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->addr_len=%d\n",
 624                    bond_dev, slave_dev, 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
 837                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 838                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 839
 840                         if (bond_is_lb(bond))
 841                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 842                 } else {
 843                         if (bond_uses_primary(bond)) {
 844                                 netdev_info(bond->dev, "making interface %s the new active one\n",
 845                                             new_active->dev->name);
 846                         }
 847                 }
 848         }
 849
 850         if (bond_uses_primary(bond))
 851                 bond_hw_addr_swap(bond, new_active, old_active);
 852
 853         if (bond_is_lb(bond)) {
 854                 bond_alb_handle_active_change(bond, new_active);
 855                 if (old_active)
 856                         bond_set_slave_inactive_flags(old_active,
 857                                                       BOND_SLAVE_NOTIFY_NOW);
 858                 if (new_active)
 859                         bond_set_slave_active_flags(new_active,
 860                                                     BOND_SLAVE_NOTIFY_NOW);
 861         } else {
 862                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 863         }
 864
 865         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
 866                 if (old_active)
 867                         bond_set_slave_inactive_flags(old_active,
 868                                                       BOND_SLAVE_NOTIFY_NOW);
 869
 870                 if (new_active) {
 871                         bool should_notify_peers = false;
 872
 873                         bond_set_slave_active_flags(new_active,
 874                                                     BOND_SLAVE_NOTIFY_NOW);
 875
 876                         if (bond->params.fail_over_mac)
 877                                 bond_do_fail_over_mac(bond, new_active,
 878                                                       old_active);
 879
 880                         if (netif_running(bond->dev)) {
 881                                 bond->send_peer_notif =
 882                                         bond->params.num_peer_notif;
 883                                 should_notify_peers =
 884                                         bond_should_notify_peers(bond);
 885                         }
 886
 887                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 888                         if (should_notify_peers)
 889                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 890                                                          bond->dev);
 891                 }
 892         }
 893
 894         /* resend IGMP joins since active slave has changed or
 895          * all were sent on curr_active_slave.
 896          * resend only if bond is brought up with the affected
 897          * bonding modes and the retransmission is enabled
 898          */
 899         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 900             ((bond_uses_primary(bond) && new_active) ||
 901              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
 902                 bond->igmp_retrans = bond->params.resend_igmp;
 903                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 904         }
 905 }
 906
 907 /**
 908  * bond_select_active_slave - select a new active slave, if needed
 909  * @bond: our bonding struct
 910  *
 911  * This functions should be called when one of the following occurs:
 912  * - The old curr_active_slave has been released or lost its link.
 913  * - The primary_slave has got its link back.
 914  * - A slave has got its link back and there's no old curr_active_slave.
 915  *
 916  * Caller must hold RTNL.
 917  */
 918 void bond_select_active_slave(struct bonding *bond)
 919 {
 920         struct slave *best_slave;
 921         int rv;
 922
 923         ASSERT_RTNL();
 924
 925         best_slave = bond_find_best_slave(bond);
 926         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
 927                 bond_change_active_slave(bond, best_slave);
 928                 rv = bond_set_carrier(bond);
 929                 if (!rv)
 930                         return;
 931
 932                 if (netif_carrier_ok(bond->dev)) {
 933                         netdev_info(bond->dev, "first active interface up!\n");
 934                 } else {
 935                         netdev_info(bond->dev, "now running without any active interface!\n");
 936                 }
 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_ALL_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_ALL_CSUM | NETIF_F_SG | NETIF_F_RXCSUM |\
1076                                  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
1186         skb->dev = bond->dev;
1187
1188         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1189             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1190             skb->pkt_type == PACKET_HOST) {
1191
1192                 if (unlikely(skb_cow_head(skb,
1193                                           skb->data - skb_mac_header(skb)))) {
1194                         kfree_skb(skb);
1195                         return RX_HANDLER_CONSUMED;
1196                 }
1197                 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1198         }
1199
1200         return ret;
1201 }
1202
1203 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1204                                       struct net_device *slave_dev,
1205                                       struct slave *slave)
1206 {
1207         int err;
1208
1209         err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1210         if (err)
1211                 return err;
1212         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1213         return 0;
1214 }
1215
1216 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1217                                   struct net_device *slave_dev)
1218 {
1219         netdev_upper_dev_unlink(slave_dev, bond_dev);
1220         slave_dev->flags &= ~IFF_SLAVE;
1221         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1222 }
1223
1224 static struct slave *bond_alloc_slave(struct bonding *bond)
1225 {
1226         struct slave *slave = NULL;
1227
1228         slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1229         if (!slave)
1230                 return NULL;
1231
1232         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1233                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1234                                                GFP_KERNEL);
1235                 if (!SLAVE_AD_INFO(slave)) {
1236                         kfree(slave);
1237                         return NULL;
1238                 }
1239         }
1240         return slave;
1241 }
1242
1243 static void bond_free_slave(struct slave *slave)
1244 {
1245         struct bonding *bond = bond_get_bond_by_slave(slave);
1246
1247         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1248                 kfree(SLAVE_AD_INFO(slave));
1249
1250         kfree(slave);
1251 }
1252
1253 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1254 {
1255         info->bond_mode = BOND_MODE(bond);
1256         info->miimon = bond->params.miimon;
1257         info->num_slaves = bond->slave_cnt;
1258 }
1259
1260 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1261 {
1262         strcpy(info->slave_name, slave->dev->name);
1263         info->link = slave->link;
1264         info->state = bond_slave_state(slave);
1265         info->link_failure_count = slave->link_failure_count;
1266 }
1267
1268 static void bond_netdev_notify(struct net_device *dev,
1269                                struct netdev_bonding_info *info)
1270 {
1271         rtnl_lock();
1272         netdev_bonding_info_change(dev, info);
1273         rtnl_unlock();
1274 }
1275
1276 static void bond_netdev_notify_work(struct work_struct *_work)
1277 {
1278         struct netdev_notify_work *w =
1279                 container_of(_work, struct netdev_notify_work, work.work);
1280
1281         bond_netdev_notify(w->dev, &w->bonding_info);
1282         dev_put(w->dev);
1283         kfree(w);
1284 }
1285
1286 void bond_queue_slave_event(struct slave *slave)
1287 {
1288         struct bonding *bond = slave->bond;
1289         struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1290
1291         if (!nnw)
1292                 return;
1293
1294         dev_hold(slave->dev);
1295         nnw->dev = slave->dev;
1296         bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1297         bond_fill_ifbond(bond, &nnw->bonding_info.master);
1298         INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1299
1300         queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1301 }
1302
1303 /* enslave device <slave> to bond device <master> */
1304 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1305 {
1306         struct bonding *bond = netdev_priv(bond_dev);
1307         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1308         struct slave *new_slave = NULL, *prev_slave;
1309         struct sockaddr addr;
1310         int link_reporting;
1311         int res = 0, i;
1312
1313         if (!bond->params.use_carrier &&
1314             slave_dev->ethtool_ops->get_link == NULL &&
1315             slave_ops->ndo_do_ioctl == NULL) {
1316                 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1317                             slave_dev->name);
1318         }
1319
1320         /* already enslaved */
1321         if (slave_dev->flags & IFF_SLAVE) {
1322                 netdev_dbg(bond_dev, "Error: Device was already enslaved\n");
1323                 return -EBUSY;
1324         }
1325
1326         if (bond_dev == slave_dev) {
1327                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1328                 return -EPERM;
1329         }
1330
1331         /* vlan challenged mutual exclusion */
1332         /* no need to lock since we're protected by rtnl_lock */
1333         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1334                 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1335                            slave_dev->name);
1336                 if (vlan_uses_dev(bond_dev)) {
1337                         netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1338                                    slave_dev->name, bond_dev->name);
1339                         return -EPERM;
1340                 } else {
1341                         netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1342                                     slave_dev->name, slave_dev->name,
1343                                     bond_dev->name);
1344                 }
1345         } else {
1346                 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1347                            slave_dev->name);
1348         }
1349
1350         /* Old ifenslave binaries are no longer supported.  These can
1351          * be identified with moderate accuracy by the state of the slave:
1352          * the current ifenslave will set the interface down prior to
1353          * enslaving it; the old ifenslave will not.
1354          */
1355         if ((slave_dev->flags & IFF_UP)) {
1356                 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1357                            slave_dev->name);
1358                 res = -EPERM;
1359                 goto err_undo_flags;
1360         }
1361
1362         /* set bonding device ether type by slave - bonding netdevices are
1363          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1364          * there is a need to override some of the type dependent attribs/funcs.
1365          *
1366          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1367          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1368          */
1369         if (!bond_has_slaves(bond)) {
1370                 if (bond_dev->type != slave_dev->type) {
1371                         netdev_dbg(bond_dev, "change device type from %d to %d\n",
1372                                    bond_dev->type, slave_dev->type);
1373
1374                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1375                                                        bond_dev);
1376                         res = notifier_to_errno(res);
1377                         if (res) {
1378                                 netdev_err(bond_dev, "refused to change device type\n");
1379                                 res = -EBUSY;
1380                                 goto err_undo_flags;
1381                         }
1382
1383                         /* Flush unicast and multicast addresses */
1384                         dev_uc_flush(bond_dev);
1385                         dev_mc_flush(bond_dev);
1386
1387                         if (slave_dev->type != ARPHRD_ETHER)
1388                                 bond_setup_by_slave(bond_dev, slave_dev);
1389                         else {
1390                                 ether_setup(bond_dev);
1391                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1392                         }
1393
1394                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1395                                                  bond_dev);
1396                 }
1397         } else if (bond_dev->type != slave_dev->type) {
1398                 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1399                            slave_dev->name, slave_dev->type, bond_dev->type);
1400                 res = -EINVAL;
1401                 goto err_undo_flags;
1402         }
1403
1404         if (slave_ops->ndo_set_mac_address == NULL) {
1405                 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1406                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1407                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1408                         if (!bond_has_slaves(bond)) {
1409                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1410                                 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1411                         } else {
1412                                 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");
1413                                 res = -EOPNOTSUPP;
1414                                 goto err_undo_flags;
1415                         }
1416                 }
1417         }
1418
1419         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1420
1421         /* If this is the first slave, then we need to set the master's hardware
1422          * address to be the same as the slave's.
1423          */
1424         if (!bond_has_slaves(bond) &&
1425             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1426                 bond_set_dev_addr(bond->dev, slave_dev);
1427
1428         new_slave = bond_alloc_slave(bond);
1429         if (!new_slave) {
1430                 res = -ENOMEM;
1431                 goto err_undo_flags;
1432         }
1433
1434         new_slave->bond = bond;
1435         new_slave->dev = slave_dev;
1436         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1437          * is set via sysfs or module option if desired.
1438          */
1439         new_slave->queue_id = 0;
1440
1441         /* Save slave's original mtu and then set it to match the bond */
1442         new_slave->original_mtu = slave_dev->mtu;
1443         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1444         if (res) {
1445                 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1446                 goto err_free;
1447         }
1448
1449         /* Save slave's original ("permanent") mac address for modes
1450          * that need it, and for restoring it upon release, and then
1451          * set it to the master's address
1452          */
1453         ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1454
1455         if (!bond->params.fail_over_mac ||
1456             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1457                 /* Set slave to master's mac address.  The application already
1458                  * set the master's mac address to that of the first slave
1459                  */
1460                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1461                 addr.sa_family = slave_dev->type;
1462                 res = dev_set_mac_address(slave_dev, &addr);
1463                 if (res) {
1464                         netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1465                         goto err_restore_mtu;
1466                 }
1467         }
1468
1469         /* set slave flag before open to prevent IPv6 addrconf */
1470         slave_dev->flags |= IFF_SLAVE;
1471
1472         /* open the slave since the application closed it */
1473         res = dev_open(slave_dev);
1474         if (res) {
1475                 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1476                 goto err_restore_mac;
1477         }
1478
1479         slave_dev->priv_flags |= IFF_BONDING;
1480         /* initialize slave stats */
1481         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1482
1483         if (bond_is_lb(bond)) {
1484                 /* bond_alb_init_slave() must be called before all other stages since
1485                  * it might fail and we do not want to have to undo everything
1486                  */
1487                 res = bond_alb_init_slave(bond, new_slave);
1488                 if (res)
1489                         goto err_close;
1490         }
1491
1492         /* If the mode uses primary, then the following is handled by
1493          * bond_change_active_slave().
1494          */
1495         if (!bond_uses_primary(bond)) {
1496                 /* set promiscuity level to new slave */
1497                 if (bond_dev->flags & IFF_PROMISC) {
1498                         res = dev_set_promiscuity(slave_dev, 1);
1499                         if (res)
1500                                 goto err_close;
1501                 }
1502
1503                 /* set allmulti level to new slave */
1504                 if (bond_dev->flags & IFF_ALLMULTI) {
1505                         res = dev_set_allmulti(slave_dev, 1);
1506                         if (res)
1507                                 goto err_close;
1508                 }
1509
1510                 netif_addr_lock_bh(bond_dev);
1511
1512                 dev_mc_sync_multiple(slave_dev, bond_dev);
1513                 dev_uc_sync_multiple(slave_dev, bond_dev);
1514
1515                 netif_addr_unlock_bh(bond_dev);
1516         }
1517
1518         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1519                 /* add lacpdu mc addr to mc list */
1520                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1521
1522                 dev_mc_add(slave_dev, lacpdu_multicast);
1523         }
1524
1525         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1526         if (res) {
1527                 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1528                            slave_dev->name);
1529                 goto err_close;
1530         }
1531
1532         prev_slave = bond_last_slave(bond);
1533
1534         new_slave->delay = 0;
1535         new_slave->link_failure_count = 0;
1536
1537         bond_update_speed_duplex(new_slave);
1538
1539         new_slave->last_rx = jiffies -
1540                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1541         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1542                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1543
1544         if (bond->params.miimon && !bond->params.use_carrier) {
1545                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1546
1547                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1548                         /* miimon is set but a bonded network driver
1549                          * does not support ETHTOOL/MII and
1550                          * arp_interval is not set.  Note: if
1551                          * use_carrier is enabled, we will never go
1552                          * here (because netif_carrier is always
1553                          * supported); thus, we don't need to change
1554                          * the messages for netif_carrier.
