drivers/net/bonding/bond_main.c

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