))
3194                         bond_update_slave_arr(bond, NULL);
3195                 break;
3196         case NETDEV_CHANGEMTU:
3197                 /* TODO: Should slaves be allowed to
3198                  * independently alter their MTU?  For
3199                  * an active-backup bond, slaves need
3200                  * not be the same type of device, so
3201                  * MTUs may vary.  For other modes,
3202                  * slaves arguably should have the
3203                  * same MTUs. To do this, we'd need to
3204                  * take over the slave's change_mtu
3205                  * function for the duration of their
3206                  * servitude.
3207                  */
3208                 break;
3209         case NETDEV_CHANGENAME:
3210                 /* we don't care if we don't have primary set */
3211                 if (!bond_uses_primary(bond) ||
3212                     !bond->params.primary[0])
3213                         break;
3214
3215                 if (slave == primary) {
3216                         /* slave's name changed - he's no longer primary */
3217                         RCU_INIT_POINTER(bond->primary_slave, NULL);
3218                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3219                         /* we have a new primary slave */
3220                         rcu_assign_pointer(bond->primary_slave, slave);
3221                 } else { /* we didn't change primary - exit */
3222                         break;
3223                 }
3224
3225                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3226                             primary ? slave_dev->name : "none");
3227
3228                 block_netpoll_tx();
3229                 bond_select_active_slave(bond);
3230                 unblock_netpoll_tx();
3231                 break;
3232         case NETDEV_FEAT_CHANGE:
3233                 bond_compute_features(bond);
3234                 break;
3235         case NETDEV_RESEND_IGMP:
3236                 /* Propagate to master device */
3237                 call_netdevice_notifiers(event, slave->bond->dev);
3238                 break;
3239         default:
3240                 break;
3241         }
3242
3243         return NOTIFY_DONE;
3244 }
3245
3246 /* bond_netdev_event: handle netdev notifier chain events.
3247  *
3248  * This function receives events for the netdev chain.  The caller (an
3249  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3250  * locks for us to safely manipulate the slave devices (RTNL lock,
3251  * dev_probe_lock).
3252  */
3253 static int bond_netdev_event(struct notifier_block *this,
3254                              unsigned long event, void *ptr)
3255 {
3256         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3257
3258         netdev_dbg(event_dev, "event: %lx\n", event);
3259
3260         if (!(event_dev->priv_flags & IFF_BONDING))
3261                 return NOTIFY_DONE;
3262
3263         if (event_dev->flags & IFF_MASTER) {
3264                 int ret;
3265
3266                 netdev_dbg(event_dev, "IFF_MASTER\n");
3267                 ret = bond_master_netdev_event(event, event_dev);
3268                 if (ret != NOTIFY_DONE)
3269                         return ret;
3270         }
3271
3272         if (event_dev->flags & IFF_SLAVE) {
3273                 netdev_dbg(event_dev, "IFF_SLAVE\n");
3274                 return bond_slave_netdev_event(event, event_dev);
3275         }
3276
3277         return NOTIFY_DONE;
3278 }
3279
3280 static struct notifier_block bond_netdev_notifier = {
3281         .notifier_call = bond_netdev_event,
3282 };
3283
3284 /*---------------------------- Hashing Policies -----------------------------*/
3285
3286 /* L2 hash helper */
3287 static inline u32 bond_eth_hash(struct sk_buff *skb)
3288 {
3289         struct ethhdr *ep, hdr_tmp;
3290
3291         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3292         if (ep)
3293                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3294         return 0;
3295 }
3296
3297 /* Extract the appropriate headers based on bond's xmit policy */
3298 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3299                               struct flow_keys *fk)
3300 {
3301         const struct ipv6hdr *iph6;
3302         const struct iphdr *iph;
3303         int noff, proto = -1;
3304
3305         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3306                 return skb_flow_dissect_flow_keys(skb, fk, 0);
3307
3308         fk->ports.ports = 0;
3309         noff = skb_network_offset(skb);
3310         if (skb->protocol == htons(ETH_P_IP)) {
3311                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3312                         return false;
3313                 iph = ip_hdr(skb);
3314                 iph_to_flow_copy_v4addrs(fk, iph);
3315                 noff += iph->ihl << 2;
3316                 if (!ip_is_fragment(iph))
3317                         proto = iph->protocol;
3318         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3319                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3320                         return false;
3321                 iph6 = ipv6_hdr(skb);
3322                 iph_to_flow_copy_v6addrs(fk, iph6);
3323                 noff += sizeof(*iph6);
3324                 proto = iph6->nexthdr;
3325         } else {
3326                 return false;
3327         }
3328         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3329                 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3330
3331         return true;
3332 }
3333
3334 /**
3335  * bond_xmit_hash - generate a hash value based on the xmit policy
3336  * @bond: bonding device
3337  * @skb: buffer to use for headers
3338  *
3339  * This function will extract the necessary headers from the skb buffer and use
3340  * them to generate a hash based on the xmit_policy set in the bonding device
3341  */
3342 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3343 {
3344         struct flow_keys flow;
3345         u32 hash;
3346
3347         if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3348             skb->l4_hash)
3349                 return skb->hash;
3350
3351         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3352             !bond_flow_dissect(bond, skb, &flow))
3353                 return bond_eth_hash(skb);
3354
3355         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3356             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3357                 hash = bond_eth_hash(skb);
3358         else
3359                 hash = (__force u32)flow.ports.ports;
3360         hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3361                 (__force u32)flow_get_u32_src(&flow);
3362         hash ^= (hash >> 16);
3363         hash ^= (hash >> 8);
3364
3365         return hash >> 1;
3366 }
3367
3368 /*-------------------------- Device entry points ----------------------------*/
3369
3370 void bond_work_init_all(struct bonding *bond)
3371 {
3372         INIT_DELAYED_WORK(&bond->mcast_work,
3373                           bond_resend_igmp_join_requests_delayed);
3374         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3375         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3376         INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3377         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3378         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3379 }
3380
3381 static void bond_work_cancel_all(struct bonding *bond)
3382 {
3383         cancel_delayed_work_sync(&bond->mii_work);
3384         cancel_delayed_work_sync(&bond->arp_work);
3385         cancel_delayed_work_sync(&bond->alb_work);
3386         cancel_delayed_work_sync(&bond->ad_work);
3387         cancel_delayed_work_sync(&bond->mcast_work);
3388         cancel_delayed_work_sync(&bond->slave_arr_work);
3389 }
3390
3391 static int bond_open(struct net_device *bond_dev)
3392 {
3393         struct bonding *bond = netdev_priv(bond_dev);
3394         struct list_head *iter;
3395         struct slave *slave;
3396
3397         /* reset slave->backup and slave->inactive */
3398         if (bond_has_slaves(bond)) {
3399                 bond_for_each_slave(bond, slave, iter) {
3400                         if (bond_uses_primary(bond) &&
3401                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3402                                 bond_set_slave_inactive_flags(slave,
3403                                                               BOND_SLAVE_NOTIFY_NOW);
3404                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3405                                 bond_set_slave_active_flags(slave,
3406                                                             BOND_SLAVE_NOTIFY_NOW);
3407                         }
3408                 }
3409         }
3410
3411         if (bond_is_lb(bond)) {
3412                 /* bond_alb_initialize must be called before the timer
3413                  * is started.