1555                          */
1556                         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",
1557                                     slave_dev->name);
1558                 } else if (link_reporting == -1) {
1559                         /* unable get link status using mii/ethtool */
1560                         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",
1561                                     slave_dev->name);
1562                 }
1563         }
1564
1565         /* check for initial state */
1566         if (bond->params.miimon) {
1567                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1568                         if (bond->params.updelay) {
1569                                 bond_set_slave_link_state(new_slave,
1570                                                           BOND_LINK_BACK);
1571                                 new_slave->delay = bond->params.updelay;
1572                         } else {
1573                                 bond_set_slave_link_state(new_slave,
1574                                                           BOND_LINK_UP);
1575                         }
1576                 } else {
1577                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN);
1578                 }
1579         } else if (bond->params.arp_interval) {
1580                 bond_set_slave_link_state(new_slave,
1581                                           (netif_carrier_ok(slave_dev) ?
1582                                           BOND_LINK_UP : BOND_LINK_DOWN));
1583         } else {
1584                 bond_set_slave_link_state(new_slave, BOND_LINK_UP);
1585         }
1586
1587         if (new_slave->link != BOND_LINK_DOWN)
1588                 new_slave->last_link_up = jiffies;
1589         netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1590                    new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1591                    (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1592
1593         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1594                 /* if there is a primary slave, remember it */
1595                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1596                         rcu_assign_pointer(bond->primary_slave, new_slave);
1597                         bond->force_primary = true;
1598                 }
1599         }
1600
1601         switch (BOND_MODE(bond)) {
1602         case BOND_MODE_ACTIVEBACKUP:
1603                 bond_set_slave_inactive_flags(new_slave,
1604                                               BOND_SLAVE_NOTIFY_NOW);
1605                 break;
1606         case BOND_MODE_8023AD:
1607                 /* in 802.3ad mode, the internal mechanism
1608                  * will activate the slaves in the selected
1609                  * aggregator
1610                  */
1611                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1612                 /* if this is the first slave */
1613                 if (!prev_slave) {
1614                         SLAVE_AD_INFO(new_slave)->id = 1;
1615                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1616                          * can be called only after the mac address of the bond is set
1617                          */
1618                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1619                 } else {
1620                         SLAVE_AD_INFO(new_slave)->id =
1621                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1622                 }
1623
1624                 bond_3ad_bind_slave(new_slave);
1625                 break;
1626         case BOND_MODE_TLB:
1627         case BOND_MODE_ALB:
1628                 bond_set_active_slave(new_slave);
1629                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1630                 break;
1631         default:
1632                 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1633
1634                 /* always active in trunk mode */
1635                 bond_set_active_slave(new_slave);
1636
1637                 /* In trunking mode there is little meaning to curr_active_slave
1638                  * anyway (it holds no special properties of the bond device),
1639                  * so we can change it without calling change_active_interface()
1640                  */
1641                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1642                     new_slave->link == BOND_LINK_UP)
1643                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1644
1645                 break;
1646         } /* switch(bond_mode) */
1647
1648 #ifdef CONFIG_NET_POLL_CONTROLLER
1649         slave_dev->npinfo = bond->dev->npinfo;
1650         if (slave_dev->npinfo) {
1651                 if (slave_enable_netpoll(new_slave)) {
1652                         netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1653                         res = -EBUSY;
1654                         goto err_detach;
1655                 }
1656         }
1657 #endif
1658
1659         if (!(bond_dev->features & NETIF_F_LRO))
1660                 dev_disable_lro(slave_dev);
1661
1662         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1663                                          new_slave);
1664         if (res) {
1665                 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1666                 goto err_detach;
1667         }
1668
1669         res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1670         if (res) {
1671                 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1672                 goto err_unregister;
1673         }
1674
1675         res = bond_sysfs_slave_add(new_slave);
1676         if (res) {
1677                 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1678                 goto err_upper_unlink;
1679         }
1680
1681         bond->slave_cnt++;
1682         bond_compute_features(bond);
1683         bond_set_carrier(bond);
1684
1685         if (bond_uses_primary(bond)) {
1686                 block_netpoll_tx();
1687                 bond_select_active_slave(bond);
1688                 unblock_netpoll_tx();
1689         }
1690
1691         if (bond_mode_uses_xmit_hash(bond))
1692                 bond_update_slave_arr(bond, NULL);
1693
1694         netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1695                     slave_dev->name,
1696                     bond_is_active_slave(new_slave) ? "an active" : "a backup",
1697                     new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1698
1699         /* enslave is successful */
1700         bond_queue_slave_event(new_slave);
1701         return 0;
1702
1703 /* Undo stages on error */
1704 err_upper_unlink:
1705         bond_upper_dev_unlink(bond_dev, slave_dev);
1706
1707 err_unregister:
1708         netdev_rx_handler_unregister(slave_dev);
1709
1710 err_detach:
1711         if (!bond_uses_primary(bond))
1712                 bond_hw_addr_flush(bond_dev, slave_dev);
1713
1714         vlan_vids_del_by_dev(slave_dev, bond_dev);
1715         if (rcu_access_pointer(bond->primary_slave) == new_slave)
1716                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1717         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1718                 block_netpoll_tx();
1719                 bond_change_active_slave(bond, NULL);
1720                 bond_select_active_slave(bond);
1721                 unblock_netpoll_tx();
1722         }
1723         /* either primary_slave or curr_active_slave might've changed */
1724         synchronize_rcu();
1725         slave_disable_netpoll(new_slave);
1726
1727 err_close:
1728         slave_dev->priv_flags &= ~IFF_BONDING;
1729         dev_close(slave_dev);
1730
1731 err_restore_mac:
1732         slave_dev->flags &= ~IFF_SLAVE;
1733         if (!bond->params.fail_over_mac ||
1734             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1735                 /* XXX TODO - fom follow mode needs to change master's
1736                  * MAC if this slave's MAC is in use by the bond, or at
1737                  * least print a warning.
1738                  */
1739                 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1740                 addr.sa_family = slave_dev->type;
1741                 dev_set_mac_address(slave_dev, &addr);
1742         }
1743
1744 err_restore_mtu:
1745         dev_set_mtu(slave_dev, new_slave->original_mtu);
1746
1747 err_free:
1748         bond_free_slave(new_slave);
1749
1750 err_undo_flags:
1751         /* Enslave of first slave has failed and we need to fix master's mac */
1752         if (!bond_has_slaves(bond)) {
1753                 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1754                                             slave_dev->dev_addr))
1755                         eth_hw_addr_random(bond_dev);
1756                 if (bond_dev->type != ARPHRD_ETHER) {
1757                         dev_close(bond_dev);
1758                         ether_setup(bond_dev);
1759                         bond_dev->flags |= IFF_MASTER;
1760                         bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1761                 }
1762         }
1763
1764         return res;
1765 }
1766
1767 /* Try to release the slave device <slave> from the bond device <master>
1768  * It is legal to access curr_active_slave without a lock because all the function
1769  * is RTNL-locked. If "all" is true it means that the function is being called
1770  * while destroying a bond interface and all slaves are being released.
1771  *
1772  * The rules for slave state should be:
1773  *   for Active/Backup:
1774  *     Active stays on all backups go down
1775  *   for Bonded connections:
1776  *     The first up interface should be left on and all others downed.
1777  */
1778 static int __bond_release_one(struct net_device *bond_dev,
1779                               struct net_device *slave_dev,
1780                               bool all)
1781 {
1782         struct bonding *bond = netdev_priv(bond_dev);
1783         struct slave *slave, *oldcurrent;
1784         struct sockaddr addr;
1785         int old_flags = bond_dev->flags;
1786         netdev_features_t old_features = bond_dev->features;
1787
1788         /* slave is not a slave or master is not master of this slave */
1789         if (!(slave_dev->flags & IFF_SLAVE) ||
1790             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1791                 netdev_dbg(bond_dev, "cannot release %s\n",
1792                            slave_dev->name);
1793                 return -EINVAL;
1794         }
1795
1796         block_netpoll_tx();
1797
1798         slave = bond_get_slave_by_dev(bond, slave_dev);
1799         if (!slave) {
1800                 /* not a slave of this bond */
1801                 netdev_info(bond_dev, "%s not enslaved\n",
1802                             slave_dev->name);
1803                 unblock_netpoll_tx();
1804                 return -EINVAL;
1805         }
1806
1807         bond_sysfs_slave_del(slave);
1808
1809         /* recompute stats just before removing the slave */
1810         bond_get_stats(bond->dev, &bond->bond_stats);
1811
1812         bond_upper_dev_unlink(bond_dev, slave_dev);
1813         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1814          * for this slave anymore.
1815          */
1816         netdev_rx_handler_unregister(slave_dev);
1817
1818         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1819                 bond_3ad_unbind_slave(slave);
1820
1821         if (bond_mode_uses_xmit_hash(bond))
1822                 bond_update_slave_arr(bond, slave);
1823
1824         netdev_info(bond_dev, "Releasing %s interface %s\n",
1825                     bond_is_active_slave(slave) ? "active" : "backup",
1826                     slave_dev->name);
1827
1828         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1829
1830         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1831
1832         if (!all && (!bond->params.fail_over_mac ||
1833                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1834                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1835                     bond_has_slaves(bond))
1836                         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",
1837                                     slave_dev->name, slave->perm_hwaddr,
1838                                     bond_dev->name, slave_dev->name);
1839         }
1840
1841         if (rtnl_dereference(bond->primary_slave) == slave)
1842                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1843
1844         if (oldcurrent == slave)
1845                 bond_change_active_slave(bond, NULL);
1846
1847         if (bond_is_lb(bond)) {
1848                 /* Must be called only after the slave has been
1849                  * detached from the list and the curr_active_slave
1850                  * has been cleared (if our_slave == old_current),
1851                  * but before a new active slave is selected.
1852                  */
1853                 bond_alb_deinit_slave(bond, slave);
1854         }
1855
1856         if (all) {
1857                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1858         } else if (oldcurrent == slave) {
1859                 /* Note that we hold RTNL over this sequence, so there
1860                  * is no concern that another slave add/remove event
1861                  * will interfere.
1862                  */
1863                 bond_select_active_slave(bond);
1864         }
1865
1866         if (!bond_has_slaves(bond)) {
1867                 bond_set_carrier(bond);
1868                 eth_hw_addr_random(bond_dev);
1869         }
1870
1871         unblock_netpoll_tx();
1872         synchronize_rcu();
1873         bond->slave_cnt--;
1874
1875         if (!bond_has_slaves(bond)) {
1876                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1877                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1878         }
1879
1880         bond_compute_features(bond);
1881         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1882             (old_features & NETIF_F_VLAN_CHALLENGED))
1883                 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1884                             slave_dev->name, bond_dev->name);
1885
1886         vlan_vids_del_by_dev(slave_dev, bond_dev);
1887
1888         /* If the mode uses primary, then this case was handled above by
1889          * bond_change_active_slave(..., NULL)
1890          */
1891         if (!bond_uses_primary(bond)) {
1892                 /* unset promiscuity level from slave
1893                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1894                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1895                  * value of that flag before that change, as that was the value
1896                  * when this slave was attached, so we cache at the start of the
1897                  * function and use it here. Same goes for ALLMULTI below
1898                  */
1899                 if (old_flags & IFF_PROMISC)
1900                         dev_set_promiscuity(slave_dev, -1);
1901
1902                 /* unset allmulti level from slave */
1903                 if (old_flags & IFF_ALLMULTI)
1904                         dev_set_allmulti(slave_dev, -1);
1905
1906                 bond_hw_addr_flush(bond_dev, slave_dev);
1907         }
1908
1909         slave_disable_netpoll(slave);
1910
1911         /* close slave before restoring its mac address */
1912         dev_close(slave_dev);
1913
1914         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1915             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1916                 /* restore original ("permanent") mac address */
1917                 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1918                 addr.sa_family = slave_dev->type;
1919                 dev_set_mac_address(slave_dev, &addr);
1920         }
1921
1922         dev_set_mtu(slave_dev, slave->original_mtu);
1923
1924         slave_dev->priv_flags &= ~IFF_BONDING;
1925
1926         bond_free_slave(slave);
1927
1928         return 0;
1929 }
1930
1931 /* A wrapper used because of ndo_del_link */
1932 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1933 {
1934         return __bond_release_one(bond_dev, slave_dev, false);
1935 }
1936
1937 /* First release a slave and then destroy the bond if no more slaves are left.
1938  * Must be under rtnl_lock when this function is called.
1939  */
1940 static int  bond_release_and_destroy(struct net_device *bond_dev,
1941                                      struct net_device *slave_dev)
1942 {
1943         struct bonding *bond = netdev_priv(bond_dev);
1944         int ret;
1945
1946         ret = bond_release(bond_dev, slave_dev);
1947         if (ret == 0 && !bond_has_slaves(bond)) {
1948                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1949                 netdev_info(bond_dev, "Destroying bond %s\n",
1950                             bond_dev->name);
1951                 bond_remove_proc_entry(bond);
1952                 unregister_netdevice(bond_dev);
1953         }
1954         return ret;
1955 }
1956
1957 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1958 {
1959         struct bonding *bond = netdev_priv(bond_dev);
1960         bond_fill_ifbond(bond, info);
1961         return 0;
1962 }
1963
1964 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1965 {
1966         struct bonding *bond = netdev_priv(bond_dev);
1967         struct list_head *iter;
1968         int i = 0, res = -ENODEV;
1969         struct slave *slave;
1970
1971         bond_for_each_slave(bond, slave, iter) {
1972                 if (i++ == (int)info->slave_id) {
1973                         res = 0;
1974                         bond_fill_ifslave(slave, info);
1975                         break;
1976                 }
1977         }
1978
1979         return res;
1980 }
1981
1982 /*-------------------------------- Monitoring -------------------------------*/
1983
1984 /* called with rcu_read_lock() */
1985 static int bond_miimon_inspect(struct bonding *bond)
1986 {
1987         int link_state, commit = 0;
1988         struct list_head *iter;
1989         struct slave *slave;
1990         bool ignore_updelay;
1991
1992         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
1993
1994         bond_for_each_slave_rcu(bond, slave, iter) {
1995                 slave->new_link = BOND_LINK_NOCHANGE;
1996
1997                 link_state = bond_check_dev_link(bond, slave->dev, 0);
1998
1999                 switch (slave->link) {
2000                 case BOND_LINK_UP:
2001                         if (link_state)
2002                                 continue;
2003
2004                         bond_set_slave_link_state(slave, BOND_LINK_FAIL);
2005                         slave->delay = bond->params.downdelay;
2006                         if (slave->delay) {
2007                                 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2008                                             (BOND_MODE(bond) ==
2009                                              BOND_MODE_ACTIVEBACKUP) ?
2010                                              (bond_is_active_slave(slave) ?
