drivers/net/bonding/bond_main.c

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