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_keys.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 /*---------------------------- Module parameters ----------------------------*/
  86
  87 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  88
  89 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  90 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  91 static int num_peer_notif = 1;
  92 static int miimon;
  93 static int updelay;
  94 static int downdelay;
  95 static int use_carrier  = 1;
  96 static char *mode;
  97 static char *primary;
  98 static char *primary_reselect;
  99 static char *lacp_rate;
 100 static int min_links;
 101 static char *ad_select;
 102 static char *xmit_hash_policy;
 103 static int arp_interval;
 104 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 105 static char *arp_validate;
 106 static char *arp_all_targets;
 107 static char *fail_over_mac;
 108 static int all_slaves_active;
 109 static struct bond_params bonding_defaults;
 110 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 111 static int packets_per_slave = 1;
 112 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
 113
 114 module_param(max_bonds, int, 0);
 115 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 116 module_param(tx_queues, int, 0);
 117 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 118 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 119 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 120                                "failover event (alias of num_unsol_na)");
 121 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 122 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 123                                "failover event (alias of num_grat_arp)");
 124 module_param(miimon, int, 0);
 125 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 126 module_param(updelay, int, 0);
 127 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 128 module_param(downdelay, int, 0);
 129 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 130                             "in milliseconds");
 131 module_param(use_carrier, int, 0);
 132 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 133                               "0 for off, 1 for on (default)");
 134 module_param(mode, charp, 0);
 135 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 136                        "1 for active-backup, 2 for balance-xor, "
 137                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 138                        "6 for balance-alb");
 139 module_param(primary, charp, 0);
 140 MODULE_PARM_DESC(primary, "Primary network device to use");
 141 module_param(primary_reselect, charp, 0);
 142 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 143                                    "once it comes up; "
 144                                    "0 for always (default), "
 145                                    "1 for only if speed of primary is "
 146                                    "better, "
 147                                    "2 for only on active slave "
 148                                    "failure");
 149 module_param(lacp_rate, charp, 0);
 150 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 151                             "0 for slow, 1 for fast");
 152 module_param(ad_select, charp, 0);
 153 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 154                             "0 for stable (default), 1 for bandwidth, "
 155                             "2 for count");
 156 module_param(min_links, int, 0);
 157 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 158
 159 module_param(xmit_hash_policy, charp, 0);
 160 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 161                                    "0 for layer 2 (default), 1 for layer 3+4, "
 162                                    "2 for layer 2+3, 3 for encap layer 2+3, "
 163                                    "4 for encap layer 3+4");
 164 module_param(arp_interval, int, 0);
 165 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 166 module_param_array(arp_ip_target, charp, NULL, 0);
 167 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 168 module_param(arp_validate, charp, 0);
 169 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 170                                "0 for none (default), 1 for active, "
 171                                "2 for backup, 3 for all");
 172 module_param(arp_all_targets, charp, 0);
 173 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 174 module_param(fail_over_mac, charp, 0);
 175 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 176                                 "the same MAC; 0 for none (default), "
 177                                 "1 for active, 2 for follow");
 178 module_param(all_slaves_active, int, 0);
 179 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
 180                                      "by setting active flag for all slaves; "
 181                                      "0 for never (default), 1 for always.");
 182 module_param(resend_igmp, int, 0);
 183 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 184                               "link failure");
 185 module_param(packets_per_slave, int, 0);
 186 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
 187                                     "mode; 0 for a random slave, 1 packet per "
 188                                     "slave (default), >1 packets per slave.");
 189 module_param(lp_interval, uint, 0);
 190 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
 191                               "the bonding driver sends learning packets to "
 192                               "each slaves peer switch. The default is 1.");
 193
 194 /*----------------------------- Global variables ----------------------------*/
 195
 196 #ifdef CONFIG_NET_POLL_CONTROLLER
 197 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 198 #endif
 199
 200 int bond_net_id __read_mostly;
 201
 202 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 203 static int arp_ip_count;
 204 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 205 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 206 static int lacp_fast;
 207
 208 /*-------------------------- Forward declarations ---------------------------*/
 209
 210 static int bond_init(struct net_device *bond_dev);
 211 static void bond_uninit(struct net_device *bond_dev);
 212 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
 213                                                 struct rtnl_link_stats64 *stats);
 214 static void bond_slave_arr_handler(struct work_struct *work);
 215
 216 /*---------------------------- General routines -----------------------------*/
 217
 218 const char *bond_mode_name(int mode)
 219 {
 220         static const char *names[] = {
 221                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 222                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 223                 [BOND_MODE_XOR] = "load balancing (xor)",
 224                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 225                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 226                 [BOND_MODE_TLB] = "transmit load balancing",
 227                 [BOND_MODE_ALB] = "adaptive load balancing",
 228         };
 229
 230         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 231                 return "unknown";
 232
 233         return names[mode];
 234 }
 235
 236 /*---------------------------------- VLAN -----------------------------------*/
 237
 238 /**
 239  * bond_dev_queue_xmit - Prepare skb for xmit.
 240  *
 241  * @bond: bond device that got this skb for tx.
 242  * @skb: hw accel VLAN tagged skb to transmit
 243  * @slave_dev: slave that is supposed to xmit this skbuff
 244  */
 245 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 246                         struct net_device *slave_dev)
 247 {
 248         skb->dev = slave_dev;
 249
 250         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 251                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 252         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 253
 254         if (unlikely(netpoll_tx_running(bond->dev)))
 255                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 256         else
 257                 dev_queue_xmit(skb);
 258 }
 259
 260 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 261  * We don't protect the slave list iteration with a lock because:
 262  * a. This operation is performed in IOCTL context,
 263  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 264  * c. Holding a lock with BH disabled while directly calling a base driver
 265  *    entry point is generally a BAD idea.
 266  *
 267  * The design of synchronization/protection for this operation in the 8021q
 268  * module is good for one or more VLAN devices over a single physical device
 269  * and cannot be extended for a teaming solution like bonding, so there is a
 270  * potential race condition here where a net device from the vlan group might
 271  * be referenced (either by a base driver or the 8021q code) while it is being
 272  * removed from the system. However, it turns out we're not making matters
 273  * worse, and if it works for regular VLAN usage it will work here too.
 274 */
 275
 276 /**
 277  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 278  * @bond_dev: bonding net device that got called
 279  * @vid: vlan id being added
 280  */
 281 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 282                                 __be16 proto, u16 vid)
 283 {
 284         struct bonding *bond = netdev_priv(bond_dev);
 285         struct slave *slave, *rollback_slave;
 286         struct list_head *iter;
 287         int res;
 288
 289         bond_for_each_slave(bond, slave, iter) {
 290                 res = vlan_vid_add(slave->dev, proto, vid);
 291                 if (res)
 292                         goto unwind;
 293         }
 294
 295         return 0;
 296
 297 unwind:
 298         /* unwind to the slave that failed */
 299         bond_for_each_slave(bond, rollback_slave, iter) {
 300                 if (rollback_slave == slave)
 301                         break;
 302
 303                 vlan_vid_del(rollback_slave->dev, proto, vid);
 304         }
 305
 306         return res;
 307 }
 308
 309 /**
 310  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 311  * @bond_dev: bonding net device that got called
 312  * @vid: vlan id being removed
 313  */
 314 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 315                                  __be16 proto, u16 vid)
 316 {
 317         struct bonding *bond = netdev_priv(bond_dev);
 318         struct list_head *iter;
 319         struct slave *slave;
 320
 321         bond_for_each_slave(bond, slave, iter)
 322                 vlan_vid_del(slave->dev, proto, vid);
 323
 324         if (bond_is_lb(bond))
 325                 bond_alb_clear_vlan(bond, vid);
 326
 327         return 0;
 328 }
 329
 330 /*------------------------------- Link status -------------------------------*/
 331
 332 /* Set the carrier state for the master according to the state of its
 333  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 334  * do special 802.3ad magic.
 335  *
 336  * Returns zero if carrier state does not change, nonzero if it does.
 337  */
 338 int bond_set_carrier(struct bonding *bond)
 339 {
 340         struct list_head *iter;
 341         struct slave *slave;
 342
 343         if (!bond_has_slaves(bond))
 344                 goto down;
 345
 346         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 347                 return bond_3ad_set_carrier(bond);
 348
 349         bond_for_each_slave(bond, slave, iter) {
 350                 if (slave->link == BOND_LINK_UP) {
 351                         if (!netif_carrier_ok(bond->dev)) {
 352                                 netif_carrier_on(bond->dev);
 353                                 return 1;
 354                         }
 355                         return 0;
 356                 }
 357         }
 358
 359 down:
 360         if (netif_carrier_ok(bond->dev)) {
 361                 netif_carrier_off(bond->dev);
 362                 return 1;
 363         }
 364         return 0;
 365 }
 366
 367 /* Get link speed and duplex from the slave's base driver
 368  * using ethtool. If for some reason the call fails or the
 369  * values are invalid, set speed and duplex to -1,
 370  * and return.
 371  */
 372 static void bond_update_speed_duplex(struct slave *slave)
 373 {
 374         struct net_device *slave_dev = slave->dev;
 375         struct ethtool_cmd ecmd;
 376         u32 slave_speed;
 377         int res;
 378
 379         slave->speed = SPEED_UNKNOWN;
 380         slave->duplex = DUPLEX_UNKNOWN;
 381
 382         res = __ethtool_get_settings(slave_dev, &ecmd);
 383         if (res < 0)
 384                 return;
 385
 386         slave_speed = ethtool_cmd_speed(&ecmd);
 387         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 388                 return;
 389
 390         switch (ecmd.duplex) {
 391         case DUPLEX_FULL:
 392         case DUPLEX_HALF:
 393                 break;
 394         default:
 395                 return;
 396         }
 397
 398         slave->speed = slave_speed;
 399         slave->duplex = ecmd.duplex;
 400
 401         return;
 402 }
 403
 404 const char *bond_slave_link_status(s8 link)
 405 {
 406         switch (link) {
 407         case BOND_LINK_UP:
 408                 return "up";
 409         case BOND_LINK_FAIL:
 410                 return "going down";
 411         case BOND_LINK_DOWN:
 412                 return "down";
 413         case BOND_LINK_BACK:
 414                 return "going back";
 415         default:
 416                 return "unknown";
 417         }
 418 }
 419
 420 /* if <dev> supports MII link status reporting, check its link status.
 421  *
 422  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 423  * depending upon the setting of the use_carrier parameter.
 424  *
 425  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 426  * can't tell and just pretend it is), or 0, meaning that the link is
 427  * down.
 428  *
 429  * If reporting is non-zero, instead of faking link up, return -1 if
 430  * both ETHTOOL and MII ioctls fail (meaning the device does not
 431  * support them).  If use_carrier is set, return whatever it says.
 432  * It'd be nice if there was a good way to tell if a driver supports
 433  * netif_carrier, but there really isn't.
 434  */
 435 static int bond_check_dev_link(struct bonding *bond,
 436                                struct net_device *slave_dev, int reporting)
 437 {
 438         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 439         int (*ioctl)(struct net_device *, struct ifreq *, int);
 440         struct ifreq ifr;
 441         struct mii_ioctl_data *mii;
 442
 443         if (!reporting && !netif_running(slave_dev))
 444                 return 0;
 445
 446         if (bond->params.use_carrier)
 447                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 448
 449         /* Try to get link status using Ethtool first. */
 450         if (slave_dev->ethtool_ops->get_link)
 451                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 452                         BMSR_LSTATUS : 0;
 453
 454         /* Ethtool can't be used, fallback to MII ioctls. */
 455         ioctl = slave_ops->ndo_do_ioctl;
 456         if (ioctl) {
 457                 /* TODO: set pointer to correct ioctl on a per team member
 458                  *       bases to make this more efficient. that is, once
 459                  *       we determine the correct ioctl, we will always
 460                  *       call it and not the others for that team
 461                  *       member.
 462                  */
 463
 464                 /* We cannot assume that SIOCGMIIPHY will also read a
 465                  * register; not all network drivers (e.g., e100)
 466                  * support that.
 467                  */
 468
 469                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 470                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 471                 mii = if_mii(&ifr);
 472                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 473                         mii->reg_num = MII_BMSR;
 474                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 475                                 return mii->val_out & BMSR_LSTATUS;
 476                 }
 477         }
 478
 479         /* If reporting, report that either there's no dev->do_ioctl,
 480          * or both SIOCGMIIREG and get_link failed (meaning that we
 481          * cannot report link status).  If not reporting, pretend
 482          * we're ok.
 483          */
 484         return reporting ? -1 : BMSR_LSTATUS;
 485 }
 486
 487 /*----------------------------- Multicast list ------------------------------*/
 488
 489 /* Push the promiscuity flag down to appropriate slaves */
 490 static int bond_set_promiscuity(struct bonding *bond, int inc)
 491 {
 492         struct list_head *iter;
 493         int err = 0;
 494
 495         if (bond_uses_primary(bond)) {
 496                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 497
 498                 if (curr_active)
 499                         err = dev_set_promiscuity(curr_active->dev, inc);
 500         } else {
 501                 struct slave *slave;
 502
 503                 bond_for_each_slave(bond, slave, iter) {
 504                         err = dev_set_promiscuity(slave->dev, inc);
 505                         if (err)
 506                                 return err;
 507                 }
 508         }
 509         return err;
 510 }
 511
 512 /* Push the allmulti flag down to all slaves */
 513 static int bond_set_allmulti(struct bonding *bond, int inc)
 514 {
 515         struct list_head *iter;
 516         int err = 0;
 517
 518         if (bond_uses_primary(bond)) {
 519                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 520
 521                 if (curr_active)
 522                         err = dev_set_allmulti(curr_active->dev, inc);
 523         } else {
 524                 struct slave *slave;
 525
 526                 bond_for_each_slave(bond, slave, iter) {
 527                         err = dev_set_allmulti(slave->dev, inc);
 528                         if (err)
 529                                 return err;
 530                 }
 531         }
 532         return err;
 533 }
 534
 535 /* Retrieve the list of registered multicast addresses for the bonding
 536  * device and retransmit an IGMP JOIN request to the current active
 537  * slave.
 538  */
 539 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 540 {
 541         struct bonding *bond = container_of(work, struct bonding,
 542                                             mcast_work.work);
 543
 544         if (!rtnl_trylock()) {
 545                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 546                 return;
 547         }
 548         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 549
 550         if (bond->igmp_retrans > 1) {
 551                 bond->igmp_retrans--;
 552                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 553         }
 554         rtnl_unlock();
 555 }
 556
 557 /* Flush bond's hardware addresses from slave */
 558 static void bond_hw_addr_flush(struct net_device *bond_dev,
 559                                struct net_device *slave_dev)
 560 {
 561         struct bonding *bond = netdev_priv(bond_dev);
 562
 563         dev_uc_unsync(slave_dev, bond_dev);
 564         dev_mc_unsync(slave_dev, bond_dev);
 565
 566         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 567                 /* del lacpdu mc addr from mc list */
 568                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 569
 570                 dev_mc_del(slave_dev, lacpdu_multicast);
 571         }
 572 }
 573
 574 /*--------------------------- Active slave change ---------------------------*/
 575
 576 /* Update the hardware address list and promisc/allmulti for the new and
 577  * old active slaves (if any).  Modes that are not using primary keep all
 578  * slaves up date at all times; only the modes that use primary need to call
 579  * this function to swap these settings during a failover.
 580  */
 581 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 582                               struct slave *old_active)
 583 {
 584         if (old_active) {
 585                 if (bond->dev->flags & IFF_PROMISC)
 586                         dev_set_promiscuity(old_active->dev, -1);
 587
 588                 if (bond->dev->flags & IFF_ALLMULTI)
 589                         dev_set_allmulti(old_active->dev, -1);
 590
 591                 bond_hw_addr_flush(bond->dev, old_active->dev);
 592         }
 593
 594         if (new_active) {
 595                 /* FIXME: Signal errors upstream. */
 596                 if (bond->dev->flags & IFF_PROMISC)
 597                         dev_set_promiscuity(new_active->dev, 1);
 598
 599                 if (bond->dev->flags & IFF_ALLMULTI)
 600                         dev_set_allmulti(new_active->dev, 1);
 601
 602                 netif_addr_lock_bh(bond->dev);
 603                 dev_uc_sync(new_active->dev, bond->dev);
 604                 dev_mc_sync(new_active->dev, bond->dev);
 605                 netif_addr_unlock_bh(bond->dev);
 606         }
 607 }
 608
 609 /**
 610  * bond_set_dev_addr - clone slave's address to bond
 611  * @bond_dev: bond net device
 612  * @slave_dev: slave net device
 613  *
 614  * Should be called with RTNL held.