2011                                               "active " : "backup ") : "",
2012                                             slave->dev->name,
2013                                             bond->params.downdelay * bond->params.miimon);
2014                         }
2015                         /*FALLTHRU*/
2016                 case BOND_LINK_FAIL:
2017                         if (link_state) {
2018                                 /* recovered before downdelay expired */
2019                                 bond_set_slave_link_state(slave, BOND_LINK_UP);
2020                                 slave->last_link_up = jiffies;
2021                                 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2022                                             (bond->params.downdelay - slave->delay) *
2023                                             bond->params.miimon,
2024                                             slave->dev->name);
2025                                 continue;
2026                         }
2027
2028                         if (slave->delay <= 0) {
2029                                 slave->new_link = BOND_LINK_DOWN;
2030                                 commit++;
2031                                 continue;
2032                         }
2033
2034                         slave->delay--;
2035                         break;
2036
2037                 case BOND_LINK_DOWN:
2038                         if (!link_state)
2039                                 continue;
2040
2041                         bond_set_slave_link_state(slave, BOND_LINK_BACK);
2042                         slave->delay = bond->params.updelay;
2043
2044                         if (slave->delay) {
2045                                 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2046                                             slave->dev->name,
2047                                             ignore_updelay ? 0 :
2048                                             bond->params.updelay *
2049                                             bond->params.miimon);
2050                         }
2051                         /*FALLTHRU*/
2052                 case BOND_LINK_BACK:
2053                         if (!link_state) {
2054                                 bond_set_slave_link_state(slave,
2055                                                           BOND_LINK_DOWN);
2056                                 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2057                                             (bond->params.updelay - slave->delay) *
2058                                             bond->params.miimon,
2059                                             slave->dev->name);
2060
2061                                 continue;
2062                         }
2063
2064                         if (ignore_updelay)
2065                                 slave->delay = 0;
2066
2067                         if (slave->delay <= 0) {
2068                                 slave->new_link = BOND_LINK_UP;
2069                                 commit++;
2070                                 ignore_updelay = false;
2071                                 continue;
2072                         }
2073
2074                         slave->delay--;
2075                         break;
2076                 }
2077         }
2078
2079         return commit;
2080 }
2081
2082 static void bond_miimon_commit(struct bonding *bond)
2083 {
2084         struct list_head *iter;
2085         struct slave *slave, *primary;
2086
2087         bond_for_each_slave(bond, slave, iter) {
2088                 switch (slave->new_link) {
2089                 case BOND_LINK_NOCHANGE:
2090                         continue;
2091
2092                 case BOND_LINK_UP:
2093                         bond_set_slave_link_state(slave, BOND_LINK_UP);
2094                         slave->last_link_up = jiffies;
2095
2096                         primary = rtnl_dereference(bond->primary_slave);
2097                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2098                                 /* prevent it from being the active one */
2099                                 bond_set_backup_slave(slave);
2100                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2101                                 /* make it immediately active */
2102                                 bond_set_active_slave(slave);
2103                         } else if (slave != primary) {
2104                                 /* prevent it from being the active one */
2105                                 bond_set_backup_slave(slave);
2106                         }
2107
2108                         netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2109                                     slave->dev->name,
2110                                     slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2111                                     slave->duplex ? "full" : "half");
2112
2113                         /* notify ad that the link status has changed */
2114                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2115                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2116
2117                         if (bond_is_lb(bond))
2118                                 bond_alb_handle_link_change(bond, slave,
2119                                                             BOND_LINK_UP);
2120
2121                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2122                                 bond_update_slave_arr(bond, NULL);
2123
2124                         if (!bond->curr_active_slave || slave == primary)
2125                                 goto do_failover;
2126
2127                         continue;
2128
2129                 case BOND_LINK_DOWN:
2130                         if (slave->link_failure_count < UINT_MAX)
2131                                 slave->link_failure_count++;
2132
2133                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2134
2135                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2136                             BOND_MODE(bond) == BOND_MODE_8023AD)
2137                                 bond_set_slave_inactive_flags(slave,
2138                                                               BOND_SLAVE_NOTIFY_NOW);
2139
2140                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2141                                     slave->dev->name);
2142
2143                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2144                                 bond_3ad_handle_link_change(slave,
2145                                                             BOND_LINK_DOWN);
2146
2147                         if (bond_is_lb(bond))
2148                                 bond_alb_handle_link_change(bond, slave,
2149                                                             BOND_LINK_DOWN);
2150
2151                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2152                                 bond_update_slave_arr(bond, NULL);
2153
2154                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2155                                 goto do_failover;
2156
2157                         continue;
2158
2159                 default:
2160                         netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2161                                    slave->new_link, slave->dev->name);
2162                         slave->new_link = BOND_LINK_NOCHANGE;
2163
2164                         continue;
2165                 }
2166
2167 do_failover:
2168                 block_netpoll_tx();
2169                 bond_select_active_slave(bond);
2170                 unblock_netpoll_tx();
2171         }
2172
2173         bond_set_carrier(bond);
2174 }
2175
2176 /* bond_mii_monitor
2177  *
2178  * Really a wrapper that splits the mii monitor into two phases: an
2179  * inspection, then (if inspection indicates something needs to be done)
2180  * an acquisition of appropriate locks followed by a commit phase to
2181  * implement whatever link state changes are indicated.
2182  */
2183 static void bond_mii_monitor(struct work_struct *work)
2184 {
2185         struct bonding *bond = container_of(work, struct bonding,
2186                                             mii_work.work);
2187         bool should_notify_peers = false;
2188         unsigned long delay;
2189
2190         delay = msecs_to_jiffies(bond->params.miimon);
2191
2192         if (!bond_has_slaves(bond))
2193                 goto re_arm;
2194
2195         rcu_read_lock();
2196
2197         should_notify_peers = bond_should_notify_peers(bond);
2198
2199         if (bond_miimon_inspect(bond)) {
2200                 rcu_read_unlock();
2201
2202                 /* Race avoidance with bond_close cancel of workqueue */
2203                 if (!rtnl_trylock()) {
2204                         delay = 1;
2205                         should_notify_peers = false;
2206                         goto re_arm;
2207                 }
2208
2209                 bond_miimon_commit(bond);
2210
2211                 rtnl_unlock();  /* might sleep, hold no other locks */
2212         } else
2213                 rcu_read_unlock();
2214
2215 re_arm:
2216         if (bond->params.miimon)
2217                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2218
2219         if (should_notify_peers) {
2220                 if (!rtnl_trylock())
2221                         return;
2222                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2223                 rtnl_unlock();
2224         }
2225 }
2226
2227 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2228 {
2229         struct net_device *upper;
2230         struct list_head *iter;
2231         bool ret = false;
2232
2233         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2234                 return true;
2235
2236         rcu_read_lock();
2237         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2238                 if (ip == bond_confirm_addr(upper, 0, ip)) {
2239                         ret = true;
2240                         break;
2241                 }
2242         }
2243         rcu_read_unlock();
2244
2245         return ret;
2246 }
2247
2248 /* We go to the (large) trouble of VLAN tagging ARP frames because
2249  * switches in VLAN mode (especially if ports are configured as
2250  * "native" to a VLAN) might not pass non-tagged frames.
2251  */
2252 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2253                           __be32 dest_ip, __be32 src_ip,
2254                           struct bond_vlan_tag *tags)
2255 {
2256         struct sk_buff *skb;
2257         struct bond_vlan_tag *outer_tag = tags;
2258
2259         netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2260                    arp_op, slave_dev->name, &dest_ip, &src_ip);
2261
2262         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2263                          NULL, slave_dev->dev_addr, NULL);
2264
2265         if (!skb) {
2266                 net_err_ratelimited("ARP packet allocation failed\n");
2267                 return;
2268         }
2269
2270         if (!tags || tags->vlan_proto == VLAN_N_VID)
2271                 goto xmit;
2272
2273         tags++;
2274
2275         /* Go through all the tags backwards and add them to the packet */
2276         while (tags->vlan_proto != VLAN_N_VID) {
2277                 if (!tags->vlan_id) {
2278                         tags++;
2279                         continue;
2280                 }
2281
2282                 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2283                            ntohs(outer_tag->vlan_proto), tags->vlan_id);
2284                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2285                                                 tags->vlan_id);
2286                 if (!skb) {
2287                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2288                         return;
2289                 }
2290
2291                 tags++;
2292         }
2293         /* Set the outer tag */
2294         if (outer_tag->vlan_id) {
2295                 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2296                            ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2297                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2298                                        outer_tag->vlan_id);
2299         }
2300
2301 xmit:
2302         arp_xmit(skb);
2303 }
2304
2305 /* Validate the device path between the @start_dev and the @end_dev.
2306  * The path is valid if the @end_dev is reachable through device
2307  * stacking.
2308  * When the path is validated, collect any vlan information in the
2309  * path.
2310  */
2311 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2312                                               struct net_device *end_dev,
2313                                               int level)
2314 {
2315         struct bond_vlan_tag *tags;
2316         struct net_device *upper;
2317         struct list_head  *iter;
2318
2319         if (start_dev == end_dev) {
2320                 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2321                 if (!tags)
2322                         return ERR_PTR(-ENOMEM);
2323                 tags[level].vlan_proto = VLAN_N_VID;
2324                 return tags;
2325         }
2326
2327         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2328                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2329                 if (IS_ERR_OR_NULL(tags)) {
2330                         if (IS_ERR(tags))
2331                                 return tags;
2332                         continue;
2333                 }
2334                 if (is_vlan_dev(upper)) {
2335                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2336                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2337                 }
2338
2339                 return tags;
2340         }
2341
2342         return NULL;
2343 }
2344
2345 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2346 {
2347         struct rtable *rt;
2348         struct bond_vlan_tag *tags;
2349         __be32 *targets = bond->params.arp_targets, addr;
2350         int i;
2351
2352         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2353                 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2354                 tags = NULL;
2355
2356                 /* Find out through which dev should the packet go */
2357                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2358                                      RTO_ONLINK, 0);
2359                 if (IS_ERR(rt)) {
2360                         /* there's no route to target - try to send arp
2361                          * probe to generate any traffic (arp_validate=0)
2362                          */
2363                         if (bond->params.arp_validate)
2364                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2365                                                      bond->dev->name,
2366                                                      &targets[i]);
2367                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2368                                       0, tags);
2369                         continue;
2370                 }
2371
2372                 /* bond device itself */
2373                 if (rt->dst.dev == bond->dev)
2374                         goto found;
2375
2376                 rcu_read_lock();
2377                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2378                 rcu_read_unlock();
2379
2380                 if (!IS_ERR_OR_NULL(tags))
2381                         goto found;
2382
2383                 /* Not our device - skip */
2384                 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2385                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2386
2387                 ip_rt_put(rt);
2388                 continue;
2389
2390 found:
2391                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2392                 ip_rt_put(rt);
2393                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2394                               addr, tags);
2395                 kfree(tags);
2396         }
2397 }
2398
2399 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2400 {
2401         int i;
2402
2403         if (!sip || !bond_has_this_ip(bond, tip)) {
2404                 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2405                            &sip, &tip);
2406                 return;
2407         }
2408
2409         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2410         if (i == -1) {
2411                 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2412                            &sip);
2413                 return;
2414         }
2415         slave->last_rx = jiffies;
2416         slave->target_last_arp_rx[i] = jiffies;
2417 }
2418
2419 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2420                  struct slave *slave)
2421 {
2422         struct arphdr *arp = (struct arphdr *)skb->data;
2423         struct slave *curr_active_slave, *curr_arp_slave;
2424         unsigned char *arp_ptr;
2425         __be32 sip, tip;
2426         int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2427
2428         if (!slave_do_arp_validate(bond, slave)) {
2429                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2430                     !slave_do_arp_validate_only(bond))
2431                         slave->last_rx = jiffies;
2432                 return RX_HANDLER_ANOTHER;
2433         } else if (!is_arp) {
2434                 return RX_HANDLER_ANOTHER;
2435         }
2436
2437         alen = arp_hdr_len(bond->dev);
2438
2439         netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2440                    skb->dev->name);
2441
2442         if (alen > skb_headlen(skb)) {
2443                 arp = kmalloc(alen, GFP_ATOMIC);
2444                 if (!arp)
2445                         goto out_unlock;
2446                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2447                         goto out_unlock;
2448         }
2449
2450         if (arp->ar_hln != bond->dev->addr_len ||
2451             skb->pkt_type == PACKET_OTHERHOST ||
2452             skb->pkt_type == PACKET_LOOPBACK ||
2453             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2454             arp->ar_pro != htons(ETH_P_IP) ||
2455             arp->ar_pln != 4)
2456                 goto out_unlock;
2457
2458         arp_ptr = (unsigned char *)(arp + 1);
2459         arp_ptr += bond->dev->addr_len;
2460         memcpy(&sip, arp_ptr, 4);
2461         arp_ptr += 4 + bond->dev->addr_len;
2462         memcpy(&tip, arp_ptr, 4);
2463
2464         netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2465                    slave->dev->name, bond_slave_state(slave),
2466                      bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2467                      &sip, &tip);
2468
2469         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2470         curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2471
2472         /* We 'trust' the received ARP enough to validate it if:
2473          *
2474          * (a) the slave receiving the ARP is active (which includes the
2475          * current ARP slave, if any), or
2476          *
2477          * (b) the receiving slave isn't active, but there is a currently
2478          * active slave and it received valid arp reply(s) after it became
2479          * the currently active slave, or
2480          *
2481          * (c) there is an ARP slave that sent an ARP during the prior ARP
2482          * interval, and we receive an ARP reply on any slave.  We accept
2483          * these because switch FDB update delays may deliver the ARP
2484          * reply to a slave other than the sender of the ARP request.
2485          *
2486          * Note: for (b), backup slaves are receiving the broadcast ARP
2487          * request, not a reply.  This request passes from the sending
2488          * slave through the L2 switch(es) to the receiving slave.  Since
2489          * this is checking the request, sip/tip are swapped for
2490          * validation.
2491          *
2492          * This is done to avoid endless looping when we can't reach the
2493          * arp_ip_target and fool ourselves with our own arp requests.
