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