drivers/net/bonding/bond_main.c

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