2494          */
2495         if (bond_is_active_slave(slave))
2496                 bond_validate_arp(bond, slave, sip, tip);
2497         else if (curr_active_slave &&
2498                  time_after(slave_last_rx(bond, curr_active_slave),
2499                             curr_active_slave->last_link_up))
2500                 bond_validate_arp(bond, slave, tip, sip);
2501         else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2502                  bond_time_in_interval(bond,
2503                                        dev_trans_start(curr_arp_slave->dev), 1))
2504                 bond_validate_arp(bond, slave, sip, tip);
2505
2506 out_unlock:
2507         if (arp != (struct arphdr *)skb->data)
2508                 kfree(arp);
2509         return RX_HANDLER_ANOTHER;
2510 }
2511
2512 /* function to verify if we're in the arp_interval timeslice, returns true if
2513  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2514  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2515  */
2516 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2517                                   int mod)
2518 {
2519         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2520
2521         return time_in_range(jiffies,
2522                              last_act - delta_in_ticks,
2523                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2524 }
2525
2526 /* This function is called regularly to monitor each slave's link
2527  * ensuring that traffic is being sent and received when arp monitoring
2528  * is used in load-balancing mode. if the adapter has been dormant, then an
2529  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2530  * arp monitoring in active backup mode.
2531  */
2532 static void bond_loadbalance_arp_mon(struct work_struct *work)
2533 {
2534         struct bonding *bond = container_of(work, struct bonding,
2535                                             arp_work.work);
2536         struct slave *slave, *oldcurrent;
2537         struct list_head *iter;
2538         int do_failover = 0, slave_state_changed = 0;
2539
2540         if (!bond_has_slaves(bond))
2541                 goto re_arm;
2542
2543         rcu_read_lock();
2544
2545         oldcurrent = rcu_dereference(bond->curr_active_slave);
2546         /* see if any of the previous devices are up now (i.e. they have
2547          * xmt and rcv traffic). the curr_active_slave does not come into
2548          * the picture unless it is null. also, slave->last_link_up is not
2549          * needed here because we send an arp on each slave and give a slave
2550          * as long as it needs to get the tx/rx within the delta.
2551          * TODO: what about up/down delay in arp mode? it wasn't here before
2552          *       so it can wait
2553          */
2554         bond_for_each_slave_rcu(bond, slave, iter) {
2555                 unsigned long trans_start = dev_trans_start(slave->dev);
2556
2557                 if (slave->link != BOND_LINK_UP) {
2558                         if (bond_time_in_interval(bond, trans_start, 1) &&
2559                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2560
2561                                 slave->link  = BOND_LINK_UP;
2562                                 slave_state_changed = 1;
2563
2564                                 /* primary_slave has no meaning in round-robin
2565                                  * mode. the window of a slave being up and
2566                                  * curr_active_slave being null after enslaving
2567                                  * is closed.
2568                                  */
2569                                 if (!oldcurrent) {
2570                                         netdev_info(bond->dev, "link status definitely up for interface %s\n",
2571                                                     slave->dev->name);
2572                                         do_failover = 1;
2573                                 } else {
2574                                         netdev_info(bond->dev, "interface %s is now up\n",
2575                                                     slave->dev->name);
2576                                 }
2577                         }
2578                 } else {
2579                         /* slave->link == BOND_LINK_UP */
2580
2581                         /* not all switches will respond to an arp request
2582                          * when the source ip is 0, so don't take the link down
2583                          * if we don't know our ip yet
2584                          */
2585                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2586                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2587
2588                                 slave->link  = BOND_LINK_DOWN;
2589                                 slave_state_changed = 1;
2590
2591                                 if (slave->link_failure_count < UINT_MAX)
2592                                         slave->link_failure_count++;
2593
2594                                 netdev_info(bond->dev, "interface %s is now down\n",
2595                                             slave->dev->name);
2596
2597                                 if (slave == oldcurrent)
2598                                         do_failover = 1;
2599                         }
2600                 }
2601
2602                 /* note: if switch is in round-robin mode, all links
2603                  * must tx arp to ensure all links rx an arp - otherwise
2604                  * links may oscillate or not come up at all; if switch is
2605                  * in something like xor mode, there is nothing we can
2606                  * do - all replies will be rx'ed on same link causing slaves
2607                  * to be unstable during low/no traffic periods
2608                  */
2609                 if (bond_slave_is_up(slave))
2610                         bond_arp_send_all(bond, slave);
2611         }
2612
2613         rcu_read_unlock();
2614
2615         if (do_failover || slave_state_changed) {
2616                 if (!rtnl_trylock())
2617                         goto re_arm;
2618
2619                 if (slave_state_changed) {
2620                         bond_slave_state_change(bond);
2621                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2622                                 bond_update_slave_arr(bond, NULL);
2623                 }
2624                 if (do_failover) {
2625                         block_netpoll_tx();
2626                         bond_select_active_slave(bond);
2627                         unblock_netpoll_tx();
2628                 }
2629                 rtnl_unlock();
2630         }
2631
2632 re_arm:
2633         if (bond->params.arp_interval)
2634                 queue_delayed_work(bond->wq, &bond->arp_work,
2635                                    msecs_to_jiffies(bond->params.arp_interval));
2636 }
2637
2638 /* Called to inspect slaves for active-backup mode ARP monitor link state
2639  * changes.  Sets new_link in slaves to specify what action should take
2640  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2641  * to link states must be committed.
2642  *
2643  * Called with rcu_read_lock held.
2644  */
2645 static int bond_ab_arp_inspect(struct bonding *bond)
2646 {
2647         unsigned long trans_start, last_rx;
2648         struct list_head *iter;
2649         struct slave *slave;
2650         int commit = 0;
2651
2652         bond_for_each_slave_rcu(bond, slave, iter) {
2653                 slave->new_link = BOND_LINK_NOCHANGE;
2654                 last_rx = slave_last_rx(bond, slave);
2655
2656                 if (slave->link != BOND_LINK_UP) {
2657                         if (bond_time_in_interval(bond, last_rx, 1)) {
2658                                 slave->new_link = BOND_LINK_UP;
2659                                 commit++;
2660                         }
2661                         continue;
2662                 }
2663
2664                 /* Give slaves 2*delta after being enslaved or made
2665                  * active.  This avoids bouncing, as the last receive
2666                  * times need a full ARP monitor cycle to be updated.
2667                  */
2668                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2669                         continue;
2670
2671                 /* Backup slave is down if:
2672                  * - No current_arp_slave AND
2673                  * - more than 3*delta since last receive AND
2674                  * - the bond has an IP address
2675                  *
2676                  * Note: a non-null current_arp_slave indicates
2677                  * the curr_active_slave went down and we are
2678                  * searching for a new one; under this condition
2679                  * we only take the curr_active_slave down - this
2680                  * gives each slave a chance to tx/rx traffic
2681                  * before being taken out
2682                  */
2683                 if (!bond_is_active_slave(slave) &&
2684                     !rcu_access_pointer(bond->current_arp_slave) &&
2685                     !bond_time_in_interval(bond, last_rx, 3)) {
2686                         slave->new_link = BOND_LINK_DOWN;
2687                         commit++;
2688                 }
2689
2690                 /* Active slave is down if:
2691                  * - more than 2*delta since transmitting OR
2692                  * - (more than 2*delta since receive AND
2693                  *    the bond has an IP address)
2694                  */
2695                 trans_start = dev_trans_start(slave->dev);
2696                 if (bond_is_active_slave(slave) &&
2697                     (!bond_time_in_interval(bond, trans_start, 2) ||
2698                      !bond_time_in_interval(bond, last_rx, 2))) {
2699                         slave->new_link = BOND_LINK_DOWN;
2700                         commit++;
2701                 }
2702         }
2703
2704         return commit;
2705 }
2706
2707 /* Called to commit link state changes noted by inspection step of
2708  * active-backup mode ARP monitor.
2709  *
2710  * Called with RTNL hold.
2711  */
2712 static void bond_ab_arp_commit(struct bonding *bond)
2713 {
2714         unsigned long trans_start;
2715         struct list_head *iter;
2716         struct slave *slave;
2717
2718         bond_for_each_slave(bond, slave, iter) {
2719                 switch (slave->new_link) {
2720                 case BOND_LINK_NOCHANGE:
2721                         continue;
2722
2723                 case BOND_LINK_UP:
2724                         trans_start = dev_trans_start(slave->dev);
2725                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
2726                             (!rtnl_dereference(bond->curr_active_slave) &&
2727                              bond_time_in_interval(bond, trans_start, 1))) {
2728                                 struct slave *current_arp_slave;
2729
2730                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2731                                 bond_set_slave_link_state(slave, BOND_LINK_UP);
2732                                 if (current_arp_slave) {
2733                                         bond_set_slave_inactive_flags(
2734                                                 current_arp_slave,
2735                                                 BOND_SLAVE_NOTIFY_NOW);
2736                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2737                                 }
2738
2739                                 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2740                                             slave->dev->name);
2741
2742                                 if (!rtnl_dereference(bond->curr_active_slave) ||
2743                                     slave == rtnl_dereference(bond->primary_slave))
2744                                         goto do_failover;
2745
2746                         }
2747
2748                         continue;
2749
2750                 case BOND_LINK_DOWN:
2751                         if (slave->link_failure_count < UINT_MAX)
2752                                 slave->link_failure_count++;
2753
2754                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2755                         bond_set_slave_inactive_flags(slave,
2756                                                       BOND_SLAVE_NOTIFY_NOW);
2757
2758                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2759                                     slave->dev->name);
2760
2761                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
2762                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2763                                 goto do_failover;
2764                         }
2765
2766                         continue;
2767
2768                 default:
2769                         netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2770                                    slave->new_link, slave->dev->name);
2771                         continue;
2772                 }
2773
2774 do_failover:
2775                 block_netpoll_tx();
2776                 bond_select_active_slave(bond);
2777                 unblock_netpoll_tx();
2778         }
2779
2780         bond_set_carrier(bond);
2781 }
2782
2783 /* Send ARP probes for active-backup mode ARP monitor.
2784  *
2785  * Called with rcu_read_lock held.
2786  */
2787 static bool bond_ab_arp_probe(struct bonding *bond)
2788 {
2789         struct slave *slave, *before = NULL, *new_slave = NULL,
2790                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2791                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2792         struct list_head *iter;
2793         bool found = false;
2794         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2795
2796         if (curr_arp_slave && curr_active_slave)
2797                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2798                             curr_arp_slave->dev->name,
2799                             curr_active_slave->dev->name);
2800
2801         if (curr_active_slave) {
2802                 bond_arp_send_all(bond, curr_active_slave);
2803                 return should_notify_rtnl;
2804         }
2805
2806         /* if we don't have a curr_active_slave, search for the next available
2807          * backup slave from the current_arp_slave and make it the candidate
2808          * for becoming the curr_active_slave
2809          */
2810
2811         if (!curr_arp_slave) {
2812                 curr_arp_slave = bond_first_slave_rcu(bond);
2813                 if (!curr_arp_slave)
2814                         return should_notify_rtnl;
2815         }
2816
2817         bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2818
2819         bond_for_each_slave_rcu(bond, slave, iter) {
2820                 if (!found && !before && bond_slave_is_up(slave))
2821                         before = slave;
2822
2823                 if (found && !new_slave && bond_slave_is_up(slave))
2824                         new_slave = slave;
2825                 /* if the link state is up at this point, we
2826                  * mark it down - this can happen if we have
2827                  * simultaneous link failures and
2828                  * reselect_active_interface doesn't make this
2829                  * one the current slave so it is still marked
2830                  * up when it is actually down
2831                  */
2832                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2833                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2834                         if (slave->link_failure_count < UINT_MAX)
2835                                 slave->link_failure_count++;
2836
2837                         bond_set_slave_inactive_flags(slave,
2838                                                       BOND_SLAVE_NOTIFY_LATER);
2839
2840                         netdev_info(bond->dev, "backup interface %s is now down\n",
2841                                     slave->dev->name);
2842                 }
2843                 if (slave == curr_arp_slave)
2844                         found = true;
2845         }
2846
2847         if (!new_slave && before)
2848                 new_slave = before;
2849
2850         if (!new_slave)
2851                 goto check_state;
2852
2853         bond_set_slave_link_state(new_slave, BOND_LINK_BACK);
2854         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2855         bond_arp_send_all(bond, new_slave);
2856         new_slave->last_link_up = jiffies;
2857         rcu_assign_pointer(bond->current_arp_slave, new_slave);
2858
2859 check_state:
2860         bond_for_each_slave_rcu(bond, slave, iter) {
2861                 if (slave->should_notify) {
2862                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2863                         break;
2864                 }
2865         }
2866         return should_notify_rtnl;
2867 }
2868
2869 static void bond_activebackup_arp_mon(struct work_struct *work)
2870 {
2871         struct bonding *bond = container_of(work, struct bonding,
2872                                             arp_work.work);
2873         bool should_notify_peers = false;
2874         bool should_notify_rtnl = false;
2875         int delta_in_ticks;
2876
2877         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2878
2879         if (!bond_has_slaves(bond))
2880                 goto re_arm;
2881
2882         rcu_read_lock();
2883
2884         should_notify_peers = bond_should_notify_peers(bond);
2885
2886         if (bond_ab_arp_inspect(bond)) {
2887                 rcu_read_unlock();
2888
2889                 /* Race avoidance with bond_close flush of workqueue */
2890                 if (!rtnl_trylock()) {
2891                         delta_in_ticks = 1;
2892                         should_notify_peers = false;
2893                         goto re_arm;
2894                 }
2895
2896                 bond_ab_arp_commit(bond);
2897
2898                 rtnl_unlock();
2899                 rcu_read_lock();
2900         }
2901
2902         should_notify_rtnl = bond_ab_arp_probe(bond);
2903         rcu_read_unlock();
2904
2905 re_arm:
2906         if (bond->params.arp_interval)
2907                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2908
2909         if (should_notify_peers || should_notify_rtnl) {
2910                 if (!rtnl_trylock())
2911                         return;
2912
2913                 if (should_notify_peers)
2914                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2915                                                  bond->dev);
2916                 if (should_notify_rtnl)
2917                         bond_slave_state_notify(bond);
2918
2919                 rtnl_unlock();
2920         }
2921 }
2922
2923 /*-------------------------- netdev event handling --------------------------*/
2924
2925 /* Change device name */
2926 static int bond_event_changename(struct bonding *bond)
2927 {
2928         bond_remove_proc_entry(bond);
2929         bond_create_proc_entry(bond);
2930
2931         bond_debug_reregister(bond);
2932
2933         return NOTIFY_DONE;
2934 }
2935
2936 static int bond_master_netdev_event(unsigned long event,
2937                                     struct net_device *bond_dev)
2938 {
2939         struct bonding *event_bond = netdev_priv(bond_dev);
2940
2941         switch (event) {
2942         case NETDEV_CHANGENAME:
2943                 return bond_event_changename(event_bond);
2944         case NETDEV_UNREGISTER:
2945                 bond_remove_proc_entry(event_bond);
2946                 break;
2947         case NETDEV_REGISTER:
2948                 bond_create_proc_entry(event_bond);
2949                 break;
2950         case NETDEV_NOTIFY_PEERS:
2951                 if (event_bond->send_peer_notif)
2952                         event_bond->send_peer_notif--;
2953                 break;
2954         default:
2955                 break;
2956         }
2957
2958         return NOTIFY_DONE;
2959 }
2960
2961 static int bond_slave_netdev_event(unsigned long event,
2962                                    struct net_device *slave_dev)
2963 {
2964         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
2965         struct bonding *bond;
2966         struct net_device *bond_dev;
2967
2968         /* A netdev event can be generated while enslaving a device
2969          * before netdev_rx_handler_register is called in which case
2970          * slave will be NULL
2971          */
2972         if (!slave)
2973                 return NOTIFY_DONE;
2974         bond_dev = slave->bond->dev;
2975         bond = slave->bond;
2976         primary = rtnl_dereference(bond->primary_slave);
2977
2978         switch (event) {
2979         case NETDEV_UNREGISTER:
2980                 if (bond_dev->type != ARPHRD_ETHER)
2981                         bond_release_and_destroy(bond_dev, slave_dev);
2982                 else
2983                         bond_release(bond_dev, slave_dev);
2984                 break;
2985         case NETDEV_UP:
2986         case NETDEV_CHANGE:
2987                 bond_update_speed_duplex(slave);
2988                 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2989                         bond_3ad_adapter_speed_duplex_changed(slave);
2990                 /* Fallthrough */
2991         case NETDEV_DOWN:
2992                 /* Refresh slave-array if applicable!