 615  */
 616 static void bond_set_dev_addr(struct net_device *bond_dev,
 617                               struct net_device *slave_dev)
 618 {
 619         netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 620                    bond_dev, slave_dev, slave_dev->addr_len);
 621         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 622         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 623         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 624 }
 625
 626 /* bond_do_fail_over_mac
 627  *
 628  * Perform special MAC address swapping for fail_over_mac settings
 629  *
 630  * Called with RTNL
 631  */
 632 static void bond_do_fail_over_mac(struct bonding *bond,
 633                                   struct slave *new_active,
 634                                   struct slave *old_active)
 635 {
 636         u8 tmp_mac[ETH_ALEN];
 637         struct sockaddr saddr;
 638         int rv;
 639
 640         switch (bond->params.fail_over_mac) {
 641         case BOND_FOM_ACTIVE:
 642                 if (new_active)
 643                         bond_set_dev_addr(bond->dev, new_active->dev);
 644                 break;
 645         case BOND_FOM_FOLLOW:
 646                 /* if new_active && old_active, swap them
 647                  * if just old_active, do nothing (going to no active slave)
 648                  * if just new_active, set new_active to bond's MAC
 649                  */
 650                 if (!new_active)
 651                         return;
 652
 653                 if (old_active) {
 654                         ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
 655                         ether_addr_copy(saddr.sa_data,
 656                                         old_active->dev->dev_addr);
 657                         saddr.sa_family = new_active->dev->type;
 658                 } else {
 659                         ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
 660                         saddr.sa_family = bond->dev->type;
 661                 }
 662
 663                 rv = dev_set_mac_address(new_active->dev, &saddr);
 664                 if (rv) {
 665                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 666                                    -rv, new_active->dev->name);
 667                         goto out;
 668                 }
 669
 670                 if (!old_active)
 671                         goto out;
 672
 673                 ether_addr_copy(saddr.sa_data, tmp_mac);
 674                 saddr.sa_family = old_active->dev->type;
 675
 676                 rv = dev_set_mac_address(old_active->dev, &saddr);
 677                 if (rv)
 678                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 679                                    -rv, new_active->dev->name);
 680 out:
 681                 break;
 682         default:
 683                 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
 684                            bond->params.fail_over_mac);
 685                 break;
 686         }
 687
 688 }
 689
 690 static bool bond_should_change_active(struct bonding *bond)
 691 {
 692         struct slave *prim = rtnl_dereference(bond->primary_slave);
 693         struct slave *curr = rtnl_dereference(bond->curr_active_slave);
 694
 695         if (!prim || !curr || curr->link != BOND_LINK_UP)
 696                 return true;
 697         if (bond->force_primary) {
 698                 bond->force_primary = false;
 699                 return true;
 700         }
 701         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 702             (prim->speed < curr->speed ||
 703              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 704                 return false;
 705         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 706                 return false;
 707         return true;
 708 }
 709
 710 /**
 711  * find_best_interface - select the best available slave to be the active one
 712  * @bond: our bonding struct
 713  */
 714 static struct slave *bond_find_best_slave(struct bonding *bond)
 715 {
 716         struct slave *slave, *bestslave = NULL, *primary;
 717         struct list_head *iter;
 718         int mintime = bond->params.updelay;
 719
 720         primary = rtnl_dereference(bond->primary_slave);
 721         if (primary && primary->link == BOND_LINK_UP &&
 722             bond_should_change_active(bond))
 723                 return primary;
 724
 725         bond_for_each_slave(bond, slave, iter) {
 726                 if (slave->link == BOND_LINK_UP)
 727                         return slave;
 728                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
 729                     slave->delay < mintime) {
 730                         mintime = slave->delay;
 731                         bestslave = slave;
 732                 }
 733         }
 734
 735         return bestslave;
 736 }
 737
 738 static bool bond_should_notify_peers(struct bonding *bond)
 739 {
 740         struct slave *slave;
 741
 742         rcu_read_lock();
 743         slave = rcu_dereference(bond->curr_active_slave);
 744         rcu_read_unlock();
 745
 746         netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
 747                    slave ? slave->dev->name : "NULL");
 748
 749         if (!slave || !bond->send_peer_notif ||
 750             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 751                 return false;
 752
 753         return true;
 754 }
 755
 756 /**
 757  * change_active_interface - change the active slave into the specified one
 758  * @bond: our bonding struct
 759  * @new: the new slave to make the active one
 760  *
 761  * Set the new slave to the bond's settings and unset them on the old
 762  * curr_active_slave.
 763  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 764  *
 765  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 766  * because it is apparently the best available slave we have, even though its
 767  * updelay hasn't timed out yet.
 768  *
 769  * Caller must hold RTNL.
 770  */
 771 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 772 {
 773         struct slave *old_active;
 774
 775         ASSERT_RTNL();
 776
 777         old_active = rtnl_dereference(bond->curr_active_slave);
 778
 779         if (old_active == new_active)
 780                 return;
 781
 782         if (new_active) {
 783                 new_active->last_link_up = jiffies;
 784
 785                 if (new_active->link == BOND_LINK_BACK) {
 786                         if (bond_uses_primary(bond)) {
 787                                 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
 788                                             new_active->dev->name,
 789                                             (bond->params.updelay - new_active->delay) * bond->params.miimon);
 790                         }
 791
 792                         new_active->delay = 0;
 793                         bond_set_slave_link_state(new_active, BOND_LINK_UP);
 794
 795                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 796                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 797
 798                         if (bond_is_lb(bond))
 799                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 800                 } else {
 801                         if (bond_uses_primary(bond)) {
 802                                 netdev_info(bond->dev, "making interface %s the new active one\n",
 803                                             new_active->dev->name);
 804                         }
 805                 }
 806         }
 807
 808         if (bond_uses_primary(bond))
 809                 bond_hw_addr_swap(bond, new_active, old_active);
 810
 811         if (bond_is_lb(bond)) {
 812                 bond_alb_handle_active_change(bond, new_active);
 813                 if (old_active)
 814                         bond_set_slave_inactive_flags(old_active,
 815                                                       BOND_SLAVE_NOTIFY_NOW);
 816                 if (new_active)
 817                         bond_set_slave_active_flags(new_active,
 818                                                     BOND_SLAVE_NOTIFY_NOW);
 819         } else {
 820                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 821         }
 822
 823         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
 824                 if (old_active)
 825                         bond_set_slave_inactive_flags(old_active,
 826                                                       BOND_SLAVE_NOTIFY_NOW);
 827
 828                 if (new_active) {
 829                         bool should_notify_peers = false;
 830
 831                         bond_set_slave_active_flags(new_active,
 832                                                     BOND_SLAVE_NOTIFY_NOW);
 833
 834                         if (bond->params.fail_over_mac)
 835                                 bond_do_fail_over_mac(bond, new_active,
 836                                                       old_active);
 837
 838                         if (netif_running(bond->dev)) {
 839                                 bond->send_peer_notif =
 840                                         bond->params.num_peer_notif;
 841                                 should_notify_peers =
 842                                         bond_should_notify_peers(bond);
 843                         }
 844
 845                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 846                         if (should_notify_peers)
 847                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 848                                                          bond->dev);
 849                 }
 850         }
 851
 852         /* resend IGMP joins since active slave has changed or
 853          * all were sent on curr_active_slave.
 854          * resend only if bond is brought up with the affected
 855          * bonding modes and the retransmission is enabled
 856          */
 857         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 858             ((bond_uses_primary(bond) && new_active) ||
 859              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
 860                 bond->igmp_retrans = bond->params.resend_igmp;
 861                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 862         }
 863 }
 864
 865 /**
 866  * bond_select_active_slave - select a new active slave, if needed
 867  * @bond: our bonding struct
 868  *
 869  * This functions should be called when one of the following occurs:
 870  * - The old curr_active_slave has been released or lost its link.
 871  * - The primary_slave has got its link back.
 872  * - A slave has got its link back and there's no old curr_active_slave.
 873  *
 874  * Caller must hold RTNL.
 875  */
 876 void bond_select_active_slave(struct bonding *bond)
 877 {
 878         struct slave *best_slave;
 879         int rv;
 880
 881         ASSERT_RTNL();
 882
 883         best_slave = bond_find_best_slave(bond);
 884         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
 885                 bond_change_active_slave(bond, best_slave);
 886                 rv = bond_set_carrier(bond);
 887                 if (!rv)
 888                         return;
 889
 890                 if (netif_carrier_ok(bond->dev)) {
 891                         netdev_info(bond->dev, "first active interface up!\n");
 892                 } else {
 893                         netdev_info(bond->dev, "now running without any active interface!\n");
 894                 }
 895         }
 896 }
 897
 898 #ifdef CONFIG_NET_POLL_CONTROLLER
 899 static inline int slave_enable_netpoll(struct slave *slave)
 900 {
 901         struct netpoll *np;
 902         int err = 0;
 903
 904         np = kzalloc(sizeof(*np), GFP_KERNEL);
 905         err = -ENOMEM;
 906         if (!np)
 907                 goto out;
 908
 909         err = __netpoll_setup(np, slave->dev);
 910         if (err) {
 911                 kfree(np);
 912                 goto out;
 913         }
 914         slave->np = np;
 915 out:
 916         return err;
 917 }
 918 static inline void slave_disable_netpoll(struct slave *slave)
 919 {
 920         struct netpoll *np = slave->np;
 921
 922         if (!np)
 923                 return;
 924
 925         slave->np = NULL;
 926         __netpoll_free_async(np);
 927 }
 928
 929 static void bond_poll_controller(struct net_device *bond_dev)
 930 {
 931         struct bonding *bond = netdev_priv(bond_dev);
 932         struct slave *slave = NULL;
 933         struct list_head *iter;
 934         struct ad_info ad_info;
 935         struct netpoll_info *ni;
 936         const struct net_device_ops *ops;
 937
 938         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 939                 if (bond_3ad_get_active_agg_info(bond, &ad_info))
 940                         return;
 941
 942         rcu_read_lock_bh();
 943         bond_for_each_slave_rcu(bond, slave, iter) {
 944                 ops = slave->dev->netdev_ops;
 945                 if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller)
 946                         continue;
 947
 948                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 949                         struct aggregator *agg =
 950                             SLAVE_AD_INFO(slave)->port.aggregator;
 951
 952                         if (agg &&
 953                             agg->aggregator_identifier != ad_info.aggregator_id)
 954                                 continue;
 955                 }
 956
 957                 ni = rcu_dereference_bh(slave->dev->npinfo);
 958                 if (down_trylock(&ni->dev_lock))
 959                         continue;
 960                 ops->ndo_poll_controller(slave->dev);
 961                 up(&ni->dev_lock);
 962         }
 963         rcu_read_unlock_bh();
 964 }
 965
 966 static void bond_netpoll_cleanup(struct net_device *bond_dev)
 967 {
 968         struct bonding *bond = netdev_priv(bond_dev);
 969         struct list_head *iter;
 970         struct slave *slave;
 971
 972         bond_for_each_slave(bond, slave, iter)
 973                 if (bond_slave_is_up(slave))
 974                         slave_disable_netpoll(slave);
 975 }
 976
 977 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
 978 {
 979         struct bonding *bond = netdev_priv(dev);
 980         struct list_head *iter;
 981         struct slave *slave;
 982         int err = 0;
 983
 984         bond_for_each_slave(bond, slave, iter) {
 985                 err = slave_enable_netpoll(slave);
 986                 if (err) {
 987                         bond_netpoll_cleanup(dev);
 988                         break;
 989                 }
 990         }
 991         return err;
 992 }
 993 #else
 994 static inline int slave_enable_netpoll(struct slave *slave)
 995 {
 996         return 0;
 997 }
 998 static inline void slave_disable_netpoll(struct slave *slave)
 999 {
1000 }
1001 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1002 {
1003 }
1004 #endif
1005
1006 /*---------------------------------- IOCTL ----------------------------------*/
1007
1008 static netdev_features_t bond_fix_features(struct net_device *dev,
1009                                            netdev_features_t features)
1010 {
1011         struct bonding *bond = netdev_priv(dev);
1012         struct list_head *iter;
1013         netdev_features_t mask;
1014         struct slave *slave;
1015
1016         /* If any slave has the offload feature flag set,
1017          * set the offload flag on the bond.
1018          */
1019         mask = features | NETIF_F_HW_SWITCH_OFFLOAD;
1020
1021         features &= ~NETIF_F_ONE_FOR_ALL;
1022         features |= NETIF_F_ALL_FOR_ALL;
1023
1024         bond_for_each_slave(bond, slave, iter) {
1025                 features = netdev_increment_features(features,
1026                                                      slave->dev->features,
1027                                                      mask);
1028         }
1029         features = netdev_add_tso_features(features, mask);
1030
1031         return features;
1032 }
1033
1034 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1035                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1036                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1037
1038 #define BOND_ENC_FEATURES       (NETIF_F_ALL_CSUM | NETIF_F_SG | NETIF_F_RXCSUM |\
1039                                  NETIF_F_TSO)
1040
1041 static void bond_compute_features(struct bonding *bond)
1042 {
1043         unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1044                                         IFF_XMIT_DST_RELEASE_PERM;
1045         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1046         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1047         struct net_device *bond_dev = bond->dev;
1048         struct list_head *iter;
1049         struct slave *slave;
1050         unsigned short max_hard_header_len = ETH_HLEN;
1051         unsigned int gso_max_size = GSO_MAX_SIZE;
1052         u16 gso_max_segs = GSO_MAX_SEGS;
1053
1054         if (!bond_has_slaves(bond))
1055                 goto done;
1056         vlan_features &= NETIF_F_ALL_FOR_ALL;
1057
1058         bond_for_each_slave(bond, slave, iter) {
1059                 vlan_features = netdev_increment_features(vlan_features,
1060                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1061
1062                 enc_features = netdev_increment_features(enc_features,
1063                                                          slave->dev->hw_enc_features,
1064                                                          BOND_ENC_FEATURES);
1065                 dst_release_flag &= slave->dev->priv_flags;
1066                 if (slave->dev->hard_header_len > max_hard_header_len)
1067                         max_hard_header_len = slave->dev->hard_header_len;
1068
1069                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1070                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1071         }
1072
1073 done:
1074         bond_dev->vlan_features = vlan_features;
1075         bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
1076         bond_dev->hard_header_len = max_hard_header_len;
1077         bond_dev->gso_max_segs = gso_max_segs;
1078         netif_set_gso_max_size(bond_dev, gso_max_size);
1079
1080         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1081         if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1082             dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1083                 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1084
1085         netdev_change_features(bond_dev);
1086 }
1087
1088 static void bond_setup_by_slave(struct net_device *bond_dev,
1089                                 struct net_device *slave_dev)
1090 {
1091         bond_dev->header_ops        = slave_dev->header_ops;
1092
1093         bond_dev->type              = slave_dev->type;
1094         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1095         bond_dev->addr_len          = slave_dev->addr_len;
1096
1097         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1098                 slave_dev->addr_len);
1099 }
1100
1101 /* On bonding slaves other than the currently active slave, suppress
1102  * duplicates except for alb non-mcast/bcast.
1103  */
1104 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1105                                             struct slave *slave,
1106                                             struct bonding *bond)
1107 {
1108         if (bond_is_slave_inactive(slave)) {
1109                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1110                     skb->pkt_type != PACKET_BROADCAST &&
1111                     skb->pkt_type != PACKET_MULTICAST)
1112                         return false;
1113                 return true;
1114         }
1115         return false;
1116 }
1117
1118 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1119 {
1120         struct sk_buff *skb = *pskb;
1121         struct slave *slave;
1122         struct bonding *bond;
1123         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1124                           struct slave *);
1125         int ret = RX_HANDLER_ANOTHER;
1126
1127         skb = skb_share_check(skb, GFP_ATOMIC);
1128         if (unlikely(!skb))
1129                 return RX_HANDLER_CONSUMED;
1130
1131         *pskb = skb;
1132
1133         slave = bond_slave_get_rcu(skb->dev);
1134         bond = slave->bond;
1135
1136         recv_probe = ACCESS_ONCE(bond->recv_probe);
1137         if (recv_probe) {
1138                 ret = recv_probe(skb, bond, slave);
1139                 if (ret == RX_HANDLER_CONSUMED) {
1140                         consume_skb(skb);
1141                         return ret;
1142                 }
1143         }
1144
1145         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1146                 return RX_HANDLER_EXACT;
1147         }
1148
1149         skb->dev = bond->dev;
1150
1151         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1152             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1153             skb->pkt_type == PACKET_HOST) {
1154
1155                 if (unlikely(skb_cow_head(skb,
1156                                           skb->data - skb_mac_header(skb)))) {
1157                         kfree_skb(skb);
1158                         return RX_HANDLER_CONSUMED;
1159                 }
1160                 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1161         }
1162
1163         return ret;
1164 }
1165
1166 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1167                                       struct net_device *slave_dev,
1168                                       struct slave *slave)
1169 {
1170         int err;
1171
1172         err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1173         if (err)
1174                 return err;
1175         slave_dev->flags |= IFF_SLAVE;
1176         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1177         return 0;
1178 }
1179
1180 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1181                                   struct net_device *slave_dev)
1182 {
1183         netdev_upper_dev_unlink(slave_dev, bond_dev);
1184         slave_dev->flags &= ~IFF_SLAVE;
1185         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1186 }
1187
1188 static struct slave *bond_alloc_slave(struct bonding *bond)
1189 {
1190         struct slave *slave = NULL;
1191
1192         slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1193         if (!slave)
1194                 return NULL;
1195
1196         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1197                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1198                                                GFP_KERNEL);
1199                 if (!SLAVE_AD_INFO(slave)) {
1200                         kfree(slave);
1201                         return NULL;
1202                 }
1203         }
1204         return slave;
1205 }
1206
1207 static void bond_free_slave(struct slave *slave)
1208 {
1209         struct bonding *bond = bond_get_bond_by_slave(slave);
1210
1211         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1212                 kfree(SLAVE_AD_INFO(slave));
1213
1214         kfree(slave);
1215 }
1216
1217 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1218 {
1219         info->bond_mode = BOND_MODE(bond);
1220         info->miimon = bond->params.miimon;
1221         info->num_slaves = bond->slave_cnt;
1222 }
1223
1224 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1225 {
1226         strcpy(info->slave_name, slave->dev->name);
1227         info->link = slave->link;
1228         info->state = bond_slave_state(slave);
1229         info->link_failure_count = slave->link_failure_count;
1230 }
1231
1232 static void bond_netdev_notify(struct net_device *dev,
1233                                struct netdev_bonding_info *info)
1234 {
1235         rtnl_lock();
1236         netdev_bonding_info_change(dev, info);
1237         rtnl_unlock();
1238 }
1239
1240 static void bond_netdev_notify_work(struct work_struct *_work)
1241 {
1242         struct netdev_notify_work *w =
1243                 container_of(_work, struct netdev_notify_work, work.work);
1244
1245         bond_netdev_notify(w->dev, &w->bonding_info);
1246         dev_put(w->dev);
1247         kfree(w);
1248 }
1249
1250 void bond_queue_slave_event(struct slave *slave)
1251 {
1252         struct bonding *bond = slave->bond;
1253         struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1254
1255         if (!nnw)
1256                 return;
1257
1258         dev_hold(slave->dev);
1259         nnw->dev = slave->dev;
1260         bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1261         bond_fill_ifbond(bond, &nnw->bonding_info.master);
1262         INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1263
1264         queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1265 }
1266
1267 /* enslave device <slave> to bond device <master> */
1268 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1269 {
1270         struct bonding *bond = netdev_priv(bond_dev);
1271         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1272         struct slave *new_slave = NULL, *prev_slave;
1273         struct sockaddr addr;
1274         int link_reporting;
1275         int res = 0, i;
1276
1277         if (!bond->params.use_carrier &&
1278             slave_dev->ethtool_ops->get_link == NULL &&
1279             slave_ops->ndo_do_ioctl == NULL) {
1280                 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1281                             slave_dev->name);
1282         }
1283
1284         /* already enslaved */
1285         if (slave_dev->flags & IFF_SLAVE) {
1286                 netdev_dbg(bond_dev, "Error: Device was already enslaved\n");
1287                 return -EBUSY;
1288         }
1289
1290         if (bond_dev == slave_dev) {
1291                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1292                 return -EPERM;
1293         }
1294
1295         /* vlan challenged mutual exclusion */
1296         /* no need to lock since we're protected by rtnl_lock */
1297         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1298                 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1299                            slave_dev->name);
1300                 if (vlan_uses_dev(bond_dev)) {
1301                         netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1302                                    slave_dev->name, bond_dev->name);
1303                         return -EPERM;
1304                 } else {
1305                         netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1306                                     slave_dev->name, slave_dev->name,
1307                                     bond_dev->name);
1308                 }
1309         } else {
1310                 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1311                            slave_dev->name);
1312         }
1313
1314         /* Old ifenslave binaries are no longer supported.  These can
1315          * be identified with moderate accuracy by the state of the slave:
1316          * the current ifenslave will set the interface down prior to
1317          * enslaving it; the old ifenslave will not.