3414                  */
3415                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3416                         return -ENOMEM;
3417                 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
3418                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3419         }
3420
3421         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3422                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3423
3424         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3425                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3426                 bond->recv_probe = bond_arp_rcv;
3427         }
3428
3429         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3430                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3431                 /* register to receive LACPDUs */
3432                 bond->recv_probe = bond_3ad_lacpdu_recv;
3433                 bond_3ad_initiate_agg_selection(bond, 1);
3434         }
3435
3436         if (bond_mode_can_use_xmit_hash(bond))
3437                 bond_update_slave_arr(bond, NULL);
3438
3439         return 0;
3440 }
3441
3442 static int bond_close(struct net_device *bond_dev)
3443 {
3444         struct bonding *bond = netdev_priv(bond_dev);
3445
3446         bond_work_cancel_all(bond);
3447         bond->send_peer_notif = 0;
3448         if (bond_is_lb(bond))
3449                 bond_alb_deinitialize(bond);
3450         bond->recv_probe = NULL;
3451
3452         return 0;
3453 }
3454
3455 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3456  * that some drivers can provide 32bit values only.
3457  */
3458 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3459                             const struct rtnl_link_stats64 *_new,
3460                             const struct rtnl_link_stats64 *_old)
3461 {
3462         const u64 *new = (const u64 *)_new;
3463         const u64 *old = (const u64 *)_old;
3464         u64 *res = (u64 *)_res;
3465         int i;
3466
3467         for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3468                 u64 nv = new[i];
3469                 u64 ov = old[i];
3470                 s64 delta = nv - ov;
3471
3472                 /* detects if this particular field is 32bit only */
3473                 if (((nv | ov) >> 32) == 0)
3474                         delta = (s64)(s32)((u32)nv - (u32)ov);
3475
3476                 /* filter anomalies, some drivers reset their stats
3477                  * at down/up events.
3478                  */
3479                 if (delta > 0)
3480                         res[i] += delta;
3481         }
3482 }
3483
3484 static int bond_get_nest_level(struct net_device *bond_dev)
3485 {
3486         struct bonding *bond = netdev_priv(bond_dev);
3487
3488         return bond->nest_level;
3489 }
3490
3491 static void bond_get_stats(struct net_device *bond_dev,
3492                            struct rtnl_link_stats64 *stats)
3493 {
3494         struct bonding *bond = netdev_priv(bond_dev);
3495         struct rtnl_link_stats64 temp;
3496         struct list_head *iter;
3497         struct slave *slave;
3498
3499         spin_lock_nested(&bond->stats_lock, bond_get_nest_level(bond_dev));
3500         memcpy(stats, &bond->bond_stats, sizeof(*stats));
3501
3502         rcu_read_lock();
3503         bond_for_each_slave_rcu(bond, slave, iter) {
3504                 const struct rtnl_link_stats64 *new =
3505                         dev_get_stats(slave->dev, &temp);
3506
3507                 bond_fold_stats(stats, new, &slave->slave_stats);
3508
3509                 /* save off the slave stats for the next run */
3510                 memcpy(&slave->slave_stats, new, sizeof(*new));
3511         }
3512         rcu_read_unlock();
3513
3514         memcpy(&bond->bond_stats, stats, sizeof(*stats));
3515         spin_unlock(&bond->stats_lock);
3516 }
3517
3518 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3519 {
3520         struct bonding *bond = netdev_priv(bond_dev);
3521         struct net_device *slave_dev = NULL;
3522         struct ifbond k_binfo;
3523         struct ifbond __user *u_binfo = NULL;
3524         struct ifslave k_sinfo;
3525         struct ifslave __user *u_sinfo = NULL;
3526         struct mii_ioctl_data *mii = NULL;
3527         struct bond_opt_value newval;
3528         struct net *net;
3529         int res = 0;
3530
3531         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3532
3533         switch (cmd) {
3534         case SIOCGMIIPHY:
3535                 mii = if_mii(ifr);
3536                 if (!mii)
3537                         return -EINVAL;
3538
3539                 mii->phy_id = 0;
3540                 /* Fall Through */
3541         case SIOCGMIIREG:
3542                 /* We do this again just in case we were called by SIOCGMIIREG
3543                  * instead of SIOCGMIIPHY.
3544                  */
3545                 mii = if_mii(ifr);
3546                 if (!mii)
3547                         return -EINVAL;
3548
3549                 if (mii->reg_num == 1) {
3550                         mii->val_out = 0;
3551                         if (netif_carrier_ok(bond->dev))
3552                                 mii->val_out = BMSR_LSTATUS;
3553                 }
3554
3555                 return 0;
3556         case BOND_INFO_QUERY_OLD:
3557         case SIOCBONDINFOQUERY:
3558                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3559
3560                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3561                         return -EFAULT;
3562
3563                 bond_info_query(bond_dev, &k_binfo);
3564                 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3565                         return -EFAULT;
3566
3567                 return 0;
3568         case BOND_SLAVE_INFO_QUERY_OLD:
3569         case SIOCBONDSLAVEINFOQUERY:
3570                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3571
3572                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3573                         return -EFAULT;
3574
3575                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3576                 if (res == 0 &&
3577                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3578                         return -EFAULT;
3579
3580                 return res;
3581         default:
3582                 break;
3583         }
3584
3585         net = dev_net(bond_dev);
3586
3587         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3588                 return -EPERM;
3589
3590         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3591
3592         netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3593
3594         if (!slave_dev)
3595                 return -ENODEV;
3596
3597         netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3598         switch (cmd) {
3599         case BOND_ENSLAVE_OLD:
3600         case SIOCBONDENSLAVE:
3601                 res = bond_enslave(bond_dev, slave_dev, NULL);
3602                 break;
3603         case BOND_RELEASE_OLD:
3604         case SIOCBONDRELEASE:
3605                 res = bond_release(bond_dev, slave_dev);
3606                 break;
3607         case BOND_SETHWADDR_OLD:
3608         case SIOCBONDSETHWADDR:
3609                 bond_set_dev_addr(bond_dev, slave_dev);
3610                 res = 0;
3611                 break;
3612         case BOND_CHANGE_ACTIVE_OLD:
3613         case SIOCBONDCHANGEACTIVE:
3614                 bond_opt_initstr(&newval, slave_dev->name);
3615                 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
3616                                             &newval);
3617                 break;
3618         default:
3619                 res = -EOPNOTSUPP;
3620         }
3621
3622         return res;
3623 }
3624
3625 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3626 {
3627         struct bonding *bond = netdev_priv(bond_dev);
3628
3629         if (change & IFF_PROMISC)
3630                 bond_set_promiscuity(bond,
3631                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3632
3633         if (change & IFF_ALLMULTI)
3634                 bond_set_allmulti(bond,
3635                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3636 }
3637
3638 static void bond_set_rx_mode(struct net_device *bond_dev)
3639 {
3640         struct bonding *bond = netdev_priv(bond_dev);
3641         struct list_head *iter;
3642         struct slave *slave;
3643
3644         rcu_read_lock();
3645         if (bond_uses_primary(bond)) {
3646                 slave = rcu_dereference(bond->curr_active_slave);
3647                 if (slave) {
3648                         dev_uc_sync(slave->dev, bond_dev);
3649                         dev_mc_sync(slave->dev, bond_dev);
3650                 }
3651         } else {
3652                 bond_for_each_slave_rcu(bond, slave, iter) {
3653                         dev_uc_sync_multiple(slave->dev, bond_dev);
3654                         dev_mc_sync_multiple(slave->dev, bond_dev);
3655                 }
3656         }
3657         rcu_read_unlock();
3658 }
3659
3660 static int bond_neigh_init(struct neighbour *n)
3661 {
3662         struct bonding *bond = netdev_priv(n->dev);
3663         const struct net_device_ops *slave_ops;
3664         struct neigh_parms parms;
3665         struct slave *slave;
3666         int ret;
3667
3668         slave = bond_first_slave(bond);
3669         if (!slave)
3670                 return 0;
3671         slave_ops = slave->dev->netdev_ops;
3672         if (!slave_ops->ndo_neigh_setup)
3673                 return 0;
3674
3675         parms.neigh_setup = NULL;
3676         parms.neigh_cleanup = NULL;
3677         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3678         if (ret)
3679                 return ret;
3680
3681         /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3682          * after the last slave has been detached.  Assumes that all slaves
3683          * utilize the same neigh_cleanup (true at this writing as only user
3684          * is ipoib).