2993                  * If the setup does not use miimon or arpmon (mode-specific!),
2994                  * then these events will not cause the slave-array to be
2995                  * refreshed. This will cause xmit to use a slave that is not
2996                  * usable. Avoid such situation by refeshing the array at these
2997                  * events. If these (miimon/arpmon) parameters are configured
2998                  * then array gets refreshed twice and that should be fine!
2999                  */
3000                 if (bond_mode_uses_xmit_hash(bond))
3001                         bond_update_slave_arr(bond, NULL);
3002                 break;
3003         case NETDEV_CHANGEMTU:
3004                 /* TODO: Should slaves be allowed to
3005                  * independently alter their MTU?  For
3006                  * an active-backup bond, slaves need
3007                  * not be the same type of device, so
3008                  * MTUs may vary.  For other modes,
3009                  * slaves arguably should have the
3010                  * same MTUs. To do this, we'd need to
3011                  * take over the slave's change_mtu
3012                  * function for the duration of their
3013                  * servitude.
3014                  */
3015                 break;
3016         case NETDEV_CHANGENAME:
3017                 /* we don't care if we don't have primary set */
3018                 if (!bond_uses_primary(bond) ||
3019                     !bond->params.primary[0])
3020                         break;
3021
3022                 if (slave == primary) {
3023                         /* slave's name changed - he's no longer primary */
3024                         RCU_INIT_POINTER(bond->primary_slave, NULL);
3025                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3026                         /* we have a new primary slave */
3027                         rcu_assign_pointer(bond->primary_slave, slave);
3028                 } else { /* we didn't change primary - exit */
3029                         break;
3030                 }
3031
3032                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3033                             primary ? slave_dev->name : "none");
3034
3035                 block_netpoll_tx();
3036                 bond_select_active_slave(bond);
3037                 unblock_netpoll_tx();
3038                 break;
3039         case NETDEV_FEAT_CHANGE:
3040                 bond_compute_features(bond);
3041                 break;
3042         case NETDEV_RESEND_IGMP:
3043                 /* Propagate to master device */
3044                 call_netdevice_notifiers(event, slave->bond->dev);
3045                 break;
3046         default:
3047                 break;
3048         }
3049
3050         return NOTIFY_DONE;
3051 }
3052
3053 /* bond_netdev_event: handle netdev notifier chain events.
3054  *
3055  * This function receives events for the netdev chain.  The caller (an
3056  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3057  * locks for us to safely manipulate the slave devices (RTNL lock,
3058  * dev_probe_lock).
3059  */
3060 static int bond_netdev_event(struct notifier_block *this,
3061                              unsigned long event, void *ptr)
3062 {
3063         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3064
3065         netdev_dbg(event_dev, "event: %lx\n", event);
3066
3067         if (!(event_dev->priv_flags & IFF_BONDING))
3068                 return NOTIFY_DONE;
3069
3070         if (event_dev->flags & IFF_MASTER) {
3071                 netdev_dbg(event_dev, "IFF_MASTER\n");
3072                 return bond_master_netdev_event(event, event_dev);
3073         }
3074
3075         if (event_dev->flags & IFF_SLAVE) {
3076                 netdev_dbg(event_dev, "IFF_SLAVE\n");
3077                 return bond_slave_netdev_event(event, event_dev);
3078         }
3079
3080         return NOTIFY_DONE;
3081 }
3082
3083 static struct notifier_block bond_netdev_notifier = {
3084         .notifier_call = bond_netdev_event,
3085 };
3086
3087 /*---------------------------- Hashing Policies -----------------------------*/
3088
3089 /* L2 hash helper */
3090 static inline u32 bond_eth_hash(struct sk_buff *skb)
3091 {
3092         struct ethhdr *ep, hdr_tmp;
3093
3094         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3095         if (ep)
3096                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3097         return 0;
3098 }
3099
3100 /* Extract the appropriate headers based on bond's xmit policy */
3101 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3102                               struct flow_keys *fk)
3103 {
3104         const struct ipv6hdr *iph6;
3105         const struct iphdr *iph;
3106         int noff, proto = -1;
3107
3108         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3109                 return skb_flow_dissect_flow_keys(skb, fk, 0);
3110
3111         fk->ports.ports = 0;
3112         noff = skb_network_offset(skb);
3113         if (skb->protocol == htons(ETH_P_IP)) {
3114                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3115                         return false;
3116                 iph = ip_hdr(skb);
3117                 iph_to_flow_copy_v4addrs(fk, iph);
3118                 noff += iph->ihl << 2;
3119                 if (!ip_is_fragment(iph))
3120                         proto = iph->protocol;
3121         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3122                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3123                         return false;
3124                 iph6 = ipv6_hdr(skb);
3125                 iph_to_flow_copy_v6addrs(fk, iph6);
3126                 noff += sizeof(*iph6);
3127                 proto = iph6->nexthdr;
3128         } else {
3129                 return false;
3130         }
3131         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3132                 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3133
3134         return true;
3135 }
3136
3137 /**
3138  * bond_xmit_hash - generate a hash value based on the xmit policy
3139  * @bond: bonding device
3140  * @skb: buffer to use for headers
3141  *
3142  * This function will extract the necessary headers from the skb buffer and use
3143  * them to generate a hash based on the xmit_policy set in the bonding device
3144  */
3145 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3146 {
3147         struct flow_keys flow;
3148         u32 hash;
3149
3150         if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3151             skb->l4_hash)
3152                 return skb->hash;
3153
3154         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3155             !bond_flow_dissect(bond, skb, &flow))
3156                 return bond_eth_hash(skb);
3157
3158         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3159             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3160                 hash = bond_eth_hash(skb);
3161         else
3162                 hash = (__force u32)flow.ports.ports;
3163         hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3164                 (__force u32)flow_get_u32_src(&flow);
3165         hash ^= (hash >> 16);
3166         hash ^= (hash >> 8);
3167
3168         return hash;
3169 }
3170
3171 /*-------------------------- Device entry points ----------------------------*/
3172
3173 static void bond_work_init_all(struct bonding *bond)
3174 {
3175         INIT_DELAYED_WORK(&bond->mcast_work,
3176                           bond_resend_igmp_join_requests_delayed);
3177         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3178         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3179         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3180                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3181         else
3182                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3183         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3184         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3185 }
3186
3187 static void bond_work_cancel_all(struct bonding *bond)
3188 {
3189         cancel_delayed_work_sync(&bond->mii_work);
3190         cancel_delayed_work_sync(&bond->arp_work);
3191         cancel_delayed_work_sync(&bond->alb_work);
3192         cancel_delayed_work_sync(&bond->ad_work);
3193         cancel_delayed_work_sync(&bond->mcast_work);
3194         cancel_delayed_work_sync(&bond->slave_arr_work);
3195 }
3196
3197 static int bond_open(struct net_device *bond_dev)
3198 {
3199         struct bonding *bond = netdev_priv(bond_dev);
3200         struct list_head *iter;
3201         struct slave *slave;
3202
3203         /* reset slave->backup and slave->inactive */
3204         if (bond_has_slaves(bond)) {
3205                 bond_for_each_slave(bond, slave, iter) {
3206                         if (bond_uses_primary(bond) &&
3207                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3208                                 bond_set_slave_inactive_flags(slave,
3209                                                               BOND_SLAVE_NOTIFY_NOW);
3210                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3211                                 bond_set_slave_active_flags(slave,
3212                                                             BOND_SLAVE_NOTIFY_NOW);
3213                         }
3214                 }
3215         }
3216
3217         bond_work_init_all(bond);
3218
3219         if (bond_is_lb(bond)) {
3220                 /* bond_alb_initialize must be called before the timer
3221                  * is started.
3222                  */
3223                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3224                         return -ENOMEM;
3225                 if (bond->params.tlb_dynamic_lb)
3226                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3227         }
3228
3229         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3230                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3231
3232         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3233                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3234                 bond->recv_probe = bond_arp_rcv;
3235         }
3236
3237         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3238                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3239                 /* register to receive LACPDUs */
3240                 bond->recv_probe = bond_3ad_lacpdu_recv;
3241                 bond_3ad_initiate_agg_selection(bond, 1);
3242         }
3243
3244         if (bond_mode_uses_xmit_hash(bond))
3245                 bond_update_slave_arr(bond, NULL);
3246
3247         return 0;
3248 }
3249
3250 static int bond_close(struct net_device *bond_dev)
3251 {
3252         struct bonding *bond = netdev_priv(bond_dev);
3253
3254         bond_work_cancel_all(bond);
3255         bond->send_peer_notif = 0;
3256         if (bond_is_lb(bond))
3257                 bond_alb_deinitialize(bond);
3258         bond->recv_probe = NULL;
3259
3260         return 0;
3261 }
3262
3263 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3264  * that some drivers can provide 32bit values only.
3265  */
3266 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3267                             const struct rtnl_link_stats64 *_new,
3268                             const struct rtnl_link_stats64 *_old)
3269 {
3270         const u64 *new = (const u64 *)_new;
3271         const u64 *old = (const u64 *)_old;
3272         u64 *res = (u64 *)_res;
3273         int i;
3274
3275         for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3276                 u64 nv = new[i];
3277                 u64 ov = old[i];
3278
3279                 /* detects if this particular field is 32bit only */
3280                 if (((nv | ov) >> 32) == 0)
3281                         res[i] += (u32)nv - (u32)ov;
3282                 else
3283                         res[i] += nv - ov;
3284         }
3285 }
3286
3287 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3288                                                 struct rtnl_link_stats64 *stats)
3289 {
3290         struct bonding *bond = netdev_priv(bond_dev);
3291         struct rtnl_link_stats64 temp;
3292         struct list_head *iter;
3293         struct slave *slave;
3294
3295         spin_lock(&bond->stats_lock);
3296         memcpy(stats, &bond->bond_stats, sizeof(*stats));
3297
3298         rcu_read_lock();
3299         bond_for_each_slave_rcu(bond, slave, iter) {
3300                 const struct rtnl_link_stats64 *new =
3301                         dev_get_stats(slave->dev, &temp);
3302
3303                 bond_fold_stats(stats, new, &slave->slave_stats);
3304
3305                 /* save off the slave stats for the next run */
3306                 memcpy(&slave->slave_stats, new, sizeof(*new));
3307         }
3308         rcu_read_unlock();
3309
3310         memcpy(&bond->bond_stats, stats, sizeof(*stats));
3311         spin_unlock(&bond->stats_lock);
3312
3313         return stats;
3314 }
3315
3316 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3317 {
3318         struct bonding *bond = netdev_priv(bond_dev);
3319         struct net_device *slave_dev = NULL;
3320         struct ifbond k_binfo;
3321         struct ifbond __user *u_binfo = NULL;
3322         struct ifslave k_sinfo;
3323         struct ifslave __user *u_sinfo = NULL;
3324         struct mii_ioctl_data *mii = NULL;
3325         struct bond_opt_value newval;
3326         struct net *net;
3327         int res = 0;
3328
3329         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3330
3331         switch (cmd) {
3332         case SIOCGMIIPHY:
3333                 mii = if_mii(ifr);
3334                 if (!mii)
3335                         return -EINVAL;
3336
3337                 mii->phy_id = 0;
3338                 /* Fall Through */
3339         case SIOCGMIIREG:
3340                 /* We do this again just in case we were called by SIOCGMIIREG
3341                  * instead of SIOCGMIIPHY.