1318          */
1319         if ((slave_dev->flags & IFF_UP)) {
1320                 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1321                            slave_dev->name);
1322                 res = -EPERM;
1323                 goto err_undo_flags;
1324         }
1325
1326         /* set bonding device ether type by slave - bonding netdevices are
1327          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1328          * there is a need to override some of the type dependent attribs/funcs.
1329          *
1330          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1331          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1332          */
1333         if (!bond_has_slaves(bond)) {
1334                 if (bond_dev->type != slave_dev->type) {
1335                         netdev_dbg(bond_dev, "change device type from %d to %d\n",
1336                                    bond_dev->type, slave_dev->type);
1337
1338                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1339                                                        bond_dev);
1340                         res = notifier_to_errno(res);
1341                         if (res) {
1342                                 netdev_err(bond_dev, "refused to change device type\n");
1343                                 res = -EBUSY;
1344                                 goto err_undo_flags;
1345                         }
1346
1347                         /* Flush unicast and multicast addresses */
1348                         dev_uc_flush(bond_dev);
1349                         dev_mc_flush(bond_dev);
1350
1351                         if (slave_dev->type != ARPHRD_ETHER)
1352                                 bond_setup_by_slave(bond_dev, slave_dev);
1353                         else {
1354                                 ether_setup(bond_dev);
1355                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1356                         }
1357
1358                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1359                                                  bond_dev);
1360                 }
1361         } else if (bond_dev->type != slave_dev->type) {
1362                 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1363                            slave_dev->name, slave_dev->type, bond_dev->type);
1364                 res = -EINVAL;
1365                 goto err_undo_flags;
1366         }
1367
1368         if (slave_ops->ndo_set_mac_address == NULL) {
1369                 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1370                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1371                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1372                         if (!bond_has_slaves(bond)) {
1373                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1374                                 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1375                         } else {
1376                                 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");
1377                                 res = -EOPNOTSUPP;
1378                                 goto err_undo_flags;
1379                         }
1380                 }
1381         }
1382
1383         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1384
1385         /* If this is the first slave, then we need to set the master's hardware
1386          * address to be the same as the slave's.
1387          */
1388         if (!bond_has_slaves(bond) &&
1389             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1390                 bond_set_dev_addr(bond->dev, slave_dev);
1391
1392         new_slave = bond_alloc_slave(bond);
1393         if (!new_slave) {
1394                 res = -ENOMEM;
1395                 goto err_undo_flags;
1396         }
1397
1398         new_slave->bond = bond;
1399         new_slave->dev = slave_dev;
1400         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1401          * is set via sysfs or module option if desired.
1402          */
1403         new_slave->queue_id = 0;
1404
1405         /* Save slave's original mtu and then set it to match the bond */
1406         new_slave->original_mtu = slave_dev->mtu;
1407         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1408         if (res) {
1409                 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1410                 goto err_free;
1411         }
1412
1413         /* Save slave's original ("permanent") mac address for modes
1414          * that need it, and for restoring it upon release, and then
1415          * set it to the master's address
1416          */
1417         ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1418
1419         if (!bond->params.fail_over_mac ||
1420             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1421                 /* Set slave to master's mac address.  The application already
1422                  * set the master's mac address to that of the first slave
1423                  */
1424                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1425                 addr.sa_family = slave_dev->type;
1426                 res = dev_set_mac_address(slave_dev, &addr);
1427                 if (res) {
1428                         netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1429                         goto err_restore_mtu;
1430                 }
1431         }
1432
1433         /* open the slave since the application closed it */
1434         res = dev_open(slave_dev);
1435         if (res) {
1436                 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1437                 goto err_restore_mac;
1438         }
1439
1440         slave_dev->priv_flags |= IFF_BONDING;
1441         /* initialize slave stats */
1442         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1443
1444         if (bond_is_lb(bond)) {
1445                 /* bond_alb_init_slave() must be called before all other stages since
1446                  * it might fail and we do not want to have to undo everything
1447                  */
1448                 res = bond_alb_init_slave(bond, new_slave);
1449                 if (res)
1450                         goto err_close;
1451         }
1452
1453         /* If the mode uses primary, then the following is handled by
1454          * bond_change_active_slave().
1455          */
1456         if (!bond_uses_primary(bond)) {
1457                 /* set promiscuity level to new slave */
1458                 if (bond_dev->flags & IFF_PROMISC) {
1459                         res = dev_set_promiscuity(slave_dev, 1);
1460                         if (res)
1461                                 goto err_close;
1462                 }
1463
1464                 /* set allmulti level to new slave */
1465                 if (bond_dev->flags & IFF_ALLMULTI) {
1466                         res = dev_set_allmulti(slave_dev, 1);
1467                         if (res)
1468                                 goto err_close;
1469                 }
1470
1471                 netif_addr_lock_bh(bond_dev);
1472
1473                 dev_mc_sync_multiple(slave_dev, bond_dev);
1474                 dev_uc_sync_multiple(slave_dev, bond_dev);
1475
1476                 netif_addr_unlock_bh(bond_dev);
1477         }
1478
1479         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1480                 /* add lacpdu mc addr to mc list */
1481                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1482
1483                 dev_mc_add(slave_dev, lacpdu_multicast);
1484         }
1485
1486         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1487         if (res) {
1488                 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1489                            slave_dev->name);
1490                 goto err_close;
1491         }
1492
1493         prev_slave = bond_last_slave(bond);
1494
1495         new_slave->delay = 0;
1496         new_slave->link_failure_count = 0;
1497
1498         bond_update_speed_duplex(new_slave);
1499
1500         new_slave->last_rx = jiffies -
1501                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1502         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1503                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1504
1505         if (bond->params.miimon && !bond->params.use_carrier) {
1506                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1507
1508                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1509                         /* miimon is set but a bonded network driver
1510                          * does not support ETHTOOL/MII and
1511                          * arp_interval is not set.  Note: if
1512                          * use_carrier is enabled, we will never go
1513                          * here (because netif_carrier is always
1514                          * supported); thus, we don't need to change
1515                          * the messages for netif_carrier.
1516                          */
1517                         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",
1518                                     slave_dev->name);
1519                 } else if (link_reporting == -1) {
1520                         /* unable get link status using mii/ethtool */
1521                         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",
1522                                     slave_dev->name);
1523                 }
1524         }
1525
1526         /* check for initial state */
1527         if (bond->params.miimon) {
1528                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1529                         if (bond->params.updelay) {
1530                                 bond_set_slave_link_state(new_slave,
1531                                                           BOND_LINK_BACK);
1532                                 new_slave->delay = bond->params.updelay;
1533                         } else {
1534                                 bond_set_slave_link_state(new_slave,
1535                                                           BOND_LINK_UP);
1536                         }
1537                 } else {
1538                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN);
1539                 }
1540         } else if (bond->params.arp_interval) {
1541                 bond_set_slave_link_state(new_slave,
1542                                           (netif_carrier_ok(slave_dev) ?
1543                                           BOND_LINK_UP : BOND_LINK_DOWN));
1544         } else {
1545                 bond_set_slave_link_state(new_slave, BOND_LINK_UP);
1546         }
1547
1548         if (new_slave->link != BOND_LINK_DOWN)
1549                 new_slave->last_link_up = jiffies;
1550         netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1551                    new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1552                    (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1553
1554         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1555                 /* if there is a primary slave, remember it */
1556                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1557                         rcu_assign_pointer(bond->primary_slave, new_slave);
1558                         bond->force_primary = true;
1559                 }
1560         }
1561
1562         switch (BOND_MODE(bond)) {
1563         case BOND_MODE_ACTIVEBACKUP:
1564                 bond_set_slave_inactive_flags(new_slave,
1565                                               BOND_SLAVE_NOTIFY_NOW);
1566                 break;
1567         case BOND_MODE_8023AD:
1568                 /* in 802.3ad mode, the internal mechanism
1569                  * will activate the slaves in the selected
1570                  * aggregator
1571                  */
1572                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1573                 /* if this is the first slave */
1574                 if (!prev_slave) {
1575                         SLAVE_AD_INFO(new_slave)->id = 1;
1576                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1577                          * can be called only after the mac address of the bond is set
1578                          */
1579                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1580                 } else {
1581                         SLAVE_AD_INFO(new_slave)->id =
1582                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1583                 }
1584
1585                 bond_3ad_bind_slave(new_slave);
1586                 break;
1587         case BOND_MODE_TLB:
1588         case BOND_MODE_ALB:
1589                 bond_set_active_slave(new_slave);
1590                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1591                 break;
1592         default:
1593                 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1594
1595                 /* always active in trunk mode */
1596                 bond_set_active_slave(new_slave);
1597
1598                 /* In trunking mode there is little meaning to curr_active_slave
1599                  * anyway (it holds no special properties of the bond device),
1600                  * so we can change it without calling change_active_interface()
1601                  */
1602                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1603                     new_slave->link == BOND_LINK_UP)
1604                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1605
1606                 break;
1607         } /* switch(bond_mode) */
1608
1609 #ifdef CONFIG_NET_POLL_CONTROLLER
1610         slave_dev->npinfo = bond->dev->npinfo;
1611         if (slave_dev->npinfo) {
1612                 if (slave_enable_netpoll(new_slave)) {
1613                         netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1614                         res = -EBUSY;
1615                         goto err_detach;
1616                 }
1617         }
1618 #endif
1619
1620         if (!(bond_dev->features & NETIF_F_LRO))
1621                 dev_disable_lro(slave_dev);
1622
1623         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1624                                          new_slave);
1625         if (res) {
1626                 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1627                 goto err_detach;
1628         }
1629
1630         res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1631         if (res) {
1632                 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1633                 goto err_unregister;
1634         }
1635
1636         res = bond_sysfs_slave_add(new_slave);
1637         if (res) {
1638                 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1639                 goto err_upper_unlink;
1640         }
1641
1642         bond->slave_cnt++;
1643         bond_compute_features(bond);
1644         bond_set_carrier(bond);
1645
1646         if (bond_uses_primary(bond)) {
1647                 block_netpoll_tx();
1648                 bond_select_active_slave(bond);
1649                 unblock_netpoll_tx();
1650         }
1651
1652         if (bond_mode_uses_xmit_hash(bond))
1653                 bond_update_slave_arr(bond, NULL);
1654
1655         netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1656                     slave_dev->name,
1657                     bond_is_active_slave(new_slave) ? "an active" : "a backup",
1658                     new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1659
1660         /* enslave is successful */
1661         bond_queue_slave_event(new_slave);
1662         return 0;
1663
1664 /* Undo stages on error */
1665 err_upper_unlink:
1666         bond_upper_dev_unlink(bond_dev, slave_dev);
1667
1668 err_unregister:
1669         netdev_rx_handler_unregister(slave_dev);
1670
1671 err_detach:
1672         if (!bond_uses_primary(bond))
1673                 bond_hw_addr_flush(bond_dev, slave_dev);
1674
1675         vlan_vids_del_by_dev(slave_dev, bond_dev);
1676         if (rcu_access_pointer(bond->primary_slave) == new_slave)
1677                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1678         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1679                 block_netpoll_tx();
1680                 bond_change_active_slave(bond, NULL);
1681                 bond_select_active_slave(bond);
1682                 unblock_netpoll_tx();
1683         }
1684         /* either primary_slave or curr_active_slave might've changed */
1685         synchronize_rcu();
1686         slave_disable_netpoll(new_slave);
1687
1688 err_close:
1689         slave_dev->priv_flags &= ~IFF_BONDING;
1690         dev_close(slave_dev);
1691
1692 err_restore_mac:
1693         if (!bond->params.fail_over_mac ||
1694             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1695                 /* XXX TODO - fom follow mode needs to change master's
1696                  * MAC if this slave's MAC is in use by the bond, or at
1697                  * least print a warning.
1698                  */
1699                 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1700                 addr.sa_family = slave_dev->type;
1701                 dev_set_mac_address(slave_dev, &addr);
1702         }
1703
1704 err_restore_mtu:
1705         dev_set_mtu(slave_dev, new_slave->original_mtu);
1706
1707 err_free:
1708         bond_free_slave(new_slave);
1709
1710 err_undo_flags:
1711         /* Enslave of first slave has failed and we need to fix master's mac */
1712         if (!bond_has_slaves(bond) &&
1713             ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
1714                 eth_hw_addr_random(bond_dev);
1715
1716         return res;
1717 }
1718
1719 /* Try to release the slave device <slave> from the bond device <master>
1720  * It is legal to access curr_active_slave without a lock because all the function
1721  * is RTNL-locked. If "all" is true it means that the function is being called
1722  * while destroying a bond interface and all slaves are being released.
1723  *
1724  * The rules for slave state should be:
1725  *   for Active/Backup:
1726  *     Active stays on all backups go down
1727  *   for Bonded connections:
1728  *     The first up interface should be left on and all others downed.
1729  */
1730 static int __bond_release_one(struct net_device *bond_dev,
1731                               struct net_device *slave_dev,
1732                               bool all)
1733 {
1734         struct bonding *bond = netdev_priv(bond_dev);
1735         struct slave *slave, *oldcurrent;
1736         struct sockaddr addr;
1737         int old_flags = bond_dev->flags;
1738         netdev_features_t old_features = bond_dev->features;
1739
1740         /* slave is not a slave or master is not master of this slave */
1741         if (!(slave_dev->flags & IFF_SLAVE) ||
1742             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1743                 netdev_dbg(bond_dev, "cannot release %s\n",
1744                            slave_dev->name);
1745                 return -EINVAL;
1746         }
1747
1748         block_netpoll_tx();
1749
1750         slave = bond_get_slave_by_dev(bond, slave_dev);
1751         if (!slave) {
1752                 /* not a slave of this bond */
1753                 netdev_info(bond_dev, "%s not enslaved\n",
1754                             slave_dev->name);
1755                 unblock_netpoll_tx();
1756                 return -EINVAL;
1757         }
1758
1759         bond_sysfs_slave_del(slave);
1760
1761         /* recompute stats just before removing the slave */
1762         bond_get_stats(bond->dev, &bond->bond_stats);
1763
1764         bond_upper_dev_unlink(bond_dev, slave_dev);
1765         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1766          * for this slave anymore.
1767          */
1768         netdev_rx_handler_unregister(slave_dev);
1769
1770         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1771                 bond_3ad_unbind_slave(slave);
1772
1773         if (bond_mode_uses_xmit_hash(bond))
1774                 bond_update_slave_arr(bond, slave);
1775
1776         netdev_info(bond_dev, "Releasing %s interface %s\n",
1777                     bond_is_active_slave(slave) ? "active" : "backup",
1778                     slave_dev->name);
1779
1780         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1781
1782         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1783
1784         if (!all && (!bond->params.fail_over_mac ||
1785                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1786                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1787                     bond_has_slaves(bond))
1788                         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",
1789                                     slave_dev->name, slave->perm_hwaddr,
1790                                     bond_dev->name, slave_dev->name);
1791         }
1792
1793         if (rtnl_dereference(bond->primary_slave) == slave)
1794                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1795
1796         if (oldcurrent == slave)
1797                 bond_change_active_slave(bond, NULL);
1798
1799         if (bond_is_lb(bond)) {
1800                 /* Must be called only after the slave has been
1801                  * detached from the list and the curr_active_slave
1802                  * has been cleared (if our_slave == old_current),
1803                  * but before a new active slave is selected.
1804                  */
1805                 bond_alb_deinit_slave(bond, slave);
1806         }
1807
1808         if (all) {
1809                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1810         } else if (oldcurrent == slave) {
1811                 /* Note that we hold RTNL over this sequence, so there
1812                  * is no concern that another slave add/remove event
1813                  * will interfere.