3685          */
3686         n->parms->neigh_cleanup = parms.neigh_cleanup;
3687
3688         if (!parms.neigh_setup)
3689                 return 0;
3690
3691         return parms.neigh_setup(n);
3692 }
3693
3694 /* The bonding ndo_neigh_setup is called at init time beofre any
3695  * slave exists. So we must declare proxy setup function which will
3696  * be used at run time to resolve the actual slave neigh param setup.
3697  *
3698  * It's also called by master devices (such as vlans) to setup their
3699  * underlying devices. In that case - do nothing, we're already set up from
3700  * our init.
3701  */
3702 static int bond_neigh_setup(struct net_device *dev,
3703                             struct neigh_parms *parms)
3704 {
3705         /* modify only our neigh_parms */
3706         if (parms->dev == dev)
3707                 parms->neigh_setup = bond_neigh_init;
3708
3709         return 0;
3710 }
3711
3712 /* Change the MTU of all of a master's slaves to match the master */
3713 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3714 {
3715         struct bonding *bond = netdev_priv(bond_dev);
3716         struct slave *slave, *rollback_slave;
3717         struct list_head *iter;
3718         int res = 0;
3719
3720         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3721
3722         bond_for_each_slave(bond, slave, iter) {
3723                 netdev_dbg(bond_dev, "s %p c_m %p\n",
3724                            slave, slave->dev->netdev_ops->ndo_change_mtu);
3725
3726                 res = dev_set_mtu(slave->dev, new_mtu);
3727
3728                 if (res) {
3729                         /* If we failed to set the slave's mtu to the new value
3730                          * we must abort the operation even in ACTIVE_BACKUP
3731                          * mode, because if we allow the backup slaves to have
3732                          * different mtu values than the active slave we'll
3733                          * need to change their mtu when doing a failover. That
3734                          * means changing their mtu from timer context, which
3735                          * is probably not a good idea.
3736                          */
3737                         netdev_dbg(bond_dev, "err %d %s\n", res,
3738                                    slave->dev->name);
3739                         goto unwind;
3740                 }
3741         }
3742
3743         bond_dev->mtu = new_mtu;
3744
3745         return 0;
3746
3747 unwind:
3748         /* unwind from head to the slave that failed */
3749         bond_for_each_slave(bond, rollback_slave, iter) {
3750                 int tmp_res;
3751
3752                 if (rollback_slave == slave)
3753                         break;
3754
3755                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3756                 if (tmp_res) {
3757                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3758                                    tmp_res, rollback_slave->dev->name);
3759                 }
3760         }
3761
3762         return res;
3763 }
3764
3765 /* Change HW address
3766  *
3767  * Note that many devices must be down to change the HW address, and
3768  * downing the master releases all slaves.  We can make bonds full of
3769  * bonding devices to test this, however.
3770  */
3771 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3772 {
3773         struct bonding *bond = netdev_priv(bond_dev);
3774         struct slave *slave, *rollback_slave;
3775         struct sockaddr_storage *ss = addr, tmp_ss;
3776         struct list_head *iter;
3777         int res = 0;
3778
3779         if (BOND_MODE(bond) == BOND_MODE_ALB)
3780                 return bond_alb_set_mac_address(bond_dev, addr);
3781
3782
3783         netdev_dbg(bond_dev, "bond=%p\n", bond);
3784
3785         /* If fail_over_mac is enabled, do nothing and return success.
3786          * Returning an error causes ifenslave to fail.
3787          */
3788         if (bond->params.fail_over_mac &&
3789             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3790                 return 0;
3791
3792         if (!is_valid_ether_addr(ss->__data))
3793                 return -EADDRNOTAVAIL;
3794
3795         bond_for_each_slave(bond, slave, iter) {
3796                 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3797                 res = dev_set_mac_address(slave->dev, addr);
3798                 if (res) {
3799                         /* TODO: consider downing the slave
3800                          * and retry ?
3801                          * User should expect communications
3802                          * breakage anyway until ARP finish
3803                          * updating, so...
3804                          */
3805                         netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3806                         goto unwind;
3807                 }
3808         }
3809
3810         /* success */
3811         memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
3812         return 0;
3813
3814 unwind:
3815         memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
3816         tmp_ss.ss_family = bond_dev->type;
3817
3818         /* unwind from head to the slave that failed */
3819         bond_for_each_slave(bond, rollback_slave, iter) {
3820                 int tmp_res;
3821
3822                 if (rollback_slave == slave)
3823                         break;
3824
3825                 tmp_res = dev_set_mac_address(rollback_slave->dev,
3826                                               (struct sockaddr *)&tmp_ss);
3827                 if (tmp_res) {
3828                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3829                                    tmp_res, rollback_slave->dev->name);
3830                 }
3831         }
3832
3833         return res;
3834 }
3835
3836 /**
3837  * bond_xmit_slave_id - transmit skb through slave with slave_id
3838  * @bond: bonding device that is transmitting
3839  * @skb: buffer to transmit
3840  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3841  *
3842  * This function tries to transmit through slave with slave_id but in case
3843  * it fails, it tries to find the first available slave for transmission.
3844  * The skb is consumed in all cases, thus the function is void.
3845  */
3846 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3847 {
3848         struct list_head *iter;
3849         struct slave *slave;
3850         int i = slave_id;
3851
3852         /* Here we start from the slave with slave_id */
3853         bond_for_each_slave_rcu(bond, slave, iter) {
3854                 if (--i < 0) {
3855                         if (bond_slave_can_tx(slave)) {
3856                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3857                                 return;
3858                         }
3859                 }
3860         }
3861
3862         /* Here we start from the first slave up to slave_id */
3863         i = slave_id;
3864         bond_for_each_slave_rcu(bond, slave, iter) {
3865                 if (--i < 0)
3866                         break;
3867                 if (bond_slave_can_tx(slave)) {
3868                         bond_dev_queue_xmit(bond, skb, slave->dev);
3869                         return;
3870                 }
3871         }
3872         /* no slave that can tx has been found */
3873         bond_tx_drop(bond->dev, skb);
3874 }
3875
3876 /**
3877  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3878  * @bond: bonding device to use
3879  *
3880  * Based on the value of the bonding device's packets_per_slave parameter
3881  * this function generates a slave id, which is usually used as the next
3882  * slave to transmit through.