3342                  */
3343                 mii = if_mii(ifr);
3344                 if (!mii)
3345                         return -EINVAL;
3346
3347                 if (mii->reg_num == 1) {
3348                         mii->val_out = 0;
3349                         if (netif_carrier_ok(bond->dev))
3350                                 mii->val_out = BMSR_LSTATUS;
3351                 }
3352
3353                 return 0;
3354         case BOND_INFO_QUERY_OLD:
3355         case SIOCBONDINFOQUERY:
3356                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3357
3358                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3359                         return -EFAULT;
3360
3361                 res = bond_info_query(bond_dev, &k_binfo);
3362                 if (res == 0 &&
3363                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3364                         return -EFAULT;
3365
3366                 return res;
3367         case BOND_SLAVE_INFO_QUERY_OLD:
3368         case SIOCBONDSLAVEINFOQUERY:
3369                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3370
3371                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3372                         return -EFAULT;
3373
3374                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3375                 if (res == 0 &&
3376                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3377                         return -EFAULT;
3378
3379                 return res;
3380         default:
3381                 break;
3382         }
3383
3384         net = dev_net(bond_dev);
3385
3386         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3387                 return -EPERM;
3388
3389         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3390
3391         netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3392
3393         if (!slave_dev)
3394                 return -ENODEV;
3395
3396         netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3397         switch (cmd) {
3398         case BOND_ENSLAVE_OLD:
3399         case SIOCBONDENSLAVE:
3400                 res = bond_enslave(bond_dev, slave_dev);
3401                 break;
3402         case BOND_RELEASE_OLD:
3403         case SIOCBONDRELEASE:
3404                 res = bond_release(bond_dev, slave_dev);
3405                 break;
3406         case BOND_SETHWADDR_OLD:
3407         case SIOCBONDSETHWADDR:
3408                 bond_set_dev_addr(bond_dev, slave_dev);
3409                 res = 0;
3410                 break;
3411         case BOND_CHANGE_ACTIVE_OLD:
3412         case SIOCBONDCHANGEACTIVE:
3413                 bond_opt_initstr(&newval, slave_dev->name);
3414                 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3415                 break;
3416         default:
3417                 res = -EOPNOTSUPP;
3418         }
3419
3420         return res;
3421 }
3422
3423 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3424 {
3425         struct bonding *bond = netdev_priv(bond_dev);
3426
3427         if (change & IFF_PROMISC)
3428                 bond_set_promiscuity(bond,
3429                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3430
3431         if (change & IFF_ALLMULTI)
3432                 bond_set_allmulti(bond,
3433                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3434 }
3435
3436 static void bond_set_rx_mode(struct net_device *bond_dev)
3437 {
3438         struct bonding *bond = netdev_priv(bond_dev);
3439         struct list_head *iter;
3440         struct slave *slave;
3441
3442         rcu_read_lock();
3443         if (bond_uses_primary(bond)) {
3444                 slave = rcu_dereference(bond->curr_active_slave);
3445                 if (slave) {
3446                         dev_uc_sync(slave->dev, bond_dev);
3447                         dev_mc_sync(slave->dev, bond_dev);
3448                 }
3449         } else {
3450                 bond_for_each_slave_rcu(bond, slave, iter) {
3451                         dev_uc_sync_multiple(slave->dev, bond_dev);
3452                         dev_mc_sync_multiple(slave->dev, bond_dev);
3453                 }
3454         }
3455         rcu_read_unlock();
3456 }
3457
3458 static int bond_neigh_init(struct neighbour *n)
3459 {
3460         struct bonding *bond = netdev_priv(n->dev);
3461         const struct net_device_ops *slave_ops;
3462         struct neigh_parms parms;
3463         struct slave *slave;
3464         int ret;
3465
3466         slave = bond_first_slave(bond);
3467         if (!slave)
3468                 return 0;
3469         slave_ops = slave->dev->netdev_ops;
3470         if (!slave_ops->ndo_neigh_setup)
3471                 return 0;
3472
3473         parms.neigh_setup = NULL;
3474         parms.neigh_cleanup = NULL;
3475         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3476         if (ret)
3477                 return ret;
3478
3479         /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3480          * after the last slave has been detached.  Assumes that all slaves
3481          * utilize the same neigh_cleanup (true at this writing as only user
3482          * is ipoib).
3483          */
3484         n->parms->neigh_cleanup = parms.neigh_cleanup;
3485
3486         if (!parms.neigh_setup)
3487                 return 0;
3488
3489         return parms.neigh_setup(n);
3490 }
3491
3492 /* The bonding ndo_neigh_setup is called at init time beofre any
3493  * slave exists. So we must declare proxy setup function which will
3494  * be used at run time to resolve the actual slave neigh param setup.
3495  *
3496  * It's also called by master devices (such as vlans) to setup their
3497  * underlying devices. In that case - do nothing, we're already set up from
3498  * our init.
3499  */
3500 static int bond_neigh_setup(struct net_device *dev,
3501                             struct neigh_parms *parms)
3502 {
3503         /* modify only our neigh_parms */
3504         if (parms->dev == dev)
3505                 parms->neigh_setup = bond_neigh_init;
3506
3507         return 0;
3508 }
3509
3510 /* Change the MTU of all of a master's slaves to match the master */
3511 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3512 {
3513         struct bonding *bond = netdev_priv(bond_dev);
3514         struct slave *slave, *rollback_slave;
3515         struct list_head *iter;
3516         int res = 0;
3517
3518         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3519
3520         bond_for_each_slave(bond, slave, iter) {
3521                 netdev_dbg(bond_dev, "s %p c_m %p\n",
3522                            slave, slave->dev->netdev_ops->ndo_change_mtu);
3523
3524                 res = dev_set_mtu(slave->dev, new_mtu);
3525
3526                 if (res) {
3527                         /* If we failed to set the slave's mtu to the new value
3528                          * we must abort the operation even in ACTIVE_BACKUP
3529                          * mode, because if we allow the backup slaves to have
3530                          * different mtu values than the active slave we'll
3531                          * need to change their mtu when doing a failover. That
3532                          * means changing their mtu from timer context, which
3533                          * is probably not a good idea.
3534                          */
3535                         netdev_dbg(bond_dev, "err %d %s\n", res,
3536                                    slave->dev->name);
3537                         goto unwind;
3538                 }
3539         }
3540
3541         bond_dev->mtu = new_mtu;
3542
3543         return 0;
3544
3545 unwind:
3546         /* unwind from head to the slave that failed */
3547         bond_for_each_slave(bond, rollback_slave, iter) {
3548                 int tmp_res;
3549
3550                 if (rollback_slave == slave)
3551                         break;
3552
3553                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3554                 if (tmp_res) {
3555                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3556                                    tmp_res, rollback_slave->dev->name);
3557                 }
3558         }
3559
3560         return res;
3561 }
3562
3563 /* Change HW address
3564  *
3565  * Note that many devices must be down to change the HW address, and
3566  * downing the master releases all slaves.  We can make bonds full of
3567  * bonding devices to test this, however.
3568  */
3569 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3570 {
3571         struct bonding *bond = netdev_priv(bond_dev);
3572         struct slave *slave, *rollback_slave;
3573         struct sockaddr *sa = addr, tmp_sa;
3574         struct list_head *iter;
3575         int res = 0;
3576
3577         if (BOND_MODE(bond) == BOND_MODE_ALB)
3578                 return bond_alb_set_mac_address(bond_dev, addr);
3579
3580
3581         netdev_dbg(bond_dev, "bond=%p\n", bond);
3582
3583         /* If fail_over_mac is enabled, do nothing and return success.
3584          * Returning an error causes ifenslave to fail.
3585          */
3586         if (bond->params.fail_over_mac &&
3587             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3588                 return 0;
3589
3590         if (!is_valid_ether_addr(sa->sa_data))
3591                 return -EADDRNOTAVAIL;
3592
3593         bond_for_each_slave(bond, slave, iter) {
3594                 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3595                 res = dev_set_mac_address(slave->dev, addr);
3596                 if (res) {
3597                         /* TODO: consider downing the slave
3598                          * and retry ?
3599                          * User should expect communications
3600                          * breakage anyway until ARP finish
3601                          * updating, so...
3602                          */
3603                         netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3604                         goto unwind;
3605                 }
3606         }
3607
3608         /* success */
3609         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3610         return 0;
3611
3612 unwind:
3613         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3614         tmp_sa.sa_family = bond_dev->type;
3615
3616         /* unwind from head to the slave that failed */
3617         bond_for_each_slave(bond, rollback_slave, iter) {
3618                 int tmp_res;
3619
3620                 if (rollback_slave == slave)
3621                         break;
3622
3623                 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3624                 if (tmp_res) {
3625                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3626                                    tmp_res, rollback_slave->dev->name);
3627                 }
3628         }
3629
3630         return res;
3631 }
3632
3633 /**
3634  * bond_xmit_slave_id - transmit skb through slave with slave_id
3635  * @bond: bonding device that is transmitting
3636  * @skb: buffer to transmit
3637  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3638  *
3639  * This function tries to transmit through slave with slave_id but in case
3640  * it fails, it tries to find the first available slave for transmission.
3641  * The skb is consumed in all cases, thus the function is void.
3642  */
3643 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3644 {
3645         struct list_head *iter;
3646         struct slave *slave;
3647         int i = slave_id;
3648
3649         /* Here we start from the slave with slave_id */
3650         bond_for_each_slave_rcu(bond, slave, iter) {
3651                 if (--i < 0) {
3652                         if (bond_slave_can_tx(slave)) {
3653                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3654                                 return;
3655                         }
3656                 }
3657         }
3658
3659         /* Here we start from the first slave up to slave_id */
3660         i = slave_id;
3661         bond_for_each_slave_rcu(bond, slave, iter) {
3662                 if (--i < 0)
3663                         break;
3664                 if (bond_slave_can_tx(slave)) {
3665                         bond_dev_queue_xmit(bond, skb, slave->dev);
3666                         return;
3667                 }
3668         }
3669         /* no slave that can tx has been found */
3670         bond_tx_drop(bond->dev, skb);
3671 }
3672
3673 /**
3674  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3675  * @bond: bonding device to use
3676  *
3677  * Based on the value of the bonding device's packets_per_slave parameter
3678  * this function generates a slave id, which is usually used as the next
3679  * slave to transmit through.
3680  */
3681 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3682 {
3683         u32 slave_id;
3684         struct reciprocal_value reciprocal_packets_per_slave;
3685         int packets_per_slave = bond->params.packets_per_slave;
3686
3687         switch (packets_per_slave) {
3688         case 0:
3689                 slave_id = prandom_u32();
3690                 break;
3691         case 1:
3692                 slave_id = bond->rr_tx_counter;
3693                 break;
3694         default:
3695                 reciprocal_packets_per_slave =
3696                         bond->params.reciprocal_packets_per_slave;
3697                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3698                                              reciprocal_packets_per_slave);
3699                 break;
3700         }
3701         bond->rr_tx_counter++;
3702
3703         return slave_id;
3704 }
3705
3706 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3707 {
3708         struct bonding *bond = netdev_priv(bond_dev);
3709         struct iphdr *iph = ip_hdr(skb);
3710         struct slave *slave;
3711         u32 slave_id;
3712
3713         /* Start with the curr_active_slave that joined the bond as the
3714          * default for sending IGMP traffic.  For failover purposes one
3715          * needs to maintain some consistency for the interface that will
3716          * send the join/membership reports.  The curr_active_slave found
3717          * will send all of this type of traffic.
3718          */
3719         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3720                 slave = rcu_dereference(bond->curr_active_slave);
3721                 if (slave)
3722                         bond_dev_queue_xmit(bond, skb, slave->dev);
3723                 else
3724                         bond_xmit_slave_id(bond, skb, 0);
3725         } else {
3726                 int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3727
3728                 if (likely(slave_cnt)) {
3729                         slave_id = bond_rr_gen_slave_id(bond);
3730                         bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3731                 } else {
3732                         bond_tx_drop(bond_dev, skb);
3733                 }
3734         }
3735
3736         return NETDEV_TX_OK;
3737 }
3738
3739 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3740  * the bond has a usable interface.
3741  */
3742 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3743 {
3744         struct bonding *bond = netdev_priv(bond_dev);
3745         struct slave *slave;
3746
3747         slave = rcu_dereference(bond->curr_active_slave);
3748         if (slave)
3749                 bond_dev_queue_xmit(bond, skb, slave->dev);
3750         else
3751                 bond_tx_drop(bond_dev, skb);
3752
3753         return NETDEV_TX_OK;
3754 }
3755
3756 /* Use this to update slave_array when (a) it's not appropriate to update
3757  * slave_array right away (note that update_slave_array() may sleep)
3758  * and / or (b) RTNL is not held.
3759  */
3760 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3761 {
3762         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3763 }
3764
3765 /* Slave array work handler. Holds only RTNL */
3766 static void bond_slave_arr_handler(struct work_struct *work)
3767 {
3768         struct bonding *bond = container_of(work, struct bonding,
3769                                             slave_arr_work.work);
3770         int ret;
3771
3772         if (!rtnl_trylock())
3773                 goto err;
3774
3775         ret = bond_update_slave_arr(bond, NULL);
3776         rtnl_unlock();
3777         if (ret) {
3778                 pr_warn_ratelimited("Failed to update slave array from WT\n");
3779                 goto err;
3780         }
3781         return;
3782
3783 err:
3784         bond_slave_arr_work_rearm(bond, 1);
3785 }
3786
3787 /* Build the usable slaves array in control path for modes that use xmit-hash
3788  * to determine the slave interface -
3789  * (a) BOND_MODE_8023AD
3790  * (b) BOND_MODE_XOR
3791  * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3792  *
3793  * The caller is expected to hold RTNL only and NO other lock!
3794  */
3795 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3796 {
3797         struct slave *slave;
3798         struct list_head *iter;
3799         struct bond_up_slave *new_arr, *old_arr;
3800         int agg_id = 0;
3801         int ret = 0;
3802
3803 #ifdef CONFIG_LOCKDEP
3804         WARN_ON(lockdep_is_held(&bond->mode_lock));
3805 #endif
3806
3807         new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3808                           GFP_KERNEL);
3809         if (!new_arr) {
3810                 ret = -ENOMEM;
3811                 pr_err("Failed to build slave-array.\n");
3812                 goto out;
3813         }
3814         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3815                 struct ad_info ad_info;
3816
3817                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3818                         pr_debug("bond_3ad_get_active_agg_info failed\n");
3819                         kfree_rcu(new_arr, rcu);
3820                         /* No active aggragator means it's not safe to use
3821                          * the previous array.
3822                          */
3823                         old_arr = rtnl_dereference(bond->slave_arr);
3824                         if (old_arr) {
3825                                 RCU_INIT_POINTER(bond->slave_arr, NULL);
3826                                 kfree_rcu(old_arr, rcu);
3827                         }
3828                         goto out;
3829                 }
3830                 agg_id = ad_info.aggregator_id;
3831         }
3832         bond_for_each_slave(bond, slave, iter) {
3833                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3834                         struct aggregator *agg;
3835
3836                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
3837                         if (!agg || agg->aggregator_identifier != agg_id)
3838                                 continue;
3839                 }
3840                 if (!bond_slave_can_tx(slave))
3841                         continue;
3842                 if (skipslave == slave)
3843                         continue;
3844                 new_arr->arr[new_arr->count++] = slave;
3845         }
3846
3847         old_arr = rtnl_dereference(bond->slave_arr);
3848         rcu_assign_pointer(bond->slave_arr, new_arr);
3849         if (old_arr)
3850                 kfree_rcu(old_arr, rcu);
3851 out:
3852         if (ret != 0 && skipslave) {
3853                 int idx;
3854
3855                 /* Rare situation where caller has asked to skip a specific
3856                  * slave but allocation failed (most likely!). BTW this is
3857                  * only possible when the call is initiated from
3858                  * __bond_release_one(). In this situation; overwrite the
3859                  * skipslave entry in the array with the last entry from the
3860                  * array to avoid a situation where the xmit path may choose
3861                  * this to-be-skipped slave to send a packet out.