1814                  */
1815                 bond_select_active_slave(bond);
1816         }
1817
1818         if (!bond_has_slaves(bond)) {
1819                 bond_set_carrier(bond);
1820                 eth_hw_addr_random(bond_dev);
1821         }
1822
1823         unblock_netpoll_tx();
1824         synchronize_rcu();
1825         bond->slave_cnt--;
1826
1827         if (!bond_has_slaves(bond)) {
1828                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1829                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1830         }
1831
1832         bond_compute_features(bond);
1833         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1834             (old_features & NETIF_F_VLAN_CHALLENGED))
1835                 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1836                             slave_dev->name, bond_dev->name);
1837
1838         vlan_vids_del_by_dev(slave_dev, bond_dev);
1839
1840         /* If the mode uses primary, then this case was handled above by
1841          * bond_change_active_slave(..., NULL)
1842          */
1843         if (!bond_uses_primary(bond)) {
1844                 /* unset promiscuity level from slave
1845                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1846                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1847                  * value of that flag before that change, as that was the value
1848                  * when this slave was attached, so we cache at the start of the
1849                  * function and use it here. Same goes for ALLMULTI below
1850                  */
1851                 if (old_flags & IFF_PROMISC)
1852                         dev_set_promiscuity(slave_dev, -1);
1853
1854                 /* unset allmulti level from slave */
1855                 if (old_flags & IFF_ALLMULTI)
1856                         dev_set_allmulti(slave_dev, -1);
1857
1858                 bond_hw_addr_flush(bond_dev, slave_dev);
1859         }
1860
1861         slave_disable_netpoll(slave);
1862
1863         /* close slave before restoring its mac address */
1864         dev_close(slave_dev);
1865
1866         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1867             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1868                 /* restore original ("permanent") mac address */
1869                 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1870                 addr.sa_family = slave_dev->type;
1871                 dev_set_mac_address(slave_dev, &addr);
1872         }
1873
1874         dev_set_mtu(slave_dev, slave->original_mtu);
1875
1876         slave_dev->priv_flags &= ~IFF_BONDING;
1877
1878         bond_free_slave(slave);
1879
1880         return 0;
1881 }
1882
1883 /* A wrapper used because of ndo_del_link */
1884 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1885 {
1886         return __bond_release_one(bond_dev, slave_dev, false);
1887 }
1888
1889 /* First release a slave and then destroy the bond if no more slaves are left.
1890  * Must be under rtnl_lock when this function is called.
1891  */
1892 static int  bond_release_and_destroy(struct net_device *bond_dev,
1893                                      struct net_device *slave_dev)
1894 {
1895         struct bonding *bond = netdev_priv(bond_dev);
1896         int ret;
1897
1898         ret = bond_release(bond_dev, slave_dev);
1899         if (ret == 0 && !bond_has_slaves(bond)) {
1900                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1901                 netdev_info(bond_dev, "Destroying bond %s\n",
1902                             bond_dev->name);
1903                 unregister_netdevice(bond_dev);
1904         }
1905         return ret;
1906 }
1907
1908 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1909 {
1910         struct bonding *bond = netdev_priv(bond_dev);
1911         bond_fill_ifbond(bond, info);
1912         return 0;
1913 }
1914
1915 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1916 {
1917         struct bonding *bond = netdev_priv(bond_dev);
1918         struct list_head *iter;
1919         int i = 0, res = -ENODEV;
1920         struct slave *slave;
1921
1922         bond_for_each_slave(bond, slave, iter) {
1923                 if (i++ == (int)info->slave_id) {
1924                         res = 0;
1925                         bond_fill_ifslave(slave, info);
1926                         break;
1927                 }
1928         }
1929
1930         return res;
1931 }
1932
1933 /*-------------------------------- Monitoring -------------------------------*/
1934
1935 /* called with rcu_read_lock() */
1936 static int bond_miimon_inspect(struct bonding *bond)
1937 {
1938         int link_state, commit = 0;
1939         struct list_head *iter;
1940         struct slave *slave;
1941         bool ignore_updelay;
1942
1943         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
1944
1945         bond_for_each_slave_rcu(bond, slave, iter) {
1946                 slave->new_link = BOND_LINK_NOCHANGE;
1947
1948                 link_state = bond_check_dev_link(bond, slave->dev, 0);
1949
1950                 switch (slave->link) {
1951                 case BOND_LINK_UP:
1952                         if (link_state)
1953                                 continue;
1954
1955                         bond_set_slave_link_state(slave, BOND_LINK_FAIL);
1956                         slave->delay = bond->params.downdelay;
1957                         if (slave->delay) {
1958                                 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
1959                                             (BOND_MODE(bond) ==
1960                                              BOND_MODE_ACTIVEBACKUP) ?
1961                                              (bond_is_active_slave(slave) ?
1962                                               "active " : "backup ") : "",
1963                                             slave->dev->name,
1964                                             bond->params.downdelay * bond->params.miimon);
1965                         }
1966                         /*FALLTHRU*/
1967                 case BOND_LINK_FAIL:
1968                         if (link_state) {
1969                                 /* recovered before downdelay expired */
1970                                 bond_set_slave_link_state(slave, BOND_LINK_UP);
1971                                 slave->last_link_up = jiffies;
1972                                 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
1973                                             (bond->params.downdelay - slave->delay) *
1974                                             bond->params.miimon,
1975                                             slave->dev->name);
1976                                 continue;
1977                         }
1978
1979                         if (slave->delay <= 0) {
1980                                 slave->new_link = BOND_LINK_DOWN;
1981                                 commit++;
1982                                 continue;
1983                         }
1984
1985                         slave->delay--;
1986                         break;
1987
1988                 case BOND_LINK_DOWN:
1989                         if (!link_state)
1990                                 continue;
1991
1992                         bond_set_slave_link_state(slave, BOND_LINK_BACK);
1993                         slave->delay = bond->params.updelay;
1994
1995                         if (slave->delay) {
1996                                 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
1997                                             slave->dev->name,
1998                                             ignore_updelay ? 0 :
1999                                             bond->params.updelay *
2000                                             bond->params.miimon);
2001                         }
2002                         /*FALLTHRU*/
2003                 case BOND_LINK_BACK:
2004                         if (!link_state) {
2005                                 bond_set_slave_link_state(slave,
2006                                                           BOND_LINK_DOWN);
2007                                 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2008                                             (bond->params.updelay - slave->delay) *
2009                                             bond->params.miimon,
2010                                             slave->dev->name);
2011
2012                                 continue;
2013                         }
2014
2015                         if (ignore_updelay)
2016                                 slave->delay = 0;
2017
2018                         if (slave->delay <= 0) {
2019                                 slave->new_link = BOND_LINK_UP;
2020                                 commit++;
2021                                 ignore_updelay = false;
2022                                 continue;
2023                         }
2024
2025                         slave->delay--;
2026                         break;
2027                 }
2028         }
2029
2030         return commit;
2031 }
2032
2033 static void bond_miimon_commit(struct bonding *bond)
2034 {
2035         struct list_head *iter;
2036         struct slave *slave, *primary;
2037
2038         bond_for_each_slave(bond, slave, iter) {
2039                 switch (slave->new_link) {
2040                 case BOND_LINK_NOCHANGE:
2041                         continue;
2042
2043                 case BOND_LINK_UP:
2044                         bond_set_slave_link_state(slave, BOND_LINK_UP);
2045                         slave->last_link_up = jiffies;
2046
2047                         primary = rtnl_dereference(bond->primary_slave);
2048                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2049                                 /* prevent it from being the active one */
2050                                 bond_set_backup_slave(slave);
2051                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2052                                 /* make it immediately active */
2053                                 bond_set_active_slave(slave);
2054                         } else if (slave != primary) {
2055                                 /* prevent it from being the active one */
2056                                 bond_set_backup_slave(slave);
2057                         }
2058
2059                         netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2060                                     slave->dev->name,
2061                                     slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2062                                     slave->duplex ? "full" : "half");
2063
2064                         /* notify ad that the link status has changed */
2065                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2066                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2067
2068                         if (bond_is_lb(bond))
2069                                 bond_alb_handle_link_change(bond, slave,
2070                                                             BOND_LINK_UP);
2071
2072                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2073                                 bond_update_slave_arr(bond, NULL);
2074
2075                         if (!bond->curr_active_slave || slave == primary)
2076                                 goto do_failover;
2077
2078                         continue;
2079
2080                 case BOND_LINK_DOWN:
2081                         if (slave->link_failure_count < UINT_MAX)
2082                                 slave->link_failure_count++;
2083
2084                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2085
2086                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2087                             BOND_MODE(bond) == BOND_MODE_8023AD)
2088                                 bond_set_slave_inactive_flags(slave,
2089                                                               BOND_SLAVE_NOTIFY_NOW);
2090
2091                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2092                                     slave->dev->name);
2093
2094                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2095                                 bond_3ad_handle_link_change(slave,
2096                                                             BOND_LINK_DOWN);
2097
2098                         if (bond_is_lb(bond))
2099                                 bond_alb_handle_link_change(bond, slave,
2100                                                             BOND_LINK_DOWN);
2101
2102                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2103                                 bond_update_slave_arr(bond, NULL);
2104
2105                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2106                                 goto do_failover;
2107
2108                         continue;
2109
2110                 default:
2111                         netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2112                                    slave->new_link, slave->dev->name);
2113                         slave->new_link = BOND_LINK_NOCHANGE;
2114
2115                         continue;
2116                 }
2117
2118 do_failover:
2119                 block_netpoll_tx();
2120                 bond_select_active_slave(bond);
2121                 unblock_netpoll_tx();
2122         }
2123
2124         bond_set_carrier(bond);
2125 }
2126
2127 /* bond_mii_monitor
2128  *
2129  * Really a wrapper that splits the mii monitor into two phases: an
2130  * inspection, then (if inspection indicates something needs to be done)
2131  * an acquisition of appropriate locks followed by a commit phase to
2132  * implement whatever link state changes are indicated.
2133  */
2134 static void bond_mii_monitor(struct work_struct *work)
2135 {
2136         struct bonding *bond = container_of(work, struct bonding,
2137                                             mii_work.work);
2138         bool should_notify_peers = false;
2139         unsigned long delay;
2140
2141         delay = msecs_to_jiffies(bond->params.miimon);
2142
2143         if (!bond_has_slaves(bond))
2144                 goto re_arm;
2145
2146         rcu_read_lock();
2147
2148         should_notify_peers = bond_should_notify_peers(bond);
2149
2150         if (bond_miimon_inspect(bond)) {
2151                 rcu_read_unlock();
2152
2153                 /* Race avoidance with bond_close cancel of workqueue */
2154                 if (!rtnl_trylock()) {
2155                         delay = 1;
2156                         should_notify_peers = false;
2157                         goto re_arm;
2158                 }
2159
2160                 bond_miimon_commit(bond);
2161
2162                 rtnl_unlock();  /* might sleep, hold no other locks */
2163         } else
2164                 rcu_read_unlock();
2165
2166 re_arm:
2167         if (bond->params.miimon)
2168                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2169
2170         if (should_notify_peers) {
2171                 if (!rtnl_trylock())
2172                         return;
2173                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2174                 rtnl_unlock();
2175         }
2176 }
2177
2178 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2179 {
2180         struct net_device *upper;
2181         struct list_head *iter;
2182         bool ret = false;
2183
2184         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2185                 return true;
2186
2187         rcu_read_lock();
2188         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2189                 if (ip == bond_confirm_addr(upper, 0, ip)) {
2190                         ret = true;
2191                         break;
2192                 }
2193         }
2194         rcu_read_unlock();
2195
2196         return ret;
2197 }
2198
2199 /* We go to the (large) trouble of VLAN tagging ARP frames because
2200  * switches in VLAN mode (especially if ports are configured as
2201  * "native" to a VLAN) might not pass non-tagged frames.
2202  */
2203 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2204                           __be32 dest_ip, __be32 src_ip,
2205                           struct bond_vlan_tag *tags)
2206 {
2207         struct sk_buff *skb;
2208         struct bond_vlan_tag *outer_tag = tags;
2209
2210         netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2211                    arp_op, slave_dev->name, &dest_ip, &src_ip);
2212
2213         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2214                          NULL, slave_dev->dev_addr, NULL);
2215
2216         if (!skb) {
2217                 net_err_ratelimited("ARP packet allocation failed\n");
2218                 return;
2219         }
2220
2221         if (!tags || tags->vlan_proto == VLAN_N_VID)
2222                 goto xmit;
2223
2224         tags++;
2225
2226         /* Go through all the tags backwards and add them to the packet */
2227         while (tags->vlan_proto != VLAN_N_VID) {
2228                 if (!tags->vlan_id) {
2229                         tags++;
2230                         continue;
2231                 }
2232
2233                 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2234                            ntohs(outer_tag->vlan_proto), tags->vlan_id);
2235                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2236                                                 tags->vlan_id);
2237                 if (!skb) {
2238                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2239                         return;
2240                 }
2241
2242                 tags++;
2243         }
2244         /* Set the outer tag */
2245         if (outer_tag->vlan_id) {
2246                 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2247                            ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2248                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2249                                        outer_tag->vlan_id);
2250         }
2251
2252 xmit:
2253         arp_xmit(skb);
2254 }
2255
2256 /* Validate the device path between the @start_dev and the @end_dev.
2257  * The path is valid if the @end_dev is reachable through device
2258  * stacking.
2259  * When the path is validated, collect any vlan information in the
2260  * path.
2261  */
2262 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2263                                               struct net_device *end_dev,
2264                                               int level)
2265 {
2266         struct bond_vlan_tag *tags;
2267         struct net_device *upper;
2268         struct list_head  *iter;
2269
2270         if (start_dev == end_dev) {
2271                 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2272                 if (!tags)
2273                         return ERR_PTR(-ENOMEM);
2274                 tags[level].vlan_proto = VLAN_N_VID;
2275                 return tags;
2276         }
2277
2278         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2279                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2280                 if (IS_ERR_OR_NULL(tags)) {
2281                         if (IS_ERR(tags))
2282                                 return tags;
2283                         continue;
2284                 }
2285                 if (is_vlan_dev(upper)) {
2286                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2287                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2288                 }
2289
2290                 return tags;
2291         }
2292
2293         return NULL;
2294 }
2295
2296 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2297 {
2298         struct rtable *rt;
2299         struct bond_vlan_tag *tags;
2300         __be32 *targets = bond->params.arp_targets, addr;
2301         int i;
2302
2303         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2304                 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2305                 tags = NULL;
2306
2307                 /* Find out through which dev should the packet go */
2308                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2309                                      RTO_ONLINK, 0);
2310                 if (IS_ERR(rt)) {
2311                         /* there's no route to target - try to send arp
2312                          * probe to generate any traffic (arp_validate=0)
2313                          */
2314                         if (bond->params.arp_validate)
2315                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2316                                                      bond->dev->name,
2317                                                      &targets[i]);
2318                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2319                                       0, tags);
2320                         continue;
2321                 }
2322
2323                 /* bond device itself */
2324                 if (rt->dst.dev == bond->dev)
2325                         goto found;
2326
2327                 rcu_read_lock();
2328                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2329                 rcu_read_unlock();
2330
2331                 if (!IS_ERR_OR_NULL(tags))
2332                         goto found;
2333
2334                 /* Not our device - skip */
2335                 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2336                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2337
2338                 ip_rt_put(rt);
2339                 continue;
2340
2341 found:
2342                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2343                 ip_rt_put(rt);
2344                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2345                               addr, tags);
2346                 kfree(tags);
2347         }
2348 }
2349
2350 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2351 {
2352         int i;
2353
2354         if (!sip || !bond_has_this_ip(bond, tip)) {
2355                 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2356                            &sip, &tip);
2357                 return;
2358         }
2359
2360         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2361         if (i == -1) {
2362                 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2363                            &sip);
2364                 return;
2365         }
2366         slave->last_rx = jiffies;
2367         slave->target_last_arp_rx[i] = jiffies;
2368 }
2369
2370 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2371                  struct slave *slave)
2372 {
2373         struct arphdr *arp = (struct arphdr *)skb->data;
2374         struct slave *curr_active_slave;
2375         unsigned char *arp_ptr;
2376         __be32 sip, tip;
2377         int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2378
2379         if (!slave_do_arp_validate(bond, slave)) {
2380                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2381                     !slave_do_arp_validate_only(bond))
2382                         slave->last_rx = jiffies;
2383                 return RX_HANDLER_ANOTHER;
2384         } else if (!is_arp) {
2385                 return RX_HANDLER_ANOTHER;
2386         }
2387
2388         alen = arp_hdr_len(bond->dev);
2389
2390         netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2391                    skb->dev->name);
2392
2393         if (alen > skb_headlen(skb)) {
2394                 arp = kmalloc(alen, GFP_ATOMIC);
2395                 if (!arp)
2396                         goto out_unlock;
2397                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2398                         goto out_unlock;
2399         }
2400
2401         if (arp->ar_hln != bond->dev->addr_len ||
2402             skb->pkt_type == PACKET_OTHERHOST ||
2403             skb->pkt_type == PACKET_LOOPBACK ||
2404             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2405             arp->ar_pro != htons(ETH_P_IP) ||
2406             arp->ar_pln != 4)
2407                 goto out_unlock;
2408
2409         arp_ptr = (unsigned char *)(arp + 1);
2410         arp_ptr += bond->dev->addr_len;
2411         memcpy(&sip, arp_ptr, 4);
2412         arp_ptr += 4 + bond->dev->addr_len;
2413         memcpy(&tip, arp_ptr, 4);
2414
2415         netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2416                    slave->dev->name, bond_slave_state(slave),
2417                      bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2418                      &sip, &tip);
2419
2420         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2421
2422         /* Backup slaves won't see the ARP reply, but do come through
2423          * here for each ARP probe (so we swap the sip/tip to validate
2424          * the probe).  In a "redundant switch, common router" type of
2425          * configuration, the ARP probe will (hopefully) travel from
2426          * the active, through one switch, the router, then the other
2427          * switch before reaching the backup.
2428          *
2429          * We 'trust' the arp requests if there is an active slave and
2430          * it received valid arp reply(s) after it became active. This
2431          * is done to avoid endless looping when we can't reach the
2432          * arp_ip_target and fool ourselves with our own arp requests.