3883  */
3884 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3885 {
3886         u32 slave_id;
3887         struct reciprocal_value reciprocal_packets_per_slave;
3888         int packets_per_slave = bond->params.packets_per_slave;
3889
3890         switch (packets_per_slave) {
3891         case 0:
3892                 slave_id = prandom_u32();
3893                 break;
3894         case 1:
3895                 slave_id = bond->rr_tx_counter;
3896                 break;
3897         default:
3898                 reciprocal_packets_per_slave =
3899                         bond->params.reciprocal_packets_per_slave;
3900                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3901                                              reciprocal_packets_per_slave);
3902                 break;
3903         }
3904         bond->rr_tx_counter++;
3905
3906         return slave_id;
3907 }
3908
3909 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
3910                                         struct net_device *bond_dev)
3911 {
3912         struct bonding *bond = netdev_priv(bond_dev);
3913         struct slave *slave;
3914         int slave_cnt;
3915         u32 slave_id;
3916
3917         /* Start with the curr_active_slave that joined the bond as the
3918          * default for sending IGMP traffic.  For failover purposes one
3919          * needs to maintain some consistency for the interface that will
3920          * send the join/membership reports.  The curr_active_slave found
3921          * will send all of this type of traffic.
3922          */
3923         if (skb->protocol == htons(ETH_P_IP)) {
3924                 int noff = skb_network_offset(skb);
3925                 struct iphdr *iph;
3926
3927                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3928                         goto non_igmp;
3929
3930                 iph = ip_hdr(skb);
3931                 if (iph->protocol == IPPROTO_IGMP) {
3932                         slave = rcu_dereference(bond->curr_active_slave);
3933                         if (slave)
3934                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3935                         else
3936                                 bond_xmit_slave_id(bond, skb, 0);
3937                         return NETDEV_TX_OK;
3938                 }
3939         }
3940
3941 non_igmp:
3942         slave_cnt = READ_ONCE(bond->slave_cnt);
3943         if (likely(slave_cnt)) {
3944                 slave_id = bond_rr_gen_slave_id(bond);
3945                 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3946         } else {
3947                 bond_tx_drop(bond_dev, skb);
3948         }
3949         return NETDEV_TX_OK;
3950 }
3951
3952 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3953  * the bond has a usable interface.
3954  */
3955 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
3956                                           struct net_device *bond_dev)
3957 {
3958         struct bonding *bond = netdev_priv(bond_dev);
3959         struct slave *slave;
3960
3961         slave = rcu_dereference(bond->curr_active_slave);
3962         if (slave)
3963                 bond_dev_queue_xmit(bond, skb, slave->dev);
3964         else
3965                 bond_tx_drop(bond_dev, skb);
3966
3967         return NETDEV_TX_OK;
3968 }
3969
3970 /* Use this to update slave_array when (a) it's not appropriate to update
3971  * slave_array right away (note that update_slave_array() may sleep)
3972  * and / or (b) RTNL is not held.
3973  */
3974 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3975 {
3976         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3977 }
3978
3979 /* Slave array work handler. Holds only RTNL */
3980 static void bond_slave_arr_handler(struct work_struct *work)
3981 {
3982         struct bonding *bond = container_of(work, struct bonding,
3983                                             slave_arr_work.work);
3984         int ret;
3985
3986         if (!rtnl_trylock())
3987                 goto err;
3988
3989         ret = bond_update_slave_arr(bond, NULL);
3990         rtnl_unlock();
3991         if (ret) {
3992                 pr_warn_ratelimited("Failed to update slave array from WT\n");
3993                 goto err;
3994         }
3995         return;
3996
3997 err:
3998         bond_slave_arr_work_rearm(bond, 1);
3999 }
4000
4001 static void bond_skip_slave(struct bond_up_slave *slaves,
4002                             struct slave *skipslave)
4003 {
4004         int idx;
4005
4006         /* Rare situation where caller has asked to skip a specific
4007          * slave but allocation failed (most likely!). BTW this is
4008          * only possible when the call is initiated from
4009          * __bond_release_one(). In this situation; overwrite the
4010          * skipslave entry in the array with the last entry from the
4011          * array to avoid a situation where the xmit path may choose
4012          * this to-be-skipped slave to send a packet out.
4013          */
4014         for (idx = 0; slaves && idx < slaves->count; idx++) {
4015                 if (skipslave == slaves->arr[idx]) {
4016                         slaves->arr[idx] =
4017                                 slaves->arr[slaves->count - 1];
4018                         slaves->count--;
4019                         break;
4020                 }
4021         }
4022 }
4023
4024 /* Build the usable slaves array in control path for modes that use xmit-hash
4025  * to determine the slave interface -
4026  * (a) BOND_MODE_8023AD
4027  * (b) BOND_MODE_XOR
4028  * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
4029  *
4030  * The caller is expected to hold RTNL only and NO other lock!
4031  */
4032 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
4033 {
4034         struct slave *slave;
4035         struct list_head *iter;
4036         struct bond_up_slave *new_arr, *old_arr;
4037         int agg_id = 0;
4038         int ret = 0;
4039
4040 #ifdef CONFIG_LOCKDEP
4041         WARN_ON(lockdep_is_held(&bond->mode_lock));
4042 #endif
4043
4044         new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
4045                           GFP_KERNEL);
4046         if (!new_arr) {
4047                 ret = -ENOMEM;
4048                 pr_err("Failed to build slave-array.\n");
4049                 goto out;
4050         }
4051         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4052                 struct ad_info ad_info;
4053
4054                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
4055                         pr_debug("bond_3ad_get_active_agg_info failed\n");
4056                         kfree_rcu(new_arr, rcu);
4057                         /* No active aggragator means it's not safe to use
4058                          * the previous array.
4059                          */
4060                         old_arr = rtnl_dereference(bond->slave_arr);
4061                         if (old_arr) {
4062                                 RCU_INIT_POINTER(bond->slave_arr, NULL);
4063                                 kfree_rcu(old_arr, rcu);
4064                         }
4065                         goto out;
4066                 }
4067                 agg_id = ad_info.aggregator_id;
4068         }
4069         bond_for_each_slave(bond, slave, iter) {
4070                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4071                         struct aggregator *agg;
4072
4073                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
4074                         if (!agg || agg->aggregator_identifier != agg_id)
4075                                 continue;
4076                 }
4077                 if (!bond_slave_can_tx(slave))
4078                         continue;
4079                 if (skipslave == slave)
4080                         continue;
4081
4082                 netdev_dbg(bond->dev,
4083                            "Adding slave dev %s to tx hash array[%d]\n",
4084                            slave->dev->name, new_arr->count);
4085
4086                 new_arr->arr[new_arr->count++] = slave;
4087         }
4088
4089         old_arr = rtnl_dereference(bond->slave_arr);
4090         rcu_assign_pointer(bond->slave_arr, new_arr);
4091         if (old_arr)
4092                 kfree_rcu(old_arr, rcu);
4093 out:
4094         if (ret != 0 && skipslave)
4095                 bond_skip_slave(rtnl_dereference(bond->slave_arr), skipslave);
4096
4097         return ret;
4098 }
4099
4100 /* Use this Xmit function for 3AD as well as XOR modes. The current
4101  * usable slave array is formed in the control path. The xmit function
4102  * just calculates hash and sends the packet out.