3862                  */
3863                 old_arr = rtnl_dereference(bond->slave_arr);
3864                 for (idx = 0; idx < old_arr->count; idx++) {
3865                         if (skipslave == old_arr->arr[idx]) {
3866                                 old_arr->arr[idx] =
3867                                     old_arr->arr[old_arr->count-1];
3868                                 old_arr->count--;
3869                                 break;
3870                         }
3871                 }
3872         }
3873         return ret;
3874 }
3875
3876 /* Use this Xmit function for 3AD as well as XOR modes. The current
3877  * usable slave array is formed in the control path. The xmit function
3878  * just calculates hash and sends the packet out.
3879  */
3880 static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3881 {
3882         struct bonding *bond = netdev_priv(dev);
3883         struct slave *slave;
3884         struct bond_up_slave *slaves;
3885         unsigned int count;
3886
3887         slaves = rcu_dereference(bond->slave_arr);
3888         count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3889         if (likely(count)) {
3890                 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3891                 bond_dev_queue_xmit(bond, skb, slave->dev);
3892         } else {
3893                 bond_tx_drop(dev, skb);
3894         }
3895
3896         return NETDEV_TX_OK;
3897 }
3898
3899 /* in broadcast mode, we send everything to all usable interfaces. */
3900 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3901 {
3902         struct bonding *bond = netdev_priv(bond_dev);
3903         struct slave *slave = NULL;
3904         struct list_head *iter;
3905
3906         bond_for_each_slave_rcu(bond, slave, iter) {
3907                 if (bond_is_last_slave(bond, slave))
3908                         break;
3909                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3910                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3911
3912                         if (!skb2) {
3913                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3914                                                     bond_dev->name, __func__);
3915                                 continue;
3916                         }
3917                         bond_dev_queue_xmit(bond, skb2, slave->dev);
3918                 }
3919         }
3920         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3921                 bond_dev_queue_xmit(bond, skb, slave->dev);
3922         else
3923                 bond_tx_drop(bond_dev, skb);
3924
3925         return NETDEV_TX_OK;
3926 }
3927
3928 /*------------------------- Device initialization ---------------------------*/
3929
3930 /* Lookup the slave that corresponds to a qid */
3931 static inline int bond_slave_override(struct bonding *bond,
3932                                       struct sk_buff *skb)
3933 {
3934         struct slave *slave = NULL;
3935         struct list_head *iter;
3936
3937         if (!skb->queue_mapping)
3938                 return 1;
3939
3940         /* Find out if any slaves have the same mapping as this skb. */
3941         bond_for_each_slave_rcu(bond, slave, iter) {
3942                 if (slave->queue_id == skb->queue_mapping) {
3943                         if (bond_slave_is_up(slave) &&
3944                             slave->link == BOND_LINK_UP) {
3945                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3946                                 return 0;
3947                         }
3948                         /* If the slave isn't UP, use default transmit policy. */
3949                         break;
3950                 }
3951         }
3952
3953         return 1;
3954 }
3955
3956
3957 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3958                              void *accel_priv, select_queue_fallback_t fallback)
3959 {
3960         /* This helper function exists to help dev_pick_tx get the correct
3961          * destination queue.  Using a helper function skips a call to
3962          * skb_tx_hash and will put the skbs in the queue we expect on their
3963          * way down to the bonding driver.
3964          */
3965         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3966
3967         /* Save the original txq to restore before passing to the driver */
3968         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3969
3970         if (unlikely(txq >= dev->real_num_tx_queues)) {
3971                 do {
3972                         txq -= dev->real_num_tx_queues;
3973                 } while (txq >= dev->real_num_tx_queues);
3974         }
3975         return txq;
3976 }
3977
3978 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3979 {
3980         struct bonding *bond = netdev_priv(dev);
3981
3982         if (bond_should_override_tx_queue(bond) &&
3983             !bond_slave_override(bond, skb))
3984                 return NETDEV_TX_OK;
3985
3986         switch (BOND_MODE(bond)) {
3987         case BOND_MODE_ROUNDROBIN:
3988                 return bond_xmit_roundrobin(skb, dev);
3989         case BOND_MODE_ACTIVEBACKUP:
3990                 return bond_xmit_activebackup(skb, dev);
3991         case BOND_MODE_8023AD:
3992         case BOND_MODE_XOR:
3993                 return bond_3ad_xor_xmit(skb, dev);
3994         case BOND_MODE_BROADCAST:
3995                 return bond_xmit_broadcast(skb, dev);
3996         case BOND_MODE_ALB:
3997                 return bond_alb_xmit(skb, dev);
3998         case BOND_MODE_TLB:
3999                 return bond_tlb_xmit(skb, dev);
4000         default:
4001                 /* Should never happen, mode already checked */
4002                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4003                 WARN_ON_ONCE(1);
4004                 bond_tx_drop(dev, skb);
4005                 return NETDEV_TX_OK;
4006         }
4007 }
4008
4009 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4010 {
4011         struct bonding *bond = netdev_priv(dev);
4012         netdev_tx_t ret = NETDEV_TX_OK;
4013
4014         /* If we risk deadlock from transmitting this in the
4015          * netpoll path, tell netpoll to queue the frame for later tx
4016          */
4017         if (unlikely(is_netpoll_tx_blocked(dev)))
4018                 return NETDEV_TX_BUSY;
4019
4020         rcu_read_lock();
4021         if (bond_has_slaves(bond))
4022                 ret = __bond_start_xmit(skb, dev);
4023         else
4024                 bond_tx_drop(dev, skb);
4025         rcu_read_unlock();
4026
4027         return ret;
4028 }
4029
4030 static int bond_ethtool_get_settings(struct net_device *bond_dev,
4031                                      struct ethtool_cmd *ecmd)
4032 {
4033         struct bonding *bond = netdev_priv(bond_dev);
4034         unsigned long speed = 0;
4035         struct list_head *iter;
4036         struct slave *slave;
4037
4038         ecmd->duplex = DUPLEX_UNKNOWN;
4039         ecmd->port = PORT_OTHER;
4040
4041         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4042          * do not need to check mode.  Though link speed might not represent
4043          * the true receive or transmit bandwidth (not all modes are symmetric)
4044          * this is an accurate maximum.
4045          */
4046         bond_for_each_slave(bond, slave, iter) {
4047                 if (bond_slave_can_tx(slave)) {
4048                         if (slave->speed != SPEED_UNKNOWN)
4049                                 speed += slave->speed;
4050                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
4051                             slave->duplex != DUPLEX_UNKNOWN)
4052                                 ecmd->duplex = slave->duplex;
4053                 }
4054         }
4055         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
4056
4057         return 0;
4058 }
4059
4060 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4061                                      struct ethtool_drvinfo *drvinfo)
4062 {
4063         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4064         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4065         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4066                  BOND_ABI_VERSION);
4067 }
4068
4069 static const struct ethtool_ops bond_ethtool_ops = {
4070         .get_drvinfo            = bond_ethtool_get_drvinfo,
4071         .get_settings           = bond_ethtool_get_settings,
4072         .get_link               = ethtool_op_get_link,
4073 };
4074
4075 static const struct net_device_ops bond_netdev_ops = {
4076         .ndo_init               = bond_init,
4077         .ndo_uninit             = bond_uninit,
4078         .ndo_open               = bond_open,
4079         .ndo_stop               = bond_close,
4080         .ndo_start_xmit         = bond_start_xmit,
4081         .ndo_select_queue       = bond_select_queue,
4082         .ndo_get_stats64        = bond_get_stats,
4083         .ndo_do_ioctl           = bond_do_ioctl,
4084         .ndo_change_rx_flags    = bond_change_rx_flags,
4085         .ndo_set_rx_mode        = bond_set_rx_mode,
4086         .ndo_change_mtu         = bond_change_mtu,
4087         .ndo_set_mac_address    = bond_set_mac_address,
4088         .ndo_neigh_setup        = bond_neigh_setup,
4089         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4090         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4091 #ifdef CONFIG_NET_POLL_CONTROLLER
4092         .ndo_netpoll_setup      = bond_netpoll_setup,
4093         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4094         .ndo_poll_controller    = bond_poll_controller,
4095 #endif
4096         .ndo_add_slave          = bond_enslave,
4097         .ndo_del_slave          = bond_release,
4098         .ndo_fix_features       = bond_fix_features,
4099         .ndo_bridge_setlink     = switchdev_port_bridge_setlink,
4100         .ndo_bridge_getlink     = switchdev_port_bridge_getlink,
4101         .ndo_bridge_dellink     = switchdev_port_bridge_dellink,
4102         .ndo_fdb_add            = switchdev_port_fdb_add,
4103         .ndo_fdb_del            = switchdev_port_fdb_del,
4104         .ndo_fdb_dump           = switchdev_port_fdb_dump,
4105         .ndo_features_check     = passthru_features_check,
4106 };
4107
4108 static const struct device_type bond_type = {
4109         .name = "bond",
4110 };
4111
4112 static void bond_destructor(struct net_device *bond_dev)
4113 {
4114         struct bonding *bond = netdev_priv(bond_dev);
4115         if (bond->wq)
4116                 destroy_workqueue(bond->wq);
4117         free_netdev(bond_dev);
4118 }
4119
4120 void bond_setup(struct net_device *bond_dev)
4121 {
4122         struct bonding *bond = netdev_priv(bond_dev);
4123
4124         spin_lock_init(&bond->mode_lock);
4125         spin_lock_init(&bond->stats_lock);
4126         bond->params = bonding_defaults;
4127
4128         /* Initialize pointers */
4129         bond->dev = bond_dev;
4130
4131         /* Initialize the device entry points */
4132         ether_setup(bond_dev);
4133         bond_dev->netdev_ops = &bond_netdev_ops;
4134         bond_dev->ethtool_ops = &bond_ethtool_ops;
4135
4136         bond_dev->destructor = bond_destructor;
4137
4138         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4139
4140         /* Initialize the device options */
4141         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4142         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4143         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4144
4145         /* don't acquire bond device's netif_tx_lock when transmitting */
4146         bond_dev->features |= NETIF_F_LLTX;
4147
4148         /* By default, we declare the bond to be fully
4149          * VLAN hardware accelerated capable. Special
4150          * care is taken in the various xmit functions
4151          * when there are slaves that are not hw accel
4152          * capable
4153          */
4154
4155         /* Don't allow bond devices to change network namespaces. */
4156         bond_dev->features |= NETIF_F_NETNS_LOCAL;
4157
4158         bond_dev->hw_features = BOND_VLAN_FEATURES |
4159                                 NETIF_F_HW_VLAN_CTAG_TX |
4160                                 NETIF_F_HW_VLAN_CTAG_RX |
4161                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4162
4163         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4164         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4165         bond_dev->features |= bond_dev->hw_features;
4166 }
4167
4168 /* Destroy a bonding device.
4169  * Must be under rtnl_lock when this function is called.
4170  */
4171 static void bond_uninit(struct net_device *bond_dev)
4172 {
4173         struct bonding *bond = netdev_priv(bond_dev);
4174         struct list_head *iter;
4175         struct slave *slave;
4176         struct bond_up_slave *arr;
4177
4178         bond_netpoll_cleanup(bond_dev);
4179
4180         /* Release the bonded slaves */
4181         bond_for_each_slave(bond, slave, iter)
4182                 __bond_release_one(bond_dev, slave->dev, true);
4183         netdev_info(bond_dev, "Released all slaves\n");
4184
4185         arr = rtnl_dereference(bond->slave_arr);
4186         if (arr) {
4187                 RCU_INIT_POINTER(bond->slave_arr, NULL);
4188                 kfree_rcu(arr, rcu);
4189         }
4190
4191         list_del(&bond->bond_list);
4192
4193         bond_debug_unregister(bond);
4194 }
4195
4196 /*------------------------- Module initialization ---------------------------*/
4197
4198 static int bond_check_params(struct bond_params *params)
4199 {
4200         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4201         struct bond_opt_value newval;
4202         const struct bond_opt_value *valptr;
4203         int arp_all_targets_value;
4204         u16 ad_actor_sys_prio = 0;
4205         u16 ad_user_port_key = 0;
4206
4207         /* Convert string parameters. */
4208         if (mode) {
4209                 bond_opt_initstr(&newval, mode);
4210                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4211                 if (!valptr) {
4212                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4213                         return -EINVAL;
4214                 }
4215                 bond_mode = valptr->value;
4216         }
4217
4218         if (xmit_hash_policy) {
4219                 if ((bond_mode != BOND_MODE_XOR) &&
4220                     (bond_mode != BOND_MODE_8023AD) &&
4221                     (bond_mode != BOND_MODE_TLB)) {
4222                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4223                                 bond_mode_name(bond_mode));
4224                 } else {
4225                         bond_opt_initstr(&newval, xmit_hash_policy);
4226                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4227                                                 &newval);
4228                         if (!valptr) {
4229                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4230                                        xmit_hash_policy);
4231                                 return -EINVAL;
4232                         }
4233                         xmit_hashtype = valptr->value;
4234                 }
4235         }
4236
4237         if (lacp_rate) {
4238                 if (bond_mode != BOND_MODE_8023AD) {
4239                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4240                                 bond_mode_name(bond_mode));
4241                 } else {
4242                         bond_opt_initstr(&newval, lacp_rate);
4243                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4244                                                 &newval);
4245                         if (!valptr) {
4246                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4247                                        lacp_rate);
4248                                 return -EINVAL;
4249                         }
4250                         lacp_fast = valptr->value;
4251                 }
4252         }
4253
4254         if (ad_select) {
4255                 bond_opt_initstr(&newval, ad_select);
4256                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4257                                         &newval);
4258                 if (!valptr) {
4259                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4260                         return -EINVAL;
4261                 }
4262                 params->ad_select = valptr->value;
4263                 if (bond_mode != BOND_MODE_8023AD)
4264                         pr_warn("ad_select param only affects 802.3ad mode\n");
4265         } else {
4266                 params->ad_select = BOND_AD_STABLE;
4267         }
4268
4269         if (max_bonds < 0) {
4270                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4271                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4272                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4273         }
4274
4275         if (miimon < 0) {
4276                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4277                         miimon, INT_MAX);
4278                 miimon = 0;
4279         }
4280
4281         if (updelay < 0) {
4282                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4283                         updelay, INT_MAX);
4284                 updelay = 0;
4285         }
4286
4287         if (downdelay < 0) {
4288                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4289                         downdelay, INT_MAX);
4290                 downdelay = 0;
4291         }
4292
4293         if ((use_carrier != 0) && (use_carrier != 1)) {
4294                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4295                         use_carrier);
4296                 use_carrier = 1;
4297         }
4298
4299         if (num_peer_notif < 0 || num_peer_notif > 255) {
4300                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4301                         num_peer_notif);
4302                 num_peer_notif = 1;
4303         }
4304
4305         /* reset values for 802.3ad/TLB/ALB */
4306         if (!bond_mode_uses_arp(bond_mode)) {
4307                 if (!miimon) {
4308                         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");
4309                         pr_warn("Forcing miimon to 100msec\n");
4310                         miimon = BOND_DEFAULT_MIIMON;
4311                 }
4312         }
4313
4314         if (tx_queues < 1 || tx_queues > 255) {
4315                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4316                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4317                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4318         }
4319
4320         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4321                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4322                         all_slaves_active);
4323                 all_slaves_active = 0;
4324         }
4325
4326         if (resend_igmp < 0 || resend_igmp > 255) {
4327                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4328                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4329                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4330         }
4331
4332         bond_opt_initval(&newval, packets_per_slave);
4333         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4334                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4335                         packets_per_slave, USHRT_MAX);
4336                 packets_per_slave = 1;
4337         }
4338
4339         if (bond_mode == BOND_MODE_ALB) {
4340                 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",
4341                           updelay);
4342         }
4343
4344         if (!miimon) {
4345                 if (updelay || downdelay) {
4346                         /* just warn the user the up/down delay will have
4347                          * no effect since miimon is zero...