2433          */
2434
2435         if (bond_is_active_slave(slave))
2436                 bond_validate_arp(bond, slave, sip, tip);
2437         else if (curr_active_slave &&
2438                  time_after(slave_last_rx(bond, curr_active_slave),
2439                             curr_active_slave->last_link_up))
2440                 bond_validate_arp(bond, slave, tip, sip);
2441
2442 out_unlock:
2443         if (arp != (struct arphdr *)skb->data)
2444                 kfree(arp);
2445         return RX_HANDLER_ANOTHER;
2446 }
2447
2448 /* function to verify if we're in the arp_interval timeslice, returns true if
2449  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2450  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2451  */
2452 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2453                                   int mod)
2454 {
2455         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2456
2457         return time_in_range(jiffies,
2458                              last_act - delta_in_ticks,
2459                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2460 }
2461
2462 /* This function is called regularly to monitor each slave's link
2463  * ensuring that traffic is being sent and received when arp monitoring
2464  * is used in load-balancing mode. if the adapter has been dormant, then an
2465  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2466  * arp monitoring in active backup mode.
2467  */
2468 static void bond_loadbalance_arp_mon(struct work_struct *work)
2469 {
2470         struct bonding *bond = container_of(work, struct bonding,
2471                                             arp_work.work);
2472         struct slave *slave, *oldcurrent;
2473         struct list_head *iter;
2474         int do_failover = 0, slave_state_changed = 0;
2475
2476         if (!bond_has_slaves(bond))
2477                 goto re_arm;
2478
2479         rcu_read_lock();
2480
2481         oldcurrent = rcu_dereference(bond->curr_active_slave);
2482         /* see if any of the previous devices are up now (i.e. they have
2483          * xmt and rcv traffic). the curr_active_slave does not come into
2484          * the picture unless it is null. also, slave->last_link_up is not
2485          * needed here because we send an arp on each slave and give a slave
2486          * as long as it needs to get the tx/rx within the delta.
2487          * TODO: what about up/down delay in arp mode? it wasn't here before
2488          *       so it can wait
2489          */
2490         bond_for_each_slave_rcu(bond, slave, iter) {
2491                 unsigned long trans_start = dev_trans_start(slave->dev);
2492
2493                 if (slave->link != BOND_LINK_UP) {
2494                         if (bond_time_in_interval(bond, trans_start, 1) &&
2495                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2496
2497                                 slave->link  = BOND_LINK_UP;
2498                                 slave_state_changed = 1;
2499
2500                                 /* primary_slave has no meaning in round-robin
2501                                  * mode. the window of a slave being up and
2502                                  * curr_active_slave being null after enslaving
2503                                  * is closed.
2504                                  */
2505                                 if (!oldcurrent) {
2506                                         netdev_info(bond->dev, "link status definitely up for interface %s\n",
2507                                                     slave->dev->name);
2508                                         do_failover = 1;
2509                                 } else {
2510                                         netdev_info(bond->dev, "interface %s is now up\n",
2511                                                     slave->dev->name);
2512                                 }
2513                         }
2514                 } else {
2515                         /* slave->link == BOND_LINK_UP */
2516
2517                         /* not all switches will respond to an arp request
2518                          * when the source ip is 0, so don't take the link down
2519                          * if we don't know our ip yet
2520                          */
2521                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2522                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2523
2524                                 slave->link  = BOND_LINK_DOWN;
2525                                 slave_state_changed = 1;
2526
2527                                 if (slave->link_failure_count < UINT_MAX)
2528                                         slave->link_failure_count++;
2529
2530                                 netdev_info(bond->dev, "interface %s is now down\n",
2531                                             slave->dev->name);
2532
2533                                 if (slave == oldcurrent)
2534                                         do_failover = 1;
2535                         }
2536                 }
2537
2538                 /* note: if switch is in round-robin mode, all links
2539                  * must tx arp to ensure all links rx an arp - otherwise
2540                  * links may oscillate or not come up at all; if switch is
2541                  * in something like xor mode, there is nothing we can
2542                  * do - all replies will be rx'ed on same link causing slaves
2543                  * to be unstable during low/no traffic periods
2544                  */
2545                 if (bond_slave_is_up(slave))
2546                         bond_arp_send_all(bond, slave);
2547         }
2548
2549         rcu_read_unlock();
2550
2551         if (do_failover || slave_state_changed) {
2552                 if (!rtnl_trylock())
2553                         goto re_arm;
2554
2555                 if (slave_state_changed) {
2556                         bond_slave_state_change(bond);
2557                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2558                                 bond_update_slave_arr(bond, NULL);
2559                 }
2560                 if (do_failover) {
2561                         block_netpoll_tx();
2562                         bond_select_active_slave(bond);
2563                         unblock_netpoll_tx();
2564                 }
2565                 rtnl_unlock();
2566         }
2567
2568 re_arm:
2569         if (bond->params.arp_interval)
2570                 queue_delayed_work(bond->wq, &bond->arp_work,
2571                                    msecs_to_jiffies(bond->params.arp_interval));
2572 }
2573
2574 /* Called to inspect slaves for active-backup mode ARP monitor link state
2575  * changes.  Sets new_link in slaves to specify what action should take
2576  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2577  * to link states must be committed.
2578  *
2579  * Called with rcu_read_lock held.
2580  */
2581 static int bond_ab_arp_inspect(struct bonding *bond)
2582 {
2583         unsigned long trans_start, last_rx;
2584         struct list_head *iter;
2585         struct slave *slave;
2586         int commit = 0;
2587
2588         bond_for_each_slave_rcu(bond, slave, iter) {
2589                 slave->new_link = BOND_LINK_NOCHANGE;
2590                 last_rx = slave_last_rx(bond, slave);
2591
2592                 if (slave->link != BOND_LINK_UP) {
2593                         if (bond_time_in_interval(bond, last_rx, 1)) {
2594                                 slave->new_link = BOND_LINK_UP;
2595                                 commit++;
2596                         }
2597                         continue;
2598                 }
2599
2600                 /* Give slaves 2*delta after being enslaved or made
2601                  * active.  This avoids bouncing, as the last receive
2602                  * times need a full ARP monitor cycle to be updated.
2603                  */
2604                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2605                         continue;
2606
2607                 /* Backup slave is down if:
2608                  * - No current_arp_slave AND
2609                  * - more than 3*delta since last receive AND
2610                  * - the bond has an IP address
2611                  *
2612                  * Note: a non-null current_arp_slave indicates
2613                  * the curr_active_slave went down and we are
2614                  * searching for a new one; under this condition
2615                  * we only take the curr_active_slave down - this
2616                  * gives each slave a chance to tx/rx traffic
2617                  * before being taken out
2618                  */
2619                 if (!bond_is_active_slave(slave) &&
2620                     !rcu_access_pointer(bond->current_arp_slave) &&
2621                     !bond_time_in_interval(bond, last_rx, 3)) {
2622                         slave->new_link = BOND_LINK_DOWN;
2623                         commit++;
2624                 }
2625
2626                 /* Active slave is down if:
2627                  * - more than 2*delta since transmitting OR
2628                  * - (more than 2*delta since receive AND
2629                  *    the bond has an IP address)
2630                  */
2631                 trans_start = dev_trans_start(slave->dev);
2632                 if (bond_is_active_slave(slave) &&
2633                     (!bond_time_in_interval(bond, trans_start, 2) ||
2634                      !bond_time_in_interval(bond, last_rx, 2))) {
2635                         slave->new_link = BOND_LINK_DOWN;
2636                         commit++;
2637                 }
2638         }
2639
2640         return commit;
2641 }
2642
2643 /* Called to commit link state changes noted by inspection step of
2644  * active-backup mode ARP monitor.
2645  *
2646  * Called with RTNL hold.
2647  */
2648 static void bond_ab_arp_commit(struct bonding *bond)
2649 {
2650         unsigned long trans_start;
2651         struct list_head *iter;
2652         struct slave *slave;
2653
2654         bond_for_each_slave(bond, slave, iter) {
2655                 switch (slave->new_link) {
2656                 case BOND_LINK_NOCHANGE:
2657                         continue;
2658
2659                 case BOND_LINK_UP:
2660                         trans_start = dev_trans_start(slave->dev);
2661                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
2662                             (!rtnl_dereference(bond->curr_active_slave) &&
2663                              bond_time_in_interval(bond, trans_start, 1))) {
2664                                 struct slave *current_arp_slave;
2665
2666                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2667                                 bond_set_slave_link_state(slave, BOND_LINK_UP);
2668                                 if (current_arp_slave) {
2669                                         bond_set_slave_inactive_flags(
2670                                                 current_arp_slave,
2671                                                 BOND_SLAVE_NOTIFY_NOW);
2672                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2673                                 }
2674
2675                                 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2676                                             slave->dev->name);
2677
2678                                 if (!rtnl_dereference(bond->curr_active_slave) ||
2679                                     slave == rtnl_dereference(bond->primary_slave))
2680                                         goto do_failover;
2681
2682                         }
2683
2684                         continue;
2685
2686                 case BOND_LINK_DOWN:
2687                         if (slave->link_failure_count < UINT_MAX)
2688                                 slave->link_failure_count++;
2689
2690                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2691                         bond_set_slave_inactive_flags(slave,
2692                                                       BOND_SLAVE_NOTIFY_NOW);
2693
2694                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2695                                     slave->dev->name);
2696
2697                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
2698                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2699                                 goto do_failover;
2700                         }
2701
2702                         continue;
2703
2704                 default:
2705                         netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2706                                    slave->new_link, slave->dev->name);
2707                         continue;
2708                 }
2709
2710 do_failover:
2711                 block_netpoll_tx();
2712                 bond_select_active_slave(bond);
2713                 unblock_netpoll_tx();
2714         }
2715
2716         bond_set_carrier(bond);
2717 }
2718
2719 /* Send ARP probes for active-backup mode ARP monitor.
2720  *
2721  * Called with rcu_read_lock held.
2722  */
2723 static bool bond_ab_arp_probe(struct bonding *bond)
2724 {
2725         struct slave *slave, *before = NULL, *new_slave = NULL,
2726                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2727                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2728         struct list_head *iter;
2729         bool found = false;
2730         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2731
2732         if (curr_arp_slave && curr_active_slave)
2733                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2734                             curr_arp_slave->dev->name,
2735                             curr_active_slave->dev->name);
2736
2737         if (curr_active_slave) {
2738                 bond_arp_send_all(bond, curr_active_slave);
2739                 return should_notify_rtnl;
2740         }
2741
2742         /* if we don't have a curr_active_slave, search for the next available
2743          * backup slave from the current_arp_slave and make it the candidate
2744          * for becoming the curr_active_slave
2745          */
2746
2747         if (!curr_arp_slave) {
2748                 curr_arp_slave = bond_first_slave_rcu(bond);
2749                 if (!curr_arp_slave)
2750                         return should_notify_rtnl;
2751         }
2752
2753         bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2754
2755         bond_for_each_slave_rcu(bond, slave, iter) {
2756                 if (!found && !before && bond_slave_is_up(slave))
2757                         before = slave;
2758
2759                 if (found && !new_slave && bond_slave_is_up(slave))
2760                         new_slave = slave;
2761                 /* if the link state is up at this point, we
2762                  * mark it down - this can happen if we have
2763                  * simultaneous link failures and
2764                  * reselect_active_interface doesn't make this
2765                  * one the current slave so it is still marked
2766                  * up when it is actually down
2767                  */
2768                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2769                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2770                         if (slave->link_failure_count < UINT_MAX)
2771                                 slave->link_failure_count++;
2772
2773                         bond_set_slave_inactive_flags(slave,
2774                                                       BOND_SLAVE_NOTIFY_LATER);
2775
2776                         netdev_info(bond->dev, "backup interface %s is now down\n",
2777                                     slave->dev->name);
2778                 }
2779                 if (slave == curr_arp_slave)
2780                         found = true;
2781         }
2782
2783         if (!new_slave && before)
2784                 new_slave = before;
2785
2786         if (!new_slave)
2787                 goto check_state;
2788
2789         bond_set_slave_link_state(new_slave, BOND_LINK_BACK);
2790         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2791         bond_arp_send_all(bond, new_slave);
2792         new_slave->last_link_up = jiffies;
2793         rcu_assign_pointer(bond->current_arp_slave, new_slave);
2794
2795 check_state:
2796         bond_for_each_slave_rcu(bond, slave, iter) {
2797                 if (slave->should_notify) {
2798                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2799                         break;
2800                 }
2801         }
2802         return should_notify_rtnl;
2803 }
2804
2805 static void bond_activebackup_arp_mon(struct work_struct *work)
2806 {
2807         struct bonding *bond = container_of(work, struct bonding,
2808                                             arp_work.work);
2809         bool should_notify_peers = false;
2810         bool should_notify_rtnl = false;
2811         int delta_in_ticks;
2812
2813         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2814
2815         if (!bond_has_slaves(bond))
2816                 goto re_arm;
2817
2818         rcu_read_lock();
2819
2820         should_notify_peers = bond_should_notify_peers(bond);
2821
2822         if (bond_ab_arp_inspect(bond)) {
2823                 rcu_read_unlock();
2824
2825                 /* Race avoidance with bond_close flush of workqueue */
2826                 if (!rtnl_trylock()) {
2827                         delta_in_ticks = 1;
2828                         should_notify_peers = false;
2829                         goto re_arm;
2830                 }
2831
2832                 bond_ab_arp_commit(bond);
2833
2834                 rtnl_unlock();
2835                 rcu_read_lock();
2836         }
2837
2838         should_notify_rtnl = bond_ab_arp_probe(bond);
2839         rcu_read_unlock();
2840
2841 re_arm:
2842         if (bond->params.arp_interval)
2843                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2844
2845         if (should_notify_peers || should_notify_rtnl) {
2846                 if (!rtnl_trylock())
2847                         return;
2848
2849                 if (should_notify_peers)
2850                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2851                                                  bond->dev);
2852                 if (should_notify_rtnl)
2853                         bond_slave_state_notify(bond);
2854
2855                 rtnl_unlock();
2856         }
2857 }
2858
2859 /*-------------------------- netdev event handling --------------------------*/
2860
2861 /* Change device name */
2862 static int bond_event_changename(struct bonding *bond)
2863 {
2864         bond_remove_proc_entry(bond);
2865         bond_create_proc_entry(bond);
2866
2867         bond_debug_reregister(bond);
2868
2869         return NOTIFY_DONE;
2870 }
2871
2872 static int bond_master_netdev_event(unsigned long event,
2873                                     struct net_device *bond_dev)
2874 {
2875         struct bonding *event_bond = netdev_priv(bond_dev);
2876
2877         switch (event) {
2878         case NETDEV_CHANGENAME:
2879                 return bond_event_changename(event_bond);
2880         case NETDEV_UNREGISTER:
2881                 bond_remove_proc_entry(event_bond);
2882                 break;
2883         case NETDEV_REGISTER:
2884                 bond_create_proc_entry(event_bond);
2885                 break;
2886         case NETDEV_NOTIFY_PEERS:
2887                 if (event_bond->send_peer_notif)
2888                         event_bond->send_peer_notif--;
2889                 break;
2890         default:
2891                 break;
2892         }
2893
2894         return NOTIFY_DONE;
2895 }
2896
2897 static int bond_slave_netdev_event(unsigned long event,
2898                                    struct net_device *slave_dev)
2899 {
2900         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
2901         struct bonding *bond;
2902         struct net_device *bond_dev;
2903         u32 old_speed;
2904         u8 old_duplex;
2905
2906         /* A netdev event can be generated while enslaving a device
2907          * before netdev_rx_handler_register is called in which case
2908          * slave will be NULL
2909          */
2910         if (!slave)
2911                 return NOTIFY_DONE;
2912         bond_dev = slave->bond->dev;
2913         bond = slave->bond;
2914         primary = rtnl_dereference(bond->primary_slave);
2915
2916         switch (event) {
2917         case NETDEV_UNREGISTER:
2918                 if (bond_dev->type != ARPHRD_ETHER)
2919                         bond_release_and_destroy(bond_dev, slave_dev);
2920                 else
2921                         bond_release(bond_dev, slave_dev);
2922                 break;
2923         case NETDEV_UP:
2924         case NETDEV_CHANGE:
2925                 old_speed = slave->speed;
2926                 old_duplex = slave->duplex;
2927
2928                 bond_update_speed_duplex(slave);
2929
2930                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2931                         if (old_speed != slave->speed)
2932                                 bond_3ad_adapter_speed_changed(slave);
2933                         if (old_duplex != slave->duplex)
2934                                 bond_3ad_adapter_duplex_changed(slave);
2935                 }
2936                 /* Fallthrough */
2937         case NETDEV_DOWN:
2938                 /* Refresh slave-array if applicable!
2939                  * If the setup does not use miimon or arpmon (mode-specific!),
2940                  * then these events will not cause the slave-array to be
2941                  * refreshed. This will cause xmit to use a slave that is not
2942                  * usable. Avoid such situation by refeshing the array at these
2943                  * events. If these (miimon/arpmon) parameters are configured
2944                  * then array gets refreshed twice and that should be fine!
2945                  */
2946                 if (bond_mode_uses_xmit_hash(bond))
2947                         bond_update_slave_arr(bond, NULL);
2948                 break;
2949         case NETDEV_CHANGEMTU:
2950                 /* TODO: Should slaves be allowed to
2951                  * independently alter their MTU?  For
2952                  * an active-backup bond, slaves need
2953                  * not be the same type of device, so
2954                  * MTUs may vary.  For other modes,
2955                  * slaves arguably should have the
2956                  * same MTUs. To do this, we'd need to
2957                  * take over the slave's change_mtu
2958                  * function for the duration of their
2959                  * servitude.
2960                  */
2961                 break;
2962         case NETDEV_CHANGENAME:
2963                 /* we don't care if we don't have primary set */
2964                 if (!bond_uses_primary(bond) ||
2965                     !bond->params.primary[0])
2966                         break;
2967
2968                 if (slave == primary) {
2969                         /* slave's name changed - he's no longer primary */
2970                         RCU_INIT_POINTER(bond->primary_slave, NULL);
2971                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2972                         /* we have a new primary slave */
2973                         rcu_assign_pointer(bond->primary_slave, slave);
2974                 } else { /* we didn't change primary - exit */
2975                         break;
2976                 }
2977
2978                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
2979                             primary ? slave_dev->name : "none");
2980
2981                 block_netpoll_tx();
2982                 bond_select_active_slave(bond);
2983                 unblock_netpoll_tx();
2984                 break;
2985         case NETDEV_FEAT_CHANGE:
2986                 bond_compute_features(bond);
2987                 break;
2988         case NETDEV_RESEND_IGMP:
2989                 /* Propagate to master device */
2990                 call_netdevice_notifiers(event, slave->bond->dev);
2991                 break;
2992         default:
2993                 break;
2994         }
2995
2996         return NOTIFY_DONE;
2997 }
2998
2999 /* bond_netdev_event: handle netdev notifier chain events.