4103  */
4104 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
4105                                      struct net_device *dev)
4106 {
4107         struct bonding *bond = netdev_priv(dev);
4108         struct slave *slave;
4109         struct bond_up_slave *slaves;
4110         unsigned int count;
4111
4112         slaves = rcu_dereference(bond->slave_arr);
4113         count = slaves ? READ_ONCE(slaves->count) : 0;
4114         if (likely(count)) {
4115                 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
4116                 bond_dev_queue_xmit(bond, skb, slave->dev);
4117         } else {
4118                 bond_tx_drop(dev, skb);
4119         }
4120
4121         return NETDEV_TX_OK;
4122 }
4123
4124 /* in broadcast mode, we send everything to all usable interfaces. */
4125 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
4126                                        struct net_device *bond_dev)
4127 {
4128         struct bonding *bond = netdev_priv(bond_dev);
4129         struct slave *slave = NULL;
4130         struct list_head *iter;
4131
4132         bond_for_each_slave_rcu(bond, slave, iter) {
4133                 if (bond_is_last_slave(bond, slave))
4134                         break;
4135                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4136                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4137
4138                         if (!skb2) {
4139                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4140                                                     bond_dev->name, __func__);
4141                                 continue;
4142                         }
4143                         bond_dev_queue_xmit(bond, skb2, slave->dev);
4144                 }
4145         }
4146         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4147                 bond_dev_queue_xmit(bond, skb, slave->dev);
4148         else
4149                 bond_tx_drop(bond_dev, skb);
4150
4151         return NETDEV_TX_OK;
4152 }
4153
4154 /*------------------------- Device initialization ---------------------------*/
4155
4156 /* Lookup the slave that corresponds to a qid */
4157 static inline int bond_slave_override(struct bonding *bond,
4158                                       struct sk_buff *skb)
4159 {
4160         struct slave *slave = NULL;
4161         struct list_head *iter;
4162
4163         if (!skb_rx_queue_recorded(skb))
4164                 return 1;
4165
4166         /* Find out if any slaves have the same mapping as this skb. */
4167         bond_for_each_slave_rcu(bond, slave, iter) {
4168                 if (slave->queue_id == skb_get_queue_mapping(skb)) {
4169                         if (bond_slave_is_up(slave) &&
4170                             slave->link == BOND_LINK_UP) {
4171                                 bond_dev_queue_xmit(bond, skb, slave->dev);
4172                                 return 0;
4173                         }
4174                         /* If the slave isn't UP, use default transmit policy. */
4175                         break;
4176                 }
4177         }
4178
4179         return 1;
4180 }
4181
4182
4183 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4184                              struct net_device *sb_dev,
4185                              select_queue_fallback_t fallback)
4186 {
4187         /* This helper function exists to help dev_pick_tx get the correct
4188          * destination queue.  Using a helper function skips a call to
4189          * skb_tx_hash and will put the skbs in the queue we expect on their
4190          * way down to the bonding driver.
4191          */
4192         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4193
4194         /* Save the original txq to restore before passing to the driver */
4195         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
4196
4197         if (unlikely(txq >= dev->real_num_tx_queues)) {
4198                 do {
4199                         txq -= dev->real_num_tx_queues;
4200                 } while (txq >= dev->real_num_tx_queues);
4201         }
4202         return txq;
4203 }
4204
4205 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4206 {
4207         struct bonding *bond = netdev_priv(dev);
4208
4209         if (bond_should_override_tx_queue(bond) &&
4210             !bond_slave_override(bond, skb))
4211                 return NETDEV_TX_OK;
4212
4213         switch (BOND_MODE(bond)) {
4214         case BOND_MODE_ROUNDROBIN:
4215                 return bond_xmit_roundrobin(skb, dev);
4216         case BOND_MODE_ACTIVEBACKUP:
4217                 return bond_xmit_activebackup(skb, dev);
4218         case BOND_MODE_8023AD:
4219         case BOND_MODE_XOR:
4220                 return bond_3ad_xor_xmit(skb, dev);
4221         case BOND_MODE_BROADCAST:
4222                 return bond_xmit_broadcast(skb, dev);
4223         case BOND_MODE_ALB:
4224                 return bond_alb_xmit(skb, dev);
4225         case BOND_MODE_TLB:
4226                 return bond_tlb_xmit(skb, dev);
4227         default:
4228                 /* Should never happen, mode already checked */
4229                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4230                 WARN_ON_ONCE(1);
4231                 bond_tx_drop(dev, skb);
4232                 return NETDEV_TX_OK;
4233         }
4234 }
4235
4236 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4237 {
4238         struct bonding *bond = netdev_priv(dev);
4239         netdev_tx_t ret = NETDEV_TX_OK;
4240
4241         /* If we risk deadlock from transmitting this in the
4242          * netpoll path, tell netpoll to queue the frame for later tx
4243          */
4244         if (unlikely(is_netpoll_tx_blocked(dev)))
4245                 return NETDEV_TX_BUSY;
4246
4247         rcu_read_lock();
4248         if (bond_has_slaves(bond))
4249                 ret = __bond_start_xmit(skb, dev);
4250         else
4251                 bond_tx_drop(dev, skb);
4252         rcu_read_unlock();
4253
4254         return ret;
4255 }
4256
4257 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed)
4258 {
4259         if (speed == 0 || speed == SPEED_UNKNOWN)
4260                 speed = slave->speed;
4261         else
4262                 speed = min(speed, slave->speed);
4263
4264         return speed;
4265 }
4266
4267 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4268                                            struct ethtool_link_ksettings *cmd)
4269 {
4270         struct bonding *bond = netdev_priv(bond_dev);
4271         struct list_head *iter;
4272         struct slave *slave;
4273         u32 speed = 0;
4274
4275         cmd->base.duplex = DUPLEX_UNKNOWN;
4276         cmd->base.port = PORT_OTHER;
4277
4278         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4279          * do not need to check mode.  Though link speed might not represent
4280          * the true receive or transmit bandwidth (not all modes are symmetric)
4281          * this is an accurate maximum.