4348                          */
4349                         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",
4350                                 updelay, downdelay);
4351                 }
4352         } else {
4353                 /* don't allow arp monitoring */
4354                 if (arp_interval) {
4355                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4356                                 miimon, arp_interval);
4357                         arp_interval = 0;
4358                 }
4359
4360                 if ((updelay % miimon) != 0) {
4361                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4362                                 updelay, miimon, (updelay / miimon) * miimon);
4363                 }
4364
4365                 updelay /= miimon;
4366
4367                 if ((downdelay % miimon) != 0) {
4368                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4369                                 downdelay, miimon,
4370                                 (downdelay / miimon) * miimon);
4371                 }
4372
4373                 downdelay /= miimon;
4374         }
4375
4376         if (arp_interval < 0) {
4377                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4378                         arp_interval, INT_MAX);
4379                 arp_interval = 0;
4380         }
4381
4382         for (arp_ip_count = 0, i = 0;
4383              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4384                 __be32 ip;
4385
4386                 /* not a complete check, but good enough to catch mistakes */
4387                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4388                     !bond_is_ip_target_ok(ip)) {
4389                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4390                                 arp_ip_target[i]);
4391                         arp_interval = 0;
4392                 } else {
4393                         if (bond_get_targets_ip(arp_target, ip) == -1)
4394                                 arp_target[arp_ip_count++] = ip;
4395                         else
4396                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4397                                         &ip);
4398                 }
4399         }
4400
4401         if (arp_interval && !arp_ip_count) {
4402                 /* don't allow arping if no arp_ip_target given... */
4403                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4404                         arp_interval);
4405                 arp_interval = 0;
4406         }
4407
4408         if (arp_validate) {
4409                 if (!arp_interval) {
4410                         pr_err("arp_validate requires arp_interval\n");
4411                         return -EINVAL;
4412                 }
4413
4414                 bond_opt_initstr(&newval, arp_validate);
4415                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4416                                         &newval);
4417                 if (!valptr) {
4418                         pr_err("Error: invalid arp_validate \"%s\"\n",
4419                                arp_validate);
4420                         return -EINVAL;
4421                 }
4422                 arp_validate_value = valptr->value;
4423         } else {
4424                 arp_validate_value = 0;
4425         }
4426
4427         arp_all_targets_value = 0;
4428         if (arp_all_targets) {
4429                 bond_opt_initstr(&newval, arp_all_targets);
4430                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4431                                         &newval);
4432                 if (!valptr) {
4433                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4434                                arp_all_targets);
4435                         arp_all_targets_value = 0;
4436                 } else {
4437                         arp_all_targets_value = valptr->value;
4438                 }
4439         }
4440
4441         if (miimon) {
4442                 pr_info("MII link monitoring set to %d ms\n", miimon);
4443         } else if (arp_interval) {
4444                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4445                                           arp_validate_value);
4446                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4447                         arp_interval, valptr->string, arp_ip_count);
4448
4449                 for (i = 0; i < arp_ip_count; i++)
4450                         pr_cont(" %s", arp_ip_target[i]);
4451
4452                 pr_cont("\n");
4453
4454         } else if (max_bonds) {
4455                 /* miimon and arp_interval not set, we need one so things
4456                  * work as expected, see bonding.txt for details
4457                  */
4458                 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");
4459         }
4460
4461         if (primary && !bond_mode_uses_primary(bond_mode)) {
4462                 /* currently, using a primary only makes sense
4463                  * in active backup, TLB or ALB modes
4464                  */
4465                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4466                         primary, bond_mode_name(bond_mode));
4467                 primary = NULL;
4468         }
4469
4470         if (primary && primary_reselect) {
4471                 bond_opt_initstr(&newval, primary_reselect);
4472                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4473                                         &newval);
4474                 if (!valptr) {
4475                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4476                                primary_reselect);
4477                         return -EINVAL;
4478                 }
4479                 primary_reselect_value = valptr->value;
4480         } else {
4481                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4482         }
4483
4484         if (fail_over_mac) {
4485                 bond_opt_initstr(&newval, fail_over_mac);
4486                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4487                                         &newval);
4488                 if (!valptr) {
4489                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4490                                fail_over_mac);
4491                         return -EINVAL;
4492                 }
4493                 fail_over_mac_value = valptr->value;
4494                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4495                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4496         } else {
4497                 fail_over_mac_value = BOND_FOM_NONE;
4498         }
4499
4500         bond_opt_initstr(&newval, "default");
4501         valptr = bond_opt_parse(
4502                         bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4503                                      &newval);
4504         if (!valptr) {
4505                 pr_err("Error: No ad_actor_sys_prio default value");
4506                 return -EINVAL;
4507         }
4508         ad_actor_sys_prio = valptr->value;
4509
4510         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4511                                 &newval);
4512         if (!valptr) {
4513                 pr_err("Error: No ad_user_port_key default value");
4514                 return -EINVAL;
4515         }
4516         ad_user_port_key = valptr->value;
4517
4518         if (lp_interval == 0) {
4519                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4520                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4521                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4522         }
4523
4524         /* fill params struct with the proper values */
4525         params->mode = bond_mode;
4526         params->xmit_policy = xmit_hashtype;
4527         params->miimon = miimon;
4528         params->num_peer_notif = num_peer_notif;
4529         params->arp_interval = arp_interval;
4530         params->arp_validate = arp_validate_value;
4531         params->arp_all_targets = arp_all_targets_value;
4532         params->updelay = updelay;
4533         params->downdelay = downdelay;
4534         params->use_carrier = use_carrier;
4535         params->lacp_fast = lacp_fast;
4536         params->primary[0] = 0;
4537         params->primary_reselect = primary_reselect_value;
4538         params->fail_over_mac = fail_over_mac_value;
4539         params->tx_queues = tx_queues;
4540         params->all_slaves_active = all_slaves_active;
4541         params->resend_igmp = resend_igmp;
4542         params->min_links = min_links;
4543         params->lp_interval = lp_interval;
4544         params->packets_per_slave = packets_per_slave;
4545         params->tlb_dynamic_lb = 1; /* Default value */
4546         params->ad_actor_sys_prio = ad_actor_sys_prio;
4547         eth_zero_addr(params->ad_actor_system);
4548         params->ad_user_port_key = ad_user_port_key;
4549         if (packets_per_slave > 0) {
4550                 params->reciprocal_packets_per_slave =
4551                         reciprocal_value(packets_per_slave);
4552         } else {
4553                 /* reciprocal_packets_per_slave is unused if
4554                  * packets_per_slave is 0 or 1, just initialize it
4555                  */
4556                 params->reciprocal_packets_per_slave =
4557                         (struct reciprocal_value) { 0 };
4558         }
4559
4560         if (primary) {
4561                 strncpy(params->primary, primary, IFNAMSIZ);
4562                 params->primary[IFNAMSIZ - 1] = 0;
4563         }
4564
4565         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4566
4567         return 0;
4568 }
4569
4570 static struct lock_class_key bonding_netdev_xmit_lock_key;
4571 static struct lock_class_key bonding_netdev_addr_lock_key;
4572 static struct lock_class_key bonding_tx_busylock_key;
4573
4574 static void bond_set_lockdep_class_one(struct net_device *dev,
4575                                        struct netdev_queue *txq,
4576                                        void *_unused)
4577 {
4578         lockdep_set_class(&txq->_xmit_lock,
4579                           &bonding_netdev_xmit_lock_key);
4580 }
4581
4582 static void bond_set_lockdep_class(struct net_device *dev)
4583 {
4584         lockdep_set_class(&dev->addr_list_lock,
4585                           &bonding_netdev_addr_lock_key);
4586         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4587         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4588 }
4589
4590 /* Called from registration process */
4591 static int bond_init(struct net_device *bond_dev)
4592 {
4593         struct bonding *bond = netdev_priv(bond_dev);
4594         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4595
4596         netdev_dbg(bond_dev, "Begin bond_init\n");
4597
4598         bond->wq = create_singlethread_workqueue(bond_dev->name);
4599         if (!bond->wq)
4600                 return -ENOMEM;
4601
4602         bond_set_lockdep_class(bond_dev);
4603
4604         list_add_tail(&bond->bond_list, &bn->dev_list);
4605
4606         bond_prepare_sysfs_group(bond);
4607
4608         bond_debug_register(bond);
4609
4610         /* Ensure valid dev_addr */
4611         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4612             bond_dev->addr_assign_type == NET_ADDR_PERM)
4613                 eth_hw_addr_random(bond_dev);
4614
4615         return 0;
4616 }
4617
4618 unsigned int bond_get_num_tx_queues(void)
4619 {
4620         return tx_queues;
4621 }
4622
4623 /* Create a new bond based on the specified name and bonding parameters.
4624  * If name is NULL, obtain a suitable "bond%d" name for us.
4625  * Caller must NOT hold rtnl_lock; we need to release it here before we
4626  * set up our sysfs entries.
4627  */
4628 int bond_create(struct net *net, const char *name)
4629 {
4630         struct net_device *bond_dev;
4631         struct bonding *bond;
4632         struct alb_bond_info *bond_info;
4633         int res;
4634
4635         rtnl_lock();
4636
4637         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4638                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
4639                                    bond_setup, tx_queues);
4640         if (!bond_dev) {
4641                 pr_err("%s: eek! can't alloc netdev!\n", name);
4642                 rtnl_unlock();
4643                 return -ENOMEM;
4644         }
4645
4646         /*
4647          * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4648          * It is set to 0 by default which is wrong.
4649          */
4650         bond = netdev_priv(bond_dev);
4651         bond_info = &(BOND_ALB_INFO(bond));
4652         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4653
4654         dev_net_set(bond_dev, net);
4655         bond_dev->rtnl_link_ops = &bond_link_ops;
4656
4657         res = register_netdevice(bond_dev);
4658
4659         netif_carrier_off(bond_dev);
4660
4661         rtnl_unlock();
4662         if (res < 0)
4663                 bond_destructor(bond_dev);
4664         return res;
4665 }
4666
4667 static int __net_init bond_net_init(struct net *net)
4668 {
4669         struct bond_net *bn = net_generic(net, bond_net_id);
4670
4671         bn->net = net;
4672         INIT_LIST_HEAD(&bn->dev_list);
4673
4674         bond_create_proc_dir(bn);
4675         bond_create_sysfs(bn);
4676
4677         return 0;
4678 }
4679
4680 static void __net_exit bond_net_exit(struct net *net)
4681 {
4682         struct bond_net *bn = net_generic(net, bond_net_id);
4683         struct bonding *bond, *tmp_bond;
4684         LIST_HEAD(list);
4685
4686         bond_destroy_sysfs(bn);
4687
4688         /* Kill off any bonds created after unregistering bond rtnl ops */
4689         rtnl_lock();
4690         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4691                 unregister_netdevice_queue(bond->dev, &list);
4692         unregister_netdevice_many(&list);
4693         rtnl_unlock();
4694
4695         bond_destroy_proc_dir(bn);
4696 }
4697
4698 static struct pernet_operations bond_net_ops = {
4699         .init = bond_net_init,
4700         .exit = bond_net_exit,
4701         .id   = &bond_net_id,
4702         .size = sizeof(struct bond_net),
4703 };
4704
4705 static int __init bonding_init(void)
4706 {
4707         int i;
4708         int res;
4709
4710         pr_info("%s", bond_version);
4711
4712         res = bond_check_params(&bonding_defaults);
4713         if (res)
4714                 goto out;
4715
4716         res = register_pernet_subsys(&bond_net_ops);
4717         if (res)
4718                 goto out;
4719
4720         res = bond_netlink_init();
4721         if (res)
4722                 goto err_link;
4723
4724         bond_create_debugfs();
4725
4726         for (i = 0; i < max_bonds; i++) {
4727                 res = bond_create(&init_net, NULL);
4728                 if (res)
4729                         goto err;
4730         }
4731
4732         register_netdevice_notifier(&bond_netdev_notifier);
4733 out:
4734         return res;
4735 err:
4736         bond_destroy_debugfs();
4737         bond_netlink_fini();
4738 err_link:
4739         unregister_pernet_subsys(&bond_net_ops);
4740         goto out;
4741
4742 }
4743
4744 static void __exit bonding_exit(void)
4745 {
4746         unregister_netdevice_notifier(&bond_netdev_notifier);
4747
4748         bond_destroy_debugfs();
4749
4750         bond_netlink_fini();
4751         unregister_pernet_subsys(&bond_net_ops);
4752
4753 #ifdef CONFIG_NET_POLL_CONTROLLER
4754         /* Make sure we don't have an imbalance on our netpoll blocking */
4755         WARN_ON(atomic_read(&netpoll_block_tx));
4756 #endif
4757 }
4758
4759 module_init(bonding_init);
4760 module_exit(bonding_exit);
4761 MODULE_LICENSE("GPL");
4762 MODULE_VERSION(DRV_VERSION);
4763 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4764 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");