3000  *
3001  * This function receives events for the netdev chain.  The caller (an
3002  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3003  * locks for us to safely manipulate the slave devices (RTNL lock,
3004  * dev_probe_lock).
3005  */
3006 static int bond_netdev_event(struct notifier_block *this,
3007                              unsigned long event, void *ptr)
3008 {
3009         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3010
3011         netdev_dbg(event_dev, "event: %lx\n", event);
3012
3013         if (!(event_dev->priv_flags & IFF_BONDING))
3014                 return NOTIFY_DONE;
3015
3016         if (event_dev->flags & IFF_MASTER) {
3017                 netdev_dbg(event_dev, "IFF_MASTER\n");
3018                 return bond_master_netdev_event(event, event_dev);
3019         }
3020
3021         if (event_dev->flags & IFF_SLAVE) {
3022                 netdev_dbg(event_dev, "IFF_SLAVE\n");
3023                 return bond_slave_netdev_event(event, event_dev);
3024         }
3025
3026         return NOTIFY_DONE;
3027 }
3028
3029 static struct notifier_block bond_netdev_notifier = {
3030         .notifier_call = bond_netdev_event,
3031 };
3032
3033 /*---------------------------- Hashing Policies -----------------------------*/
3034
3035 /* L2 hash helper */
3036 static inline u32 bond_eth_hash(struct sk_buff *skb)
3037 {
3038         struct ethhdr *ep, hdr_tmp;
3039
3040         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3041         if (ep)
3042                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3043         return 0;
3044 }
3045
3046 /* Extract the appropriate headers based on bond's xmit policy */
3047 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3048                               struct flow_keys *fk)
3049 {
3050         const struct ipv6hdr *iph6;
3051         const struct iphdr *iph;
3052         int noff, proto = -1;
3053
3054         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3055                 return skb_flow_dissect(skb, fk);
3056
3057         fk->ports = 0;
3058         noff = skb_network_offset(skb);
3059         if (skb->protocol == htons(ETH_P_IP)) {
3060                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3061                         return false;
3062                 iph = ip_hdr(skb);
3063                 fk->src = iph->saddr;
3064                 fk->dst = iph->daddr;
3065                 noff += iph->ihl << 2;
3066                 if (!ip_is_fragment(iph))
3067                         proto = iph->protocol;
3068         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3069                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3070                         return false;
3071                 iph6 = ipv6_hdr(skb);
3072                 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
3073                 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
3074                 noff += sizeof(*iph6);
3075                 proto = iph6->nexthdr;
3076         } else {
3077                 return false;
3078         }
3079         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3080                 fk->ports = skb_flow_get_ports(skb, noff, proto);
3081
3082         return true;
3083 }
3084
3085 /**
3086  * bond_xmit_hash - generate a hash value based on the xmit policy
3087  * @bond: bonding device
3088  * @skb: buffer to use for headers
3089  *
3090  * This function will extract the necessary headers from the skb buffer and use
3091  * them to generate a hash based on the xmit_policy set in the bonding device
3092  */
3093 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3094 {
3095         struct flow_keys flow;
3096         u32 hash;
3097
3098         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3099             !bond_flow_dissect(bond, skb, &flow))
3100                 return bond_eth_hash(skb);
3101
3102         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3103             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3104                 hash = bond_eth_hash(skb);
3105         else
3106                 hash = (__force u32)flow.ports;
3107         hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
3108         hash ^= (hash >> 16);
3109         hash ^= (hash >> 8);
3110
3111         return hash;
3112 }
3113
3114 /*-------------------------- Device entry points ----------------------------*/
3115
3116 static void bond_work_init_all(struct bonding *bond)
3117 {
3118         INIT_DELAYED_WORK(&bond->mcast_work,
3119                           bond_resend_igmp_join_requests_delayed);
3120         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3121         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3122         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3123                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3124         else
3125                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3126         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3127         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3128 }
3129
3130 static void bond_work_cancel_all(struct bonding *bond)
3131 {
3132         cancel_delayed_work_sync(&bond->mii_work);
3133         cancel_delayed_work_sync(&bond->arp_work);
3134         cancel_delayed_work_sync(&bond->alb_work);
3135         cancel_delayed_work_sync(&bond->ad_work);
3136         cancel_delayed_work_sync(&bond->mcast_work);
3137         cancel_delayed_work_sync(&bond->slave_arr_work);
3138 }
3139
3140 static int bond_open(struct net_device *bond_dev)
3141 {
3142         struct bonding *bond = netdev_priv(bond_dev);
3143         struct list_head *iter;
3144         struct slave *slave;
3145
3146         /* reset slave->backup and slave->inactive */
3147         if (bond_has_slaves(bond)) {
3148                 bond_for_each_slave(bond, slave, iter) {
3149                         if (bond_uses_primary(bond) &&
3150                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3151                                 bond_set_slave_inactive_flags(slave,
3152                                                               BOND_SLAVE_NOTIFY_NOW);
3153                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3154                                 bond_set_slave_active_flags(slave,
3155                                                             BOND_SLAVE_NOTIFY_NOW);
3156                         }
3157                 }
3158         }
3159
3160         bond_work_init_all(bond);
3161
3162         if (bond_is_lb(bond)) {
3163                 /* bond_alb_initialize must be called before the timer
3164                  * is started.
3165                  */
3166                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3167                         return -ENOMEM;
3168                 if (bond->params.tlb_dynamic_lb)
3169                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3170         }
3171
3172         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3173                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3174
3175         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3176                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3177                 bond->recv_probe = bond_arp_rcv;
3178         }
3179
3180         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3181                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3182                 /* register to receive LACPDUs */
3183                 bond->recv_probe = bond_3ad_lacpdu_recv;
3184                 bond_3ad_initiate_agg_selection(bond, 1);
3185         }
3186
3187         if (bond_mode_uses_xmit_hash(bond))
3188                 bond_update_slave_arr(bond, NULL);
3189
3190         return 0;
3191 }
3192
3193 static int bond_close(struct net_device *bond_dev)
3194 {
3195         struct bonding *bond = netdev_priv(bond_dev);
3196
3197         bond_work_cancel_all(bond);
3198         bond->send_peer_notif = 0;
3199         if (bond_is_lb(bond))
3200                 bond_alb_deinitialize(bond);
3201         bond->recv_probe = NULL;
3202
3203         return 0;
3204 }
3205
3206 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3207                                                 struct rtnl_link_stats64 *stats)
3208 {
3209         struct bonding *bond = netdev_priv(bond_dev);
3210         struct rtnl_link_stats64 temp;
3211         struct list_head *iter;
3212         struct slave *slave;
3213
3214         memcpy(stats, &bond->bond_stats, sizeof(*stats));
3215
3216         bond_for_each_slave(bond, slave, iter) {
3217                 const struct rtnl_link_stats64 *sstats =
3218                         dev_get_stats(slave->dev, &temp);
3219                 struct rtnl_link_stats64 *pstats = &slave->slave_stats;
3220
3221                 stats->rx_packets +=  sstats->rx_packets - pstats->rx_packets;
3222                 stats->rx_bytes += sstats->rx_bytes - pstats->rx_bytes;
3223                 stats->rx_errors += sstats->rx_errors - pstats->rx_errors;
3224                 stats->rx_dropped += sstats->rx_dropped - pstats->rx_dropped;
3225
3226                 stats->tx_packets += sstats->tx_packets - pstats->tx_packets;;
3227                 stats->tx_bytes += sstats->tx_bytes - pstats->tx_bytes;
3228                 stats->tx_errors += sstats->tx_errors - pstats->tx_errors;
3229                 stats->tx_dropped += sstats->tx_dropped - pstats->tx_dropped;
3230
3231                 stats->multicast += sstats->multicast - pstats->multicast;
3232                 stats->collisions += sstats->collisions - pstats->collisions;
3233
3234                 stats->rx_length_errors += sstats->rx_length_errors - pstats->rx_length_errors;
3235                 stats->rx_over_errors += sstats->rx_over_errors - pstats->rx_over_errors;
3236                 stats->rx_crc_errors += sstats->rx_crc_errors - pstats->rx_crc_errors;
3237                 stats->rx_frame_errors += sstats->rx_frame_errors - pstats->rx_frame_errors;
3238                 stats->rx_fifo_errors += sstats->rx_fifo_errors - pstats->rx_fifo_errors;
3239                 stats->rx_missed_errors += sstats->rx_missed_errors - pstats->rx_missed_errors;
3240
3241                 stats->tx_aborted_errors += sstats->tx_aborted_errors - pstats->tx_aborted_errors;
3242                 stats->tx_carrier_errors += sstats->tx_carrier_errors - pstats->tx_carrier_errors;
3243                 stats->tx_fifo_errors += sstats->tx_fifo_errors - pstats->tx_fifo_errors;
3244                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors - pstats->tx_heartbeat_errors;
3245                 stats->tx_window_errors += sstats->tx_window_errors - pstats->tx_window_errors;
3246
3247                 /* save off the slave stats for the next run */
3248                 memcpy(pstats, sstats, sizeof(*sstats));
3249         }
3250         memcpy(&bond->bond_stats, stats, sizeof(*stats));
3251
3252         return stats;
3253 }
3254
3255 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3256 {
3257         struct bonding *bond = netdev_priv(bond_dev);
3258         struct net_device *slave_dev = NULL;
3259         struct ifbond k_binfo;
3260         struct ifbond __user *u_binfo = NULL;
3261         struct ifslave k_sinfo;
3262         struct ifslave __user *u_sinfo = NULL;
3263         struct mii_ioctl_data *mii = NULL;
3264         struct bond_opt_value newval;
3265         struct net *net;
3266         int res = 0;
3267
3268         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3269
3270         switch (cmd) {
3271         case SIOCGMIIPHY:
3272                 mii = if_mii(ifr);
3273                 if (!mii)
3274                         return -EINVAL;
3275
3276                 mii->phy_id = 0;
3277                 /* Fall Through */
3278         case SIOCGMIIREG:
3279                 /* We do this again just in case we were called by SIOCGMIIREG
3280                  * instead of SIOCGMIIPHY.
3281                  */
3282                 mii = if_mii(ifr);
3283                 if (!mii)
3284                         return -EINVAL;
3285
3286                 if (mii->reg_num == 1) {
3287                         mii->val_out = 0;
3288                         if (netif_carrier_ok(bond->dev))
3289                                 mii->val_out = BMSR_LSTATUS;
3290                 }
3291
3292                 return 0;
3293         case BOND_INFO_QUERY_OLD:
3294         case SIOCBONDINFOQUERY:
3295                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3296
3297                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3298                         return -EFAULT;
3299
3300                 res = bond_info_query(bond_dev, &k_binfo);
3301                 if (res == 0 &&
3302                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3303                         return -EFAULT;
3304
3305                 return res;
3306         case BOND_SLAVE_INFO_QUERY_OLD:
3307         case SIOCBONDSLAVEINFOQUERY:
3308                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3309
3310                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3311                         return -EFAULT;
3312
3313                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3314                 if (res == 0 &&
3315                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3316                         return -EFAULT;
3317
3318                 return res;
3319         default:
3320                 break;
3321         }
3322
3323         net = dev_net(bond_dev);
3324
3325         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3326                 return -EPERM;
3327
3328         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3329
3330         netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3331
3332         if (!slave_dev)
3333                 return -ENODEV;
3334
3335         netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3336         switch (cmd) {
3337         case BOND_ENSLAVE_OLD:
3338         case SIOCBONDENSLAVE:
3339                 res = bond_enslave(bond_dev, slave_dev);
3340                 break;
3341         case BOND_RELEASE_OLD:
3342         case SIOCBONDRELEASE:
3343                 res = bond_release(bond_dev, slave_dev);
3344                 break;
3345         case BOND_SETHWADDR_OLD:
3346         case SIOCBONDSETHWADDR:
3347                 bond_set_dev_addr(bond_dev, slave_dev);
3348                 res = 0;
3349                 break;
3350         case BOND_CHANGE_ACTIVE_OLD:
3351         case SIOCBONDCHANGEACTIVE:
3352                 bond_opt_initstr(&newval, slave_dev->name);
3353                 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3354                 break;
3355         default:
3356                 res = -EOPNOTSUPP;
3357         }
3358
3359         return res;
3360 }
3361
3362 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3363 {
3364         struct bonding *bond = netdev_priv(bond_dev);
3365
3366         if (change & IFF_PROMISC)
3367                 bond_set_promiscuity(bond,
3368                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3369
3370         if (change & IFF_ALLMULTI)
3371                 bond_set_allmulti(bond,
3372                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3373 }
3374
3375 static void bond_set_rx_mode(struct net_device *bond_dev)
3376 {
3377         struct bonding *bond = netdev_priv(bond_dev);
3378         struct list_head *iter;
3379         struct slave *slave;
3380
3381         rcu_read_lock();
3382         if (bond_uses_primary(bond)) {
3383                 slave = rcu_dereference(bond->curr_active_slave);
3384                 if (slave) {
3385                         dev_uc_sync(slave->dev, bond_dev);
3386                         dev_mc_sync(slave->dev, bond_dev);
3387                 }
3388         } else {
3389                 bond_for_each_slave_rcu(bond, slave, iter) {
3390                         dev_uc_sync_multiple(slave->dev, bond_dev);
3391                         dev_mc_sync_multiple(slave->dev, bond_dev);
3392                 }
3393         }
3394         rcu_read_unlock();
3395 }
3396
3397 static int bond_neigh_init(struct neighbour *n)
3398 {
3399         struct bonding *bond = netdev_priv(n->dev);
3400         const struct net_device_ops *slave_ops;
3401         struct neigh_parms parms;
3402         struct slave *slave;
3403         int ret;
3404
3405         slave = bond_first_slave(bond);
3406         if (!slave)
3407                 return 0;
3408         slave_ops = slave->dev->netdev_ops;
3409         if (!slave_ops->ndo_neigh_setup)
3410                 return 0;
3411
3412         parms.neigh_setup = NULL;
3413         parms.neigh_cleanup = NULL;
3414         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3415         if (ret)
3416                 return ret;
3417
3418         /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3419          * after the last slave has been detached.  Assumes that all slaves
3420          * utilize the same neigh_cleanup (true at this writing as only user
3421          * is ipoib).
3422          */
3423         n->parms->neigh_cleanup = parms.neigh_cleanup;
3424
3425         if (!parms.neigh_setup)
3426                 return 0;
3427
3428         return parms.neigh_setup(n);
3429 }
3430
3431 /* The bonding ndo_neigh_setup is called at init time beofre any
3432  * slave exists. So we must declare proxy setup function which will
3433  * be used at run time to resolve the actual slave neigh param setup.
3434  *
3435  * It's also called by master devices (such as vlans) to setup their
3436  * underlying devices. In that case - do nothing, we're already set up from
3437  * our init.
3438  */
3439 static int bond_neigh_setup(struct net_device *dev,
3440                             struct neigh_parms *parms)
3441 {
3442         /* modify only our neigh_parms */
3443         if (parms->dev == dev)
3444                 parms->neigh_setup = bond_neigh_init;
3445
3446         return 0;
3447 }
3448
3449 /* Change the MTU of all of a master's slaves to match the master */
3450 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3451 {
3452         struct bonding *bond = netdev_priv(bond_dev);
3453         struct slave *slave, *rollback_slave;
3454         struct list_head *iter;
3455         int res = 0;
3456
3457         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3458
3459         bond_for_each_slave(bond, slave, iter) {
3460                 netdev_dbg(bond_dev, "s %p c_m %p\n",
3461                            slave, slave->dev->netdev_ops->ndo_change_mtu);
3462
3463                 res = dev_set_mtu(slave->dev, new_mtu);
3464
3465                 if (res) {
3466                         /* If we failed to set the slave's mtu to the new value
3467                          * we must abort the operation even in ACTIVE_BACKUP
3468                          * mode, because if we allow the backup slaves to have
3469                          * different mtu values than the active slave we'll
3470                          * need to change their mtu when doing a failover. That
3471                          * means changing their mtu from timer context, which
3472                          * is probably not a good idea.
3473                          */
3474                         netdev_dbg(bond_dev, "err %d %s\n", res,
3475                                    slave->dev->name);
3476                         goto unwind;
3477                 }
3478         }
3479
3480         bond_dev->mtu = new_mtu;
3481
3482         return 0;
3483
3484 unwind:
3485         /* unwind from head to the slave that failed */
3486         bond_for_each_slave(bond, rollback_slave, iter) {
3487                 int tmp_res;
3488
3489                 if (rollback_slave == slave)
3490                         break;
3491
3492                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3493                 if (tmp_res) {
3494                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3495                                    tmp_res, rollback_slave->dev->name);
3496                 }
3497         }
3498
3499         return res;
3500 }
3501
3502 /* Change HW address
3503  *
3504  * Note that many devices must be down to change the HW address, and
3505  * downing the master releases all slaves.  We can make bonds full of
3506  * bonding devices to test this, however.
3507  */
3508 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3509 {
3510         struct bonding *bond = netdev_priv(bond_dev);
3511         struct slave *slave, *rollback_slave;
3512         struct sockaddr *sa = addr, tmp_sa;
3513         struct list_head *iter;
3514         int res = 0;
3515
3516         if (BOND_MODE(bond) == BOND_MODE_ALB)
3517                 return bond_alb_set_mac_address(bond_dev, addr);
3518
3519
3520         netdev_dbg(bond_dev, "bond=%p\n", bond);
3521
3522         /* If fail_over_mac is enabled, do nothing and return success.