4282          */
4283         bond_for_each_slave(bond, slave, iter) {
4284                 if (bond_slave_can_tx(slave)) {
4285                         if (slave->speed != SPEED_UNKNOWN) {
4286                                 if (BOND_MODE(bond) == BOND_MODE_BROADCAST)
4287                                         speed = bond_mode_bcast_speed(slave,
4288                                                                       speed);
4289                                 else
4290                                         speed += slave->speed;
4291                         }
4292                         if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4293                             slave->duplex != DUPLEX_UNKNOWN)
4294                                 cmd->base.duplex = slave->duplex;
4295                 }
4296         }
4297         cmd->base.speed = speed ? : SPEED_UNKNOWN;
4298
4299         return 0;
4300 }
4301
4302 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4303                                      struct ethtool_drvinfo *drvinfo)
4304 {
4305         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4306         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4307         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4308                  BOND_ABI_VERSION);
4309 }
4310
4311 static const struct ethtool_ops bond_ethtool_ops = {
4312         .get_drvinfo            = bond_ethtool_get_drvinfo,
4313         .get_link               = ethtool_op_get_link,
4314         .get_link_ksettings     = bond_ethtool_get_link_ksettings,
4315 };
4316
4317 static const struct net_device_ops bond_netdev_ops = {
4318         .ndo_init               = bond_init,
4319         .ndo_uninit             = bond_uninit,
4320         .ndo_open               = bond_open,
4321         .ndo_stop               = bond_close,
4322         .ndo_start_xmit         = bond_start_xmit,
4323         .ndo_select_queue       = bond_select_queue,
4324         .ndo_get_stats64        = bond_get_stats,
4325         .ndo_do_ioctl           = bond_do_ioctl,
4326         .ndo_change_rx_flags    = bond_change_rx_flags,
4327         .ndo_set_rx_mode        = bond_set_rx_mode,
4328         .ndo_change_mtu         = bond_change_mtu,
4329         .ndo_set_mac_address    = bond_set_mac_address,
4330         .ndo_neigh_setup        = bond_neigh_setup,
4331         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4332         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4333         .ndo_get_lock_subclass  = bond_get_nest_level,
4334 #ifdef CONFIG_NET_POLL_CONTROLLER
4335         .ndo_netpoll_setup      = bond_netpoll_setup,
4336         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4337         .ndo_poll_controller    = bond_poll_controller,
4338 #endif
4339         .ndo_add_slave          = bond_enslave,
4340         .ndo_del_slave          = bond_release,
4341         .ndo_fix_features       = bond_fix_features,
4342         .ndo_features_check     = passthru_features_check,
4343 };
4344
4345 static const struct device_type bond_type = {
4346         .name = "bond",
4347 };
4348
4349 static void bond_destructor(struct net_device *bond_dev)
4350 {
4351         struct bonding *bond = netdev_priv(bond_dev);
4352         if (bond->wq)
4353                 destroy_workqueue(bond->wq);
4354 }
4355
4356 void bond_setup(struct net_device *bond_dev)
4357 {
4358         struct bonding *bond = netdev_priv(bond_dev);
4359
4360         spin_lock_init(&bond->mode_lock);
4361         spin_lock_init(&bond->stats_lock);
4362         bond->params = bonding_defaults;
4363
4364         /* Initialize pointers */
4365         bond->dev = bond_dev;
4366
4367         /* Initialize the device entry points */
4368         ether_setup(bond_dev);
4369         bond_dev->max_mtu = ETH_MAX_MTU;
4370         bond_dev->netdev_ops = &bond_netdev_ops;
4371         bond_dev->ethtool_ops = &bond_ethtool_ops;
4372
4373         bond_dev->needs_free_netdev = true;
4374         bond_dev->priv_destructor = bond_destructor;
4375
4376         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4377
4378         /* Initialize the device options */
4379         bond_dev->flags |= IFF_MASTER;
4380         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4381         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4382
4383         /* don't acquire bond device's netif_tx_lock when transmitting */
4384         bond_dev->features |= NETIF_F_LLTX;
4385
4386         /* By default, we declare the bond to be fully
4387          * VLAN hardware accelerated capable. Special
4388          * care is taken in the various xmit functions
4389          * when there are slaves that are not hw accel
4390          * capable
4391          */
4392
4393         /* Don't allow bond devices to change network namespaces. */
4394         bond_dev->features |= NETIF_F_NETNS_LOCAL;
4395
4396         bond_dev->hw_features = BOND_VLAN_FEATURES |
4397                                 NETIF_F_HW_VLAN_CTAG_RX |
4398                                 NETIF_F_HW_VLAN_CTAG_FILTER |
4399                                 NETIF_F_HW_VLAN_STAG_RX |
4400                                 NETIF_F_HW_VLAN_STAG_FILTER;
4401
4402         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4;
4403         bond_dev->features |= bond_dev->hw_features;
4404         bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
4405 }
4406
4407 /* Destroy a bonding device.
4408  * Must be under rtnl_lock when this function is called.
4409  */
4410 static void bond_uninit(struct net_device *bond_dev)
4411 {
4412         struct bonding *bond = netdev_priv(bond_dev);
4413         struct list_head *iter;
4414         struct slave *slave;
4415         struct bond_up_slave *arr;
4416
4417         bond_netpoll_cleanup(bond_dev);
4418
4419         /* Release the bonded slaves */
4420         bond_for_each_slave(bond, slave, iter)
4421                 __bond_release_one(bond_dev, slave->dev, true, true);
4422         netdev_info(bond_dev, "Released all slaves\n");
4423
4424         arr = rtnl_dereference(bond->slave_arr);
4425         if (arr) {
4426                 RCU_INIT_POINTER(bond->slave_arr, NULL);
4427                 kfree_rcu(arr, rcu);
4428         }
4429
4430         list_del(&bond->bond_list);
4431
4432         bond_debug_unregister(bond);
4433 }
4434
4435 /*------------------------- Module initialization ---------------------------*/
4436
4437 static int bond_check_params(struct bond_params *params)
4438 {
4439         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4440         struct bond_opt_value newval;
4441         const struct bond_opt_value *valptr;
4442         int arp_all_targets_value = 0;
4443         u16 ad_actor_sys_prio = 0;
4444         u16 ad_user_port_key = 0;
4445         __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
4446         int arp_ip_count;
4447         int bond_mode   = BOND_MODE_ROUNDROBIN;
4448         int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4449         int lacp_fast = 0;
4450         int tlb_dynamic_lb;
4451
4452         /* Convert string parameters. */
4453         if (mode) {
4454                 bond_opt_initstr(&newval, mode);
4455                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4456                 if (!valptr) {
4457                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4458                         return -EINVAL;
4459                 }
4460                 bond_mode = valptr->value;
4461         }
4462
4463         if (xmit_hash_policy) {
4464                 if (bond_mode == BOND_MODE_ROUNDROBIN ||
4465                     bond_mode == BOND_MODE_ACTIVEBACKUP ||
4466                     bond_mode == BOND_MODE_BROADCAST) {
4467                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4468                                 bond_mode_name(bond_mode));
4469                 } else {
4470                         bond_opt_initstr(&newval, xmit_hash_policy);
4471                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4472                                                 &newval);
4473                         if (!valptr) {
4474                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4475                                        xmit_hash_policy);
4476                                 return -EINVAL;
4477                         }
4478                         xmit_hashtype = valptr->value;
4479                 }
4480         }
4481
4482         if (lacp_rate) {
4483                 if (bond_mode != BOND_MODE_8023AD) {
4484                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4485                                 bond_mode_name(bond_mode));
4486                 } else {
4487                         bond_opt_initstr(&newval, lacp_rate);
4488                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4489                                                 &newval);
4490                         if (!valptr) {
4491                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4492                                        lacp_rate);
4493                                 return -EINVAL;
4494                         }
4495                         lacp_fast = valptr->value;
4496                 }
4497         }
4498
4499         if (ad_select) {
4500                 bond_opt_initstr(&newval, ad_select);
4501                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4502                                         &newval);
4503                 if (!valptr) {
4504                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4505                         return -EINVAL;
4506                 }
4507                 params->ad_select = valptr->value;
4508                 if (bond_mode != BOND_MODE_8023AD)
4509                         pr_warn("ad_select param only affects 802.3ad mode\n");
4510         } else {
4511                 params->ad_select = BOND_AD_STABLE;
4512         }
4513
4514         if (max_bonds < 0) {
4515                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4516                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4517                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4518         }
4519
4520         if (miimon < 0) {
4521                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4522                         miimon, INT_MAX);
4523                 miimon = 0;
4524         }
4525
4526         if (updelay < 0) {
4527                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4528                         updelay, INT_MAX);
4529                 updelay = 0;
4530         }
4531
4532         if (downdelay < 0) {
4533                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4534                         downdelay, INT_MAX);
4535                 downdelay = 0;
4536         }
4537
4538         if ((use_carrier != 0) && (use_carrier != 1)) {
4539                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4540                         use_carrier);
4541                 use_carrier = 1;
4542         }
4543
4544         if (num_peer_notif < 0 || num_peer_notif > 255) {
4545                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4546                         num_peer_notif);
4547                 num_peer_notif = 1;
4548         }
4549
4550         /* reset values for 802.3ad/TLB/ALB */
4551         if (!bond_mode_uses_arp(bond_mode)) {
4552                 if (!miimon) {
4553                         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");
4554                         pr_warn("Forcing miimon to 100msec\n");
4555                         miimon = BOND_DEFAULT_MIIMON;
4556                 }
4557         }
4558
4559         if (tx_queues < 1 || tx_queues > 255) {
4560                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4561                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4562                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4563         }
4564
4565         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4566                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4567                         all_slaves_active);
4568                 all_slaves_active = 0;
4569         }
4570
4571         if (resend_igmp < 0 || resend_igmp > 255) {
4572                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4573                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4574                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4575         }
4576
4577         bond_opt_initval(&newval, packets_per_slave);
4578         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4579                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4580                         packets_per_slave, USHRT_MAX);
4581                 packets_per_slave = 1;
4582         }
4583
4584         if (bond_mode == BOND_MODE_ALB) {
4585                 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",
4586                           updelay);
4587         }
4588
4589         if (!miimon) {
4590                 if (updelay || downdelay) {
4591                         /* just warn the user the up/down delay will have
4592                          * no effect since miimon is zero...
4593                          */
4594                         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",
4595                                 updelay, downdelay);
4596                 }
4597         } else {
4598                 /* don't allow arp monitoring */
4599                 if (arp_interval) {
4600                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4601                                 miimon, arp_interval);
4602                         arp_interval = 0;
4603                 }
4604
4605                 if ((updelay % miimon) != 0) {
4606                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4607                                 updelay, miimon, (updelay / miimon) * miimon);
4608                 }
4609
4610                 updelay /= miimon;
4611
4612                 if ((downdelay % miimon) != 0) {
4613                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4614                                 downdelay, miimon,
4615                                 (downdelay / miimon) * miimon);
4616                 }
4617
4618                 downdelay /= miimon;
4619         }
4620
4621         if (arp_interval < 0) {
4622                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4623                         arp_interval, INT_MAX);
4624                 arp_interval = 0;
4625         }
4626
4627         for (arp_ip_count = 0, i = 0;
4628              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4629                 __be32 ip;
4630
4631                 /* not a complete check, but good enough to catch mistakes */
4632                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4633                     !bond_is_ip_target_ok(ip)) {
4634                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4635                                 arp_ip_target[i]);
4636                         arp_interval = 0;
4637                 } else {
4638                         if (bond_get_targets_ip(arp_target, ip) == -1)
4639                                 arp_target[arp_ip_count++] = ip;
4640                         else
4641                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4642                                         &ip);
4643                 }
4644         }
4645
4646         if (arp_interval && !arp_ip_count) {
4647                 /* don't allow arping if no arp_ip_target given... */
4648                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4649                         arp_interval);
4650                 arp_interval = 0;
4651         }
4652
4653         if (arp_validate) {
4654                 if (!arp_interval) {
4655                         pr_err("arp_validate requires arp_interval\n");
4656                         return -EINVAL;
4657                 }
4658
4659                 bond_opt_initstr(&newval, arp_validate);
4660                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4661                                         &newval);
4662                 if (!valptr) {
4663                         pr_err("Error: invalid arp_validate \"%s\"\n",
4664                                arp_validate);
4665                         return -EINVAL;
4666                 }
4667                 arp_validate_value = valptr->value;
4668         } else {
4669                 arp_validate_value = 0;
4670         }
4671
4672         if (arp_all_targets) {
4673                 bond_opt_initstr(&newval, arp_all_targets);
4674                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4675                                         &newval);
4676                 if (!valptr) {
4677                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4678                                arp_all_targets);
4679                         arp_all_targets_value = 0;
4680                 } else {
4681                         arp_all_targets_value = valptr->value;
4682                 }
4683         }
4684
4685         if (miimon) {
4686                 pr_info("MII link monitoring set to %d ms\n", miimon);
4687         } else if (arp_interval) {
4688                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4689                                           arp_validate_value);
4690                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4691                         arp_interval, valptr->string, arp_ip_count);
4692
4693                 for (i = 0; i < arp_ip_count; i++)
4694                         pr_cont(" %s", arp_ip_target[i]);
4695
4696                 pr_cont("\n");
4697
4698         } else if (max_bonds) {
4699                 /* miimon and arp_interval not set, we need one so things
4700                  * work as expected, see bonding.txt for details
4701                  */
4702                 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");
4703         }
4704
4705         if (primary && !bond_mode_uses_primary(bond_mode)) {
4706                 /* currently, using a primary only makes sense
4707                  * in active backup, TLB or ALB modes
4708                  */
4709                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4710                         primary, bond_mode_name(bond_mode));
4711                 primary = NULL;
4712         }
4713
4714         if (primary && primary_reselect) {
4715                 bond_opt_initstr(&newval, primary_reselect);
4716                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4717                                         &newval);
4718                 if (!valptr) {
4719                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4720                                primary_reselect);
4721                         return -EINVAL;
4722                 }
4723                 primary_reselect_value = valptr->value;
4724         } else {
4725                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4726         }
4727
4728         if (fail_over_mac) {
4729                 bond_opt_initstr(&newval, fail_over_mac);
4730                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4731                                         &newval);
4732                 if (!valptr) {
4733                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4734                                fail_over_mac);
4735                         return -EINVAL;
4736                 }
4737                 fail_over_mac_value = valptr->value;
4738                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4739                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4740         } else {
4741                 fail_over_mac_value = BOND_FOM_NONE;
4742         }
4743
4744         bond_opt_initstr(&newval, "default");
4745         valptr = bond_opt_parse(
4746                         bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4747                                      &newval);
4748         if (!valptr) {
4749                 pr_err("Error: No ad_actor_sys_prio default value");
4750                 return -EINVAL;
4751         }
4752         ad_actor_sys_prio = valptr->value;
4753
4754         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4755                                 &newval);
4756         if (!valptr) {
4757                 pr_err("Error: No ad_user_port_key default value");
4758                 return -EINVAL;
4759         }
4760         ad_user_port_key = valptr->value;
4761
4762         bond_opt_initstr(&newval, "default");
4763         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
4764         if (!valptr) {
4765                 pr_err("Error: No tlb_dynamic_lb default value");
4766                 return -EINVAL;
4767         }
4768         tlb_dynamic_lb = valptr->value;
4769
4770         if (lp_interval == 0) {