3523          * Returning an error causes ifenslave to fail.
3524          */
3525         if (bond->params.fail_over_mac &&
3526             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3527                 return 0;
3528
3529         if (!is_valid_ether_addr(sa->sa_data))
3530                 return -EADDRNOTAVAIL;
3531
3532         bond_for_each_slave(bond, slave, iter) {
3533                 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3534                 res = dev_set_mac_address(slave->dev, addr);
3535                 if (res) {
3536                         /* TODO: consider downing the slave
3537                          * and retry ?
3538                          * User should expect communications
3539                          * breakage anyway until ARP finish
3540                          * updating, so...
3541                          */
3542                         netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3543                         goto unwind;
3544                 }
3545         }
3546
3547         /* success */
3548         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3549         return 0;
3550
3551 unwind:
3552         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3553         tmp_sa.sa_family = bond_dev->type;
3554
3555         /* unwind from head to the slave that failed */
3556         bond_for_each_slave(bond, rollback_slave, iter) {
3557                 int tmp_res;
3558
3559                 if (rollback_slave == slave)
3560                         break;
3561
3562                 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3563                 if (tmp_res) {
3564                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3565                                    tmp_res, rollback_slave->dev->name);
3566                 }
3567         }
3568
3569         return res;
3570 }
3571
3572 /**
3573  * bond_xmit_slave_id - transmit skb through slave with slave_id
3574  * @bond: bonding device that is transmitting
3575  * @skb: buffer to transmit
3576  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3577  *
3578  * This function tries to transmit through slave with slave_id but in case
3579  * it fails, it tries to find the first available slave for transmission.
3580  * The skb is consumed in all cases, thus the function is void.
3581  */
3582 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3583 {
3584         struct list_head *iter;
3585         struct slave *slave;
3586         int i = slave_id;
3587
3588         /* Here we start from the slave with slave_id */
3589         bond_for_each_slave_rcu(bond, slave, iter) {
3590                 if (--i < 0) {
3591                         if (bond_slave_can_tx(slave)) {
3592                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3593                                 return;
3594                         }
3595                 }
3596         }
3597
3598         /* Here we start from the first slave up to slave_id */
3599         i = slave_id;
3600         bond_for_each_slave_rcu(bond, slave, iter) {
3601                 if (--i < 0)
3602                         break;
3603                 if (bond_slave_can_tx(slave)) {
3604                         bond_dev_queue_xmit(bond, skb, slave->dev);
3605                         return;
3606                 }
3607         }
3608         /* no slave that can tx has been found */
3609         bond_tx_drop(bond->dev, skb);
3610 }
3611
3612 /**
3613  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3614  * @bond: bonding device to use
3615  *
3616  * Based on the value of the bonding device's packets_per_slave parameter
3617  * this function generates a slave id, which is usually used as the next
3618  * slave to transmit through.
3619  */
3620 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3621 {
3622         u32 slave_id;
3623         struct reciprocal_value reciprocal_packets_per_slave;
3624         int packets_per_slave = bond->params.packets_per_slave;
3625
3626         switch (packets_per_slave) {
3627         case 0:
3628                 slave_id = prandom_u32();
3629                 break;
3630         case 1:
3631                 slave_id = bond->rr_tx_counter;
3632                 break;
3633         default:
3634                 reciprocal_packets_per_slave =
3635                         bond->params.reciprocal_packets_per_slave;
3636                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3637                                              reciprocal_packets_per_slave);
3638                 break;
3639         }
3640         bond->rr_tx_counter++;
3641
3642         return slave_id;
3643 }
3644
3645 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3646 {
3647         struct bonding *bond = netdev_priv(bond_dev);
3648         struct iphdr *iph = ip_hdr(skb);
3649         struct slave *slave;
3650         u32 slave_id;
3651
3652         /* Start with the curr_active_slave that joined the bond as the
3653          * default for sending IGMP traffic.  For failover purposes one
3654          * needs to maintain some consistency for the interface that will
3655          * send the join/membership reports.  The curr_active_slave found
3656          * will send all of this type of traffic.
3657          */
3658         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3659                 slave = rcu_dereference(bond->curr_active_slave);
3660                 if (slave)
3661                         bond_dev_queue_xmit(bond, skb, slave->dev);
3662                 else
3663                         bond_xmit_slave_id(bond, skb, 0);
3664         } else {
3665                 int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3666
3667                 if (likely(slave_cnt)) {
3668                         slave_id = bond_rr_gen_slave_id(bond);
3669                         bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3670                 } else {
3671                         bond_tx_drop(bond_dev, skb);
3672                 }
3673         }
3674
3675         return NETDEV_TX_OK;
3676 }
3677
3678 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3679  * the bond has a usable interface.
3680  */
3681 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3682 {
3683         struct bonding *bond = netdev_priv(bond_dev);
3684         struct slave *slave;
3685
3686         slave = rcu_dereference(bond->curr_active_slave);
3687         if (slave)
3688                 bond_dev_queue_xmit(bond, skb, slave->dev);
3689         else
3690                 bond_tx_drop(bond_dev, skb);
3691
3692         return NETDEV_TX_OK;
3693 }
3694
3695 /* Use this to update slave_array when (a) it's not appropriate to update
3696  * slave_array right away (note that update_slave_array() may sleep)
3697  * and / or (b) RTNL is not held.
3698  */
3699 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3700 {
3701         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3702 }
3703
3704 /* Slave array work handler. Holds only RTNL */
3705 static void bond_slave_arr_handler(struct work_struct *work)
3706 {
3707         struct bonding *bond = container_of(work, struct bonding,
3708                                             slave_arr_work.work);
3709         int ret;
3710
3711         if (!rtnl_trylock())
3712                 goto err;
3713
3714         ret = bond_update_slave_arr(bond, NULL);
3715         rtnl_unlock();
3716         if (ret) {
3717                 pr_warn_ratelimited("Failed to update slave array from WT\n");
3718                 goto err;
3719         }
3720         return;
3721
3722 err:
3723         bond_slave_arr_work_rearm(bond, 1);
3724 }
3725
3726 /* Build the usable slaves array in control path for modes that use xmit-hash
3727  * to determine the slave interface -
3728  * (a) BOND_MODE_8023AD
3729  * (b) BOND_MODE_XOR
3730  * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3731  *
3732  * The caller is expected to hold RTNL only and NO other lock!
3733  */
3734 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3735 {
3736         struct slave *slave;
3737         struct list_head *iter;
3738         struct bond_up_slave *new_arr, *old_arr;
3739         int slaves_in_agg;
3740         int agg_id = 0;
3741         int ret = 0;
3742
3743 #ifdef CONFIG_LOCKDEP
3744         WARN_ON(lockdep_is_held(&bond->mode_lock));
3745 #endif
3746
3747         new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3748                           GFP_KERNEL);
3749         if (!new_arr) {
3750                 ret = -ENOMEM;
3751                 pr_err("Failed to build slave-array.\n");
3752                 goto out;
3753         }
3754         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3755                 struct ad_info ad_info;
3756
3757                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3758                         pr_debug("bond_3ad_get_active_agg_info failed\n");
3759                         kfree_rcu(new_arr, rcu);
3760                         /* No active aggragator means it's not safe to use
3761                          * the previous array.
3762                          */
3763                         old_arr = rtnl_dereference(bond->slave_arr);
3764                         if (old_arr) {
3765                                 RCU_INIT_POINTER(bond->slave_arr, NULL);
3766                                 kfree_rcu(old_arr, rcu);
3767                         }
3768                         goto out;
3769                 }
3770                 slaves_in_agg = ad_info.ports;
3771                 agg_id = ad_info.aggregator_id;
3772         }
3773         bond_for_each_slave(bond, slave, iter) {
3774                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3775                         struct aggregator *agg;
3776
3777                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
3778                         if (!agg || agg->aggregator_identifier != agg_id)
3779                                 continue;
3780                 }
3781                 if (!bond_slave_can_tx(slave))
3782                         continue;
3783                 if (skipslave == slave)
3784                         continue;
3785                 new_arr->arr[new_arr->count++] = slave;
3786         }
3787
3788         old_arr = rtnl_dereference(bond->slave_arr);
3789         rcu_assign_pointer(bond->slave_arr, new_arr);
3790         if (old_arr)
3791                 kfree_rcu(old_arr, rcu);
3792 out:
3793         if (ret != 0 && skipslave) {
3794                 int idx;
3795
3796                 /* Rare situation where caller has asked to skip a specific
3797                  * slave but allocation failed (most likely!). BTW this is
3798                  * only possible when the call is initiated from
3799                  * __bond_release_one(). In this situation; overwrite the
3800                  * skipslave entry in the array with the last entry from the
3801                  * array to avoid a situation where the xmit path may choose
3802                  * this to-be-skipped slave to send a packet out.
3803                  */
3804                 old_arr = rtnl_dereference(bond->slave_arr);
3805                 for (idx = 0; idx < old_arr->count; idx++) {
3806                         if (skipslave == old_arr->arr[idx]) {
3807                                 old_arr->arr[idx] =
3808                                     old_arr->arr[old_arr->count-1];
3809                                 old_arr->count--;
3810                                 break;
3811                         }
3812                 }
3813         }
3814         return ret;
3815 }
3816
3817 /* Use this Xmit function for 3AD as well as XOR modes. The current
3818  * usable slave array is formed in the control path. The xmit function
3819  * just calculates hash and sends the packet out.
3820  */
3821 static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3822 {
3823         struct bonding *bond = netdev_priv(dev);
3824         struct slave *slave;
3825         struct bond_up_slave *slaves;
3826         unsigned int count;
3827
3828         slaves = rcu_dereference(bond->slave_arr);
3829         count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3830         if (likely(count)) {
3831                 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3832                 bond_dev_queue_xmit(bond, skb, slave->dev);
3833         } else {
3834                 bond_tx_drop(dev, skb);
3835         }
3836
3837         return NETDEV_TX_OK;
3838 }
3839
3840 /* in broadcast mode, we send everything to all usable interfaces. */
3841 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3842 {
3843         struct bonding *bond = netdev_priv(bond_dev);
3844         struct slave *slave = NULL;
3845         struct list_head *iter;
3846
3847         bond_for_each_slave_rcu(bond, slave, iter) {
3848                 if (bond_is_last_slave(bond, slave))
3849                         break;
3850                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3851                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3852
3853                         if (!skb2) {
3854                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3855                                                     bond_dev->name, __func__);
3856                                 continue;
3857                         }
3858                         bond_dev_queue_xmit(bond, skb2, slave->dev);
3859                 }
3860         }
3861         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3862                 bond_dev_queue_xmit(bond, skb, slave->dev);
3863         else
3864                 bond_tx_drop(bond_dev, skb);
3865
3866         return NETDEV_TX_OK;
3867 }
3868
3869 /*------------------------- Device initialization ---------------------------*/
3870
3871 /* Lookup the slave that corresponds to a qid */
3872 static inline int bond_slave_override(struct bonding *bond,
3873                                       struct sk_buff *skb)
3874 {
3875         struct slave *slave = NULL;
3876         struct list_head *iter;
3877
3878         if (!skb->queue_mapping)
3879                 return 1;
3880
3881         /* Find out if any slaves have the same mapping as this skb. */
3882         bond_for_each_slave_rcu(bond, slave, iter) {
3883                 if (slave->queue_id == skb->queue_mapping) {
3884                         if (bond_slave_is_up(slave) &&
3885                             slave->link == BOND_LINK_UP) {
3886                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3887                                 return 0;
3888                         }
3889                         /* If the slave isn't UP, use default transmit policy. */
3890                         break;
3891                 }
3892         }
3893
3894         return 1;
3895 }
3896
3897
3898 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3899                              void *accel_priv, select_queue_fallback_t fallback)
3900 {
3901         /* This helper function exists to help dev_pick_tx get the correct
3902          * destination queue.  Using a helper function skips a call to
3903          * skb_tx_hash and will put the skbs in the queue we expect on their
3904          * way down to the bonding driver.
3905          */
3906         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3907
3908         /* Save the original txq to restore before passing to the driver */
3909         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3910
3911         if (unlikely(txq >= dev->real_num_tx_queues)) {
3912                 do {
3913                         txq -= dev->real_num_tx_queues;
3914                 } while (txq >= dev->real_num_tx_queues);
3915         }
3916         return txq;
3917 }
3918
3919 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3920 {
3921         struct bonding *bond = netdev_priv(dev);
3922
3923         if (bond_should_override_tx_queue(bond) &&
3924             !bond_slave_override(bond, skb))
3925                 return NETDEV_TX_OK;
3926
3927         switch (BOND_MODE(bond)) {
3928         case BOND_MODE_ROUNDROBIN:
3929                 return bond_xmit_roundrobin(skb, dev);
3930         case BOND_MODE_ACTIVEBACKUP:
3931                 return bond_xmit_activebackup(skb, dev);
3932         case BOND_MODE_8023AD:
3933         case BOND_MODE_XOR:
3934                 return bond_3ad_xor_xmit(skb, dev);
3935         case BOND_MODE_BROADCAST:
3936                 return bond_xmit_broadcast(skb, dev);
3937         case BOND_MODE_ALB:
3938                 return bond_alb_xmit(skb, dev);
3939         case BOND_MODE_TLB:
3940                 return bond_tlb_xmit(skb, dev);
3941         default:
3942                 /* Should never happen, mode already checked */
3943                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
3944                 WARN_ON_ONCE(1);
3945                 bond_tx_drop(dev, skb);
3946                 return NETDEV_TX_OK;
3947         }
3948 }
3949
3950 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3951 {
3952         struct bonding *bond = netdev_priv(dev);
3953         netdev_tx_t ret = NETDEV_TX_OK;
3954
3955         /* If we risk deadlock from transmitting this in the
3956          * netpoll path, tell netpoll to queue the frame for later tx
3957          */
3958         if (unlikely(is_netpoll_tx_blocked(dev)))
3959                 return NETDEV_TX_BUSY;
3960
3961         rcu_read_lock();
3962         if (bond_has_slaves(bond))
3963                 ret = __bond_start_xmit(skb, dev);
3964         else
3965                 bond_tx_drop(dev, skb);
3966         rcu_read_unlock();
3967
3968         return ret;
3969 }
3970
3971 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3972                                      struct ethtool_cmd *ecmd)
3973 {
3974         struct bonding *bond = netdev_priv(bond_dev);
3975         unsigned long speed = 0;
3976         struct list_head *iter;
3977         struct slave *slave;
3978
3979         ecmd->duplex = DUPLEX_UNKNOWN;
3980         ecmd->port = PORT_OTHER;
3981
3982         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
3983          * do not need to check mode.  Though link speed might not represent
3984          * the true receive or transmit bandwidth (not all modes are symmetric)
3985          * this is an accurate maximum.
3986          */
3987         bond_for_each_slave(bond, slave, iter) {
3988                 if (bond_slave_can_tx(slave)) {
3989                         if (slave->speed != SPEED_UNKNOWN)
3990                                 speed += slave->speed;
3991                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
3992                             slave->duplex != DUPLEX_UNKNOWN)
3993                                 ecmd->duplex = slave->duplex;
3994                 }
3995         }
3996         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3997
3998         return 0;
3999 }
4000
4001 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4002                                      struct ethtool_drvinfo *drvinfo)
4003 {
4004         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4005         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4006         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4007                  BOND_ABI_VERSION);
4008 }
4009
4010 static const struct ethtool_ops bond_ethtool_ops = {
4011         .get_drvinfo            = bond_ethtool_get_drvinfo,
4012         .get_settings           = bond_ethtool_get_settings,
4013         .get_link               = ethtool_op_get_link,
4014 };
4015
4016 static const struct net_device_ops bond_netdev_ops = {
4017         .ndo_init               = bond_init,
4018         .ndo_uninit             = bond_uninit,
4019         .ndo_open               = bond_open,
4020         .ndo_stop               = bond_close,
4021         .ndo_start_xmit         = bond_start_xmit,
4022         .ndo_select_queue       = bond_select_queue,
4023         .ndo_get_stats64        = bond_get_stats,
4024         .ndo_do_ioctl           = bond_do_ioctl,
4025         .ndo_change_rx_flags    = bond_change_rx_flags,
4026         .ndo_set_rx_mode        = bond_set_rx_mode,
4027         .ndo_change_mtu         = bond_change_mtu,
4028         .ndo_set_mac_address    = bond_set_mac_address,
4029         .ndo_neigh_setup        = bond_neigh_setup,
4030         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4031         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4032 #ifdef CONFIG_NET_POLL_CONTROLLER
4033         .ndo_netpoll_setup      = bond_netpoll_setup,
4034         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4035         .ndo_poll_controller    = bond_poll_controller,
4036 #endif
4037         .ndo_add_slave          = bond_enslave,
4038         .ndo_del_slave          = bond_release,
4039         .ndo_fix_features       = bond_fix_features,
4040         .ndo_bridge_setlink     = ndo_dflt_netdev_switch_port_bridge_setlink,
4041         .ndo_bridge_dellink     = ndo_dflt_netdev_switch_port_bridge_dellink,
4042         .ndo_features_check     = passthru_features_check,
4043 };
4044
4045 static const struct device_type bond_type = {
4046         .name = "bond",
4047 };
4048
4049 static void bond_destructor(struct net_device *bond_dev)
4050 {
4051         struct bonding *bond = netdev_priv(bond_dev);
4052         if (bond->wq)
4053                 destroy_workqueue(bond->wq);
4054         free_netdev(bond_dev);
4055 }
4056
4057 void bond_setup(struct net_device *bond_dev)
4058 {
4059         struct bonding *bond = netdev_priv(bond_dev);
4060
4061         spin_lock_init(&bond->mode_lock);
4062         bond->params = bonding_defaults;
4063
4064         /* Initialize pointers */
4065         bond->dev = bond_dev;
4066
4067         /* Initialize the device entry points */
4068         ether_setup(bond_dev);
4069         bond_dev->netdev_ops = &bond_netdev_ops;
4070         bond_dev->ethtool_ops = &bond_ethtool_ops;
4071
4072         bond_dev->destructor = bond_destructor;
4073
4074         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4075
4076         /* Initialize the device options */
4077         bond_dev->tx_queue_len = 0;
4078         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4079         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT;
4080         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4081
4082         /* don't acquire bond device's netif_tx_lock when transmitting */
4083         bond_dev->features |= NETIF_F_LLTX;
4084
4085         /* By default, we declare the bond to be fully
4086          * VLAN hardware accelerated capable. Special
4087          * care is taken in the various xmit functions
4088          * when there are slaves that are not hw accel
4089          * capable
4090          */
4091
4092         /* Don't allow bond devices to change network namespaces. */
4093         bond_dev->features |= NETIF_F_NETNS_LOCAL;
4094
4095         bond_dev->hw_features = BOND_VLAN_FEATURES |
4096                                 NETIF_F_HW_VLAN_CTAG_TX |
4097                                 NETIF_F_HW_VLAN_CTAG_RX |
4098                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4099
4100         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4101         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4102         bond_dev->features |= bond_dev->hw_features;
4103 }
4104
4105 /* Destroy a bonding device.
4106  * Must be under rtnl_lock when this function is called.
4107  */
4108 static void bond_uninit(struct net_device *bond_dev)
4109 {
4110         struct bonding *bond = netdev_priv(bond_dev);
4111         struct list_head *iter;
4112         struct slave *slave;
4113         struct bond_up_slave *arr;
4114
4115         bond_netpoll_cleanup(bond_dev);
4116
4117         /* Release the bonded slaves */
4118         bond_for_each_slave(bond, slave, iter)
4119                 __bond_release_one(bond_dev, slave->dev, true);
4120         netdev_info(bond_dev, "Released all slaves\n");
4121
4122         arr = rtnl_dereference(bond->slave_arr);
4123         if (arr) {
4124                 RCU_INIT_POINTER(bond->slave_arr, NULL);
4125                 kfree_rcu(arr, rcu);
4126         }
4127
4128         list_del(&bond->bond_list);
4129
4130         bond_debug_unregister(bond);
4131 }
4132
4133 /*------------------------- Module initialization ---------------------------*/
4134
4135 static int bond_check_params(struct bond_params *params)
4136 {
4137         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4138         struct bond_opt_value newval;
4139         const struct bond_opt_value *valptr;
4140         int arp_all_targets_value;
4141
4142         /* Convert string parameters. */
4143         if (mode) {
4144                 bond_opt_initstr(&newval, mode);
4145                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4146                 if (!valptr) {
4147                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4148                         return -EINVAL;
4149                 }
4150                 bond_mode = valptr->value;
4151         }
4152
4153         if (xmit_hash_policy) {
4154                 if ((bond_mode != BOND_MODE_XOR) &&
4155                     (bond_mode != BOND_MODE_8023AD) &&
4156                     (bond_mode != BOND_MODE_TLB)) {
4157                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4158                                 bond_mode_name(bond_mode));
4159                 } else {
4160                         bond_opt_initstr(&newval, xmit_hash_policy);
4161                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4162                                                 &newval);
4163                         if (!valptr) {
4164                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4165                                        xmit_hash_policy);
4166                                 return -EINVAL;
4167                         }
4168                         xmit_hashtype = valptr->value;
4169                 }
4170         }
4171
4172         if (lacp_rate) {
4173                 if (bond_mode != BOND_MODE_8023AD) {
4174                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4175                                 bond_mode_name(bond_mode));
4176                 } else {
4177                         bond_opt_initstr(&newval, lacp_rate);
4178                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4179                                                 &newval);
4180                         if (!valptr) {
4181                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4182                                        lacp_rate);
4183                                 return -EINVAL;
4184                         }
4185                         lacp_fast = valptr->value;
4186                 }
4187         }
4188
4189         if (ad_select) {
4190                 bond_opt_initstr(&newval, ad_select);
4191                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4192                                         &newval);
4193                 if (!valptr) {
4194                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4195                         return -EINVAL;
4196                 }
4197                 params->ad_select = valptr->value;
4198                 if (bond_mode != BOND_MODE_8023AD)
4199                         pr_warn("ad_select param only affects 802.3ad mode\n");
4200         } else {
4201                 params->ad_select = BOND_AD_STABLE;
4202         }
4203
4204         if (max_bonds < 0) {
4205                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4206                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4207                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4208         }
4209
4210         if (miimon < 0) {
4211                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4212                         miimon, INT_MAX);
4213                 miimon = 0;
4214         }
4215
4216         if (updelay < 0) {
4217                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4218                         updelay, INT_MAX);
4219                 updelay = 0;
4220         }
4221
4222         if (downdelay < 0) {
4223                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4224                         downdelay, INT_MAX);
4225                 downdelay = 0;
4226         }
4227
4228         if ((use_carrier != 0) && (use_carrier != 1)) {
4229                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4230                         use_carrier);
4231                 use_carrier = 1;
4232         }
4233
4234         if (num_peer_notif < 0 || num_peer_notif > 255) {
4235                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4236                         num_peer_notif);
4237                 num_peer_notif = 1;
4238         }
4239
4240         /* reset values for 802.3ad/TLB/ALB */
4241         if (!bond_mode_uses_arp(bond_mode)) {
4242                 if (!miimon) {
4243                         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");
4244                         pr_warn("Forcing miimon to 100msec\n");
4245                         miimon = BOND_DEFAULT_MIIMON;
4246                 }
4247         }
4248
4249         if (tx_queues < 1 || tx_queues > 255) {
4250                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4251                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4252                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4253         }
4254
4255         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4256                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4257                         all_slaves_active);
4258                 all_slaves_active = 0;
4259         }
4260
4261         if (resend_igmp < 0 || resend_igmp > 255) {
4262                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4263                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4264                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4265         }
4266
4267         bond_opt_initval(&newval, packets_per_slave);
4268         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4269                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4270                         packets_per_slave, USHRT_MAX);
4271                 packets_per_slave = 1;
4272         }
4273
4274         if (bond_mode == BOND_MODE_ALB) {
4275                 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",
4276                           updelay);
4277         }
4278
4279         if (!miimon) {
4280                 if (updelay || downdelay) {
4281                         /* just warn the user the up/down delay will have
4282                          * no effect since miimon is zero...
4283                          */
4284                         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",
4285                                 updelay, downdelay);
4286                 }
4287         } else {
4288                 /* don't allow arp monitoring */
4289                 if (arp_interval) {
4290                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4291                                 miimon, arp_interval);
4292                         arp_interval = 0;
4293                 }
4294
4295                 if ((updelay % miimon) != 0) {
4296                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4297                                 updelay, miimon, (updelay / miimon) * miimon);
4298                 }
4299
4300                 updelay /= miimon;
4301
4302                 if ((downdelay % miimon) != 0) {
4303                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4304                                 downdelay, miimon,
4305                                 (downdelay / miimon) * miimon);
4306                 }
4307
4308                 downdelay /= miimon;
4309         }
4310
4311         if (arp_interval < 0) {
4312                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4313                         arp_interval, INT_MAX);
4314                 arp_interval = 0;
4315         }
4316
4317         for (arp_ip_count = 0, i = 0;
4318              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4319                 __be32 ip;
4320
4321                 /* not a complete check, but good enough to catch mistakes */
4322                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4323                     !bond_is_ip_target_ok(ip)) {
4324                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4325                                 arp_ip_target[i]);
4326                         arp_interval = 0;
4327                 } else {
4328                         if (bond_get_targets_ip(arp_target, ip) == -1)
4329                                 arp_target[arp_ip_count++] = ip;
4330                         else
4331                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4332                                         &ip);
4333                 }
4334         }
4335
4336         if (arp_interval && !arp_ip_count) {
4337                 /* don't allow arping if no arp_ip_target given... */
4338                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4339                         arp_interval);
4340                 arp_interval = 0;
4341         }
4342
4343         if (arp_validate) {
4344                 if (!arp_interval) {
4345                         pr_err("arp_validate requires arp_interval\n");
4346                         return -EINVAL;
4347                 }
4348
4349                 bond_opt_initstr(&newval, arp_validate);
4350                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4351                                         &newval);
4352                 if (!valptr) {
4353                         pr_err("Error: invalid arp_validate \"%s\"\n",
4354                                arp_validate);
4355                         return -EINVAL;
4356                 }
4357                 arp_validate_value = valptr->value;
4358         } else {
4359                 arp_validate_value = 0;
4360         }
4361
4362         arp_all_targets_value = 0;
4363         if (arp_all_targets) {
4364                 bond_opt_initstr(&newval, arp_all_targets);
4365                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4366                                         &newval);
4367                 if (!valptr) {
4368                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4369                                arp_all_targets);
4370                         arp_all_targets_value = 0;
4371                 } else {
4372                         arp_all_targets_value = valptr->value;
4373                 }
4374         }
4375
4376         if (miimon) {
4377                 pr_info("MII link monitoring set to %d ms\n", miimon);
4378         } else if (arp_interval) {
4379                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4380                                           arp_validate_value);
4381                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4382                         arp_interval, valptr->string, arp_ip_count);
4383
4384                 for (i = 0; i < arp_ip_count; i++)
4385                         pr_cont(" %s", arp_ip_target[i]);
4386
4387                 pr_cont("\n");
4388
4389         } else if (max_bonds) {
4390                 /* miimon and arp_interval not set, we need one so things
4391                  * work as expected, see bonding.txt for details
4392                  */
4393                 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");
4394         }
4395
4396         if (primary && !bond_mode_uses_primary(bond_mode)) {
4397                 /* currently, using a primary only makes sense
4398                  * in active backup, TLB or ALB modes
4399                  */
4400                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4401                         primary, bond_mode_name(bond_mode));
4402                 primary = NULL;
4403         }
4404
4405         if (primary && primary_reselect) {
4406                 bond_opt_initstr(&newval, primary_reselect);
4407                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4408                                         &newval);
4409                 if (!valptr) {
4410                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4411                                primary_reselect);
4412                         return -EINVAL;
4413                 }
4414                 primary_reselect_value = valptr->value;
4415         } else {
4416                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4417         }
4418
4419         if (fail_over_mac) {
4420                 bond_opt_initstr(&newval, fail_over_mac);
4421                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4422                                         &newval);
4423                 if (!valptr) {
4424                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4425                                fail_over_mac);
4426                         return -EINVAL;
4427                 }
4428                 fail_over_mac_value = valptr->value;
4429                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4430                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4431         } else {
4432                 fail_over_mac_value = BOND_FOM_NONE;
4433         }
4434
4435         if (lp_interval == 0) {
4436                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4437                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4438                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4439         }
4440
4441         /* fill params struct with the proper values */
4442         params->mode = bond_mode;
4443         params->xmit_policy = xmit_hashtype;
4444         params->miimon = miimon;
4445         params->num_peer_notif = num_peer_notif;
4446         params->arp_interval = arp_interval;
4447         params->arp_validate = arp_validate_value;
4448         params->arp_all_targets = arp_all_targets_value;
4449         params->updelay = updelay;
4450         params->downdelay = downdelay;
4451         params->use_carrier = use_carrier;
4452         params->lacp_fast = lacp_fast;
4453         params->primary[0] = 0;
4454         params->primary_reselect = primary_reselect_value;
4455         params->fail_over_mac = fail_over_mac_value;
4456         params->tx_queues = tx_queues;
4457         params->all_slaves_active = all_slaves_active;
4458         params->resend_igmp = resend_igmp;
4459         params->min_links = min_links;
4460         params->lp_interval = lp_interval;
4461         params->packets_per_slave = packets_per_slave;
4462         params->tlb_dynamic_lb = 1; /* Default value */
4463         if (packets_per_slave > 0) {
4464                 params->reciprocal_packets_per_slave =
4465                         reciprocal_value(packets_per_slave);
4466         } else {
4467                 /* reciprocal_packets_per_slave is unused if
4468                  * packets_per_slave is 0 or 1, just initialize it
4469                  */
4470                 params->reciprocal_packets_per_slave =
4471                         (struct reciprocal_value) { 0 };
4472         }
4473
4474         if (primary) {
4475                 strncpy(params->primary, primary, IFNAMSIZ);
4476                 params->primary[IFNAMSIZ - 1] = 0;
4477         }
4478
4479         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4480
4481         return 0;
4482 }
4483
4484 static struct lock_class_key bonding_netdev_xmit_lock_key;
4485 static struct lock_class_key bonding_netdev_addr_lock_key;
4486 static struct lock_class_key bonding_tx_busylock_key;
4487
4488 static void bond_set_lockdep_class_one(struct net_device *dev,
4489                                        struct netdev_queue *txq,
4490                                        void *_unused)
4491 {
4492         lockdep_set_class(&txq->_xmit_lock,
4493                           &bonding_netdev_xmit_lock_key);
4494 }
4495
4496 static void bond_set_lockdep_class(struct net_device *dev)
4497 {
4498         lockdep_set_class(&dev->addr_list_lock,
4499                           &bonding_netdev_addr_lock_key);
4500         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4501         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4502 }
4503
4504 /* Called from registration process */
4505 static int bond_init(struct net_device *bond_dev)
4506 {
4507         struct bonding *bond = netdev_priv(bond_dev);
4508         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4509
4510         netdev_dbg(bond_dev, "Begin bond_init\n");
4511
4512         bond->wq = create_singlethread_workqueue(bond_dev->name);
4513         if (!bond->wq)
4514                 return -ENOMEM;
4515
4516         bond_set_lockdep_class(bond_dev);
4517
4518         list_add_tail(&bond->bond_list, &bn->dev_list);
4519
4520         bond_prepare_sysfs_group(bond);
4521
4522         bond_debug_register(bond);
4523
4524         /* Ensure valid dev_addr */
4525         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4526             bond_dev->addr_assign_type == NET_ADDR_PERM)
4527                 eth_hw_addr_random(bond_dev);
4528
4529         return 0;
4530 }
4531
4532 unsigned int bond_get_num_tx_queues(void)
4533 {
4534         return tx_queues;
4535 }
4536
4537 /* Create a new bond based on the specified name and bonding parameters.
4538  * If name is NULL, obtain a suitable "bond%d" name for us.
4539  * Caller must NOT hold rtnl_lock; we need to release it here before we
4540  * set up our sysfs entries.
4541  */
4542 int bond_create(struct net *net, const char *name)
4543 {
4544         struct net_device *bond_dev;
4545         int res;
4546
4547         rtnl_lock();
4548
4549         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4550                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
4551                                    bond_setup, tx_queues);
4552         if (!bond_dev) {
4553                 pr_err("%s: eek! can't alloc netdev!\n", name);
4554                 rtnl_unlock();
4555                 return -ENOMEM;
4556         }
4557
4558         dev_net_set(bond_dev, net);
4559         bond_dev->rtnl_link_ops = &bond_link_ops;
4560
4561         res = register_netdevice(bond_dev);
4562
4563         netif_carrier_off(bond_dev);
4564
4565         rtnl_unlock();
4566         if (res < 0)
4567                 bond_destructor(bond_dev);
4568         return res;
4569 }
4570
4571 static int __net_init bond_net_init(struct net *net)
4572 {
4573         struct bond_net *bn = net_generic(net, bond_net_id);
4574
4575         bn->net = net;
4576         INIT_LIST_HEAD(&bn->dev_list);
4577
4578         bond_create_proc_dir(bn);
4579         bond_create_sysfs(bn);
4580
4581         return 0;
4582 }
4583
4584 static void __net_exit bond_net_exit(struct net *net)
4585 {
4586         struct bond_net *bn = net_generic(net, bond_net_id);
4587         struct bonding *bond, *tmp_bond;
4588         LIST_HEAD(list);
4589
4590         bond_destroy_sysfs(bn);
4591
4592         /* Kill off any bonds created after unregistering bond rtnl ops */
4593         rtnl_lock();
4594         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4595                 unregister_netdevice_queue(bond->dev, &list);
4596         unregister_netdevice_many(&list);
4597         rtnl_unlock();
4598
4599         bond_destroy_proc_dir(bn);
4600 }
4601
4602 static struct pernet_operations bond_net_ops = {
4603         .init = bond_net_init,
4604         .exit = bond_net_exit,
4605         .id   = &bond_net_id,
4606         .size = sizeof(struct bond_net),
4607 };
4608
4609 static int __init bonding_init(void)
4610 {
4611         int i;
4612         int res;
4613
4614         pr_info("%s", bond_version);
4615
4616         res = bond_check_params(&bonding_defaults);
4617         if (res)
4618                 goto out;
4619
4620         res = register_pernet_subsys(&bond_net_ops);
4621         if (res)
4622                 goto out;
4623
4624         res = bond_netlink_init();
4625         if (res)
4626                 goto err_link;
4627
4628         bond_create_debugfs();
4629
4630         for (i = 0; i < max_bonds; i++) {
4631                 res = bond_create(&init_net, NULL);
4632                 if (res)
4633                         goto err;
4634         }
4635
4636         register_netdevice_notifier(&bond_netdev_notifier);
4637 out:
4638         return res;
4639 err:
4640         bond_destroy_debugfs();
4641         bond_netlink_fini();
4642 err_link:
4643         unregister_pernet_subsys(&bond_net_ops);
4644         goto out;
4645
4646 }
4647
4648 static void __exit bonding_exit(void)
4649 {
4650         unregister_netdevice_notifier(&bond_netdev_notifier);
4651
4652         bond_destroy_debugfs();
4653
4654         bond_netlink_fini();
4655         unregister_pernet_subsys(&bond_net_ops);
4656
4657 #ifdef CONFIG_NET_POLL_CONTROLLER
4658         /* Make sure we don't have an imbalance on our netpoll blocking */
4659         WARN_ON(atomic_read(&netpoll_block_tx));
4660 #endif
4661 }
4662
4663 module_init(bonding_init);
4664 module_exit(bonding_exit);
4665 MODULE_LICENSE("GPL");
4666 MODULE_VERSION(DRV_VERSION);
4667 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4668 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");