4771                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4772                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4773                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4774         }
4775
4776         /* fill params struct with the proper values */
4777         params->mode = bond_mode;
4778         params->xmit_policy = xmit_hashtype;
4779         params->miimon = miimon;
4780         params->num_peer_notif = num_peer_notif;
4781         params->arp_interval = arp_interval;
4782         params->arp_validate = arp_validate_value;
4783         params->arp_all_targets = arp_all_targets_value;
4784         params->updelay = updelay;
4785         params->downdelay = downdelay;
4786         params->use_carrier = use_carrier;
4787         params->lacp_fast = lacp_fast;
4788         params->primary[0] = 0;
4789         params->primary_reselect = primary_reselect_value;
4790         params->fail_over_mac = fail_over_mac_value;
4791         params->tx_queues = tx_queues;
4792         params->all_slaves_active = all_slaves_active;
4793         params->resend_igmp = resend_igmp;
4794         params->min_links = min_links;
4795         params->lp_interval = lp_interval;
4796         params->packets_per_slave = packets_per_slave;
4797         params->tlb_dynamic_lb = tlb_dynamic_lb;
4798         params->ad_actor_sys_prio = ad_actor_sys_prio;
4799         eth_zero_addr(params->ad_actor_system);
4800         params->ad_user_port_key = ad_user_port_key;
4801         if (packets_per_slave > 0) {
4802                 params->reciprocal_packets_per_slave =
4803                         reciprocal_value(packets_per_slave);
4804         } else {
4805                 /* reciprocal_packets_per_slave is unused if
4806                  * packets_per_slave is 0 or 1, just initialize it
4807                  */
4808                 params->reciprocal_packets_per_slave =
4809                         (struct reciprocal_value) { 0 };
4810         }
4811
4812         if (primary) {
4813                 strncpy(params->primary, primary, IFNAMSIZ);
4814                 params->primary[IFNAMSIZ - 1] = 0;
4815         }
4816
4817         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4818
4819         return 0;
4820 }
4821
4822 /* Called from registration process */
4823 static int bond_init(struct net_device *bond_dev)
4824 {
4825         struct bonding *bond = netdev_priv(bond_dev);
4826         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4827
4828         netdev_dbg(bond_dev, "Begin bond_init\n");
4829
4830         bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
4831         if (!bond->wq)
4832                 return -ENOMEM;
4833
4834         bond->nest_level = SINGLE_DEPTH_NESTING;
4835         netdev_lockdep_set_classes(bond_dev);
4836
4837         list_add_tail(&bond->bond_list, &bn->dev_list);
4838
4839         bond_prepare_sysfs_group(bond);
4840
4841         bond_debug_register(bond);
4842
4843         /* Ensure valid dev_addr */
4844         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4845             bond_dev->addr_assign_type == NET_ADDR_PERM)
4846                 eth_hw_addr_random(bond_dev);
4847
4848         return 0;
4849 }
4850
4851 unsigned int bond_get_num_tx_queues(void)
4852 {
4853         return tx_queues;
4854 }
4855
4856 /* Create a new bond based on the specified name and bonding parameters.
4857  * If name is NULL, obtain a suitable "bond%d" name for us.
4858  * Caller must NOT hold rtnl_lock; we need to release it here before we
4859  * set up our sysfs entries.
4860  */
4861 int bond_create(struct net *net, const char *name)
4862 {
4863         struct net_device *bond_dev;
4864         struct bonding *bond;
4865         struct alb_bond_info *bond_info;
4866         int res;
4867
4868         rtnl_lock();
4869
4870         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4871                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
4872                                    bond_setup, tx_queues);
4873         if (!bond_dev) {
4874                 pr_err("%s: eek! can't alloc netdev!\n", name);
4875                 rtnl_unlock();
4876                 return -ENOMEM;
4877         }
4878
4879         /*
4880          * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4881          * It is set to 0 by default which is wrong.
4882          */
4883         bond = netdev_priv(bond_dev);
4884         bond_info = &(BOND_ALB_INFO(bond));
4885         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4886
4887         dev_net_set(bond_dev, net);
4888         bond_dev->rtnl_link_ops = &bond_link_ops;
4889
4890         res = register_netdevice(bond_dev);
4891         if (res < 0) {
4892                 free_netdev(bond_dev);
4893                 rtnl_unlock();
4894
4895                 return res;
4896         }
4897
4898         netif_carrier_off(bond_dev);
4899
4900         bond_work_init_all(bond);
4901
4902         rtnl_unlock();
4903         return 0;
4904 }
4905
4906 static int __net_init bond_net_init(struct net *net)
4907 {
4908         struct bond_net *bn = net_generic(net, bond_net_id);
4909
4910         bn->net = net;
4911         INIT_LIST_HEAD(&bn->dev_list);
4912
4913         bond_create_proc_dir(bn);
4914         bond_create_sysfs(bn);
4915
4916         return 0;
4917 }
4918
4919 static void __net_exit bond_net_exit(struct net *net)
4920 {
4921         struct bond_net *bn = net_generic(net, bond_net_id);
4922         struct bonding *bond, *tmp_bond;
4923         LIST_HEAD(list);
4924
4925         bond_destroy_sysfs(bn);
4926
4927         /* Kill off any bonds created after unregistering bond rtnl ops */
4928         rtnl_lock();
4929         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4930                 unregister_netdevice_queue(bond->dev, &list);
4931         unregister_netdevice_many(&list);
4932         rtnl_unlock();
4933
4934         bond_destroy_proc_dir(bn);
4935 }
4936
4937 static struct pernet_operations bond_net_ops = {
4938         .init = bond_net_init,
4939         .exit = bond_net_exit,
4940         .id   = &bond_net_id,
4941         .size = sizeof(struct bond_net),
4942 };
4943
4944 static int __init bonding_init(void)
4945 {
4946         int i;
4947         int res;
4948
4949         pr_info("%s", bond_version);
4950
4951         res = bond_check_params(&bonding_defaults);
4952         if (res)
4953                 goto out;
4954
4955         res = register_pernet_subsys(&bond_net_ops);
4956         if (res)
4957                 goto out;
4958
4959         res = bond_netlink_init();
4960         if (res)
4961                 goto err_link;
4962
4963         bond_create_debugfs();
4964
4965         for (i = 0; i < max_bonds; i++) {
4966                 res = bond_create(&init_net, NULL);
4967                 if (res)
4968                         goto err;
4969         }
4970
4971         register_netdevice_notifier(&bond_netdev_notifier);
4972 out:
4973         return res;
4974 err:
4975         bond_destroy_debugfs();
4976         bond_netlink_fini();
4977 err_link:
4978         unregister_pernet_subsys(&bond_net_ops);
4979         goto out;
4980
4981 }
4982
4983 static void __exit bonding_exit(void)
4984 {
4985         unregister_netdevice_notifier(&bond_netdev_notifier);
4986
4987         bond_destroy_debugfs();
4988
4989         bond_netlink_fini();
4990         unregister_pernet_subsys(&bond_net_ops);
4991
4992 #ifdef CONFIG_NET_POLL_CONTROLLER
4993         /* Make sure we don't have an imbalance on our netpoll blocking */
4994         WARN_ON(atomic_read(&netpoll_block_tx));
4995 #endif
4996 }
4997
4998 module_init(bonding_init);
4999 module_exit(bonding_exit);
5000 MODULE_LICENSE("GPL");
5001 MODULE_VERSION(DRV_VERSION);
5002 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5003 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");