kernel/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 #include "bonding_priv.h"
  86
  87 /*---------------------------- Module parameters ----------------------------*/
  88
  89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  90
  91 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  92 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  93 static int num_peer_notif = 1;
  94 static int miimon;
  95 static int updelay;
  96 static int downdelay;
  97 static int use_carrier  = 1;
  98 static char *mode;
  99 static char *primary;
 100 static char *primary_reselect;
 101 static char *lacp_rate;
 102 static int min_links;
 103 static char *ad_select;
 104 static char *xmit_hash_policy;
 105 static int arp_interval;
 106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 107 static char *arp_validate;
 108 static char *arp_all_targets;
 109 static char *fail_over_mac;
 110 static int all_slaves_active;
 111 static struct bond_params bonding_defaults;
 112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 113 static int packets_per_slave = 1;
 114 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
 115
 116 module_param(max_bonds, int, 0);
 117 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 118 module_param(tx_queues, int, 0);
 119 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 120 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 121 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 122                                "failover event (alias of num_unsol_na)");
 123 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 124 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 125                                "failover event (alias of num_grat_arp)");
 126 module_param(miimon, int, 0);
 127 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 128 module_param(updelay, int, 0);
 129 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 130 module_param(downdelay, int, 0);
 131 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 132                             "in milliseconds");
 133 module_param(use_carrier, int, 0);
 134 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 135                               "0 for off, 1 for on (default)");
 136 module_param(mode, charp, 0);
 137 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 138                        "1 for active-backup, 2 for balance-xor, "
 139                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 140                        "6 for balance-alb");
 141 module_param(primary, charp, 0);
 142 MODULE_PARM_DESC(primary, "Primary network device to use");
 143 module_param(primary_reselect, charp, 0);
 144 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 145                                    "once it comes up; "
 146                                    "0 for always (default), "
 147                                    "1 for only if speed of primary is "
 148                                    "better, "
 149                                    "2 for only on active slave "
 150                                    "failure");
 151 module_param(lacp_rate, charp, 0);
 152 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 153                             "0 for slow, 1 for fast");
 154 module_param(ad_select, charp, 0);
 155 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 156                             "0 for stable (default), 1 for bandwidth, "
 157                             "2 for count");
 158 module_param(min_links, int, 0);
 159 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 160
 161 module_param(xmit_hash_policy, charp, 0);
 162 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 163                                    "0 for layer 2 (default), 1 for layer 3+4, "
 164                                    "2 for layer 2+3, 3 for encap layer 2+3, "
 165                                    "4 for encap layer 3+4");
 166 module_param(arp_interval, int, 0);
 167 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 168 module_param_array(arp_ip_target, charp, NULL, 0);
 169 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 170 module_param(arp_validate, charp, 0);
 171 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 172                                "0 for none (default), 1 for active, "
 173                                "2 for backup, 3 for all");
 174 module_param(arp_all_targets, charp, 0);
 175 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 176 module_param(fail_over_mac, charp, 0);
 177 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 178                                 "the same MAC; 0 for none (default), "
 179                                 "1 for active, 2 for follow");
 180 module_param(all_slaves_active, int, 0);
 181 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
 182                                      "by setting active flag for all slaves; "
 183                                      "0 for never (default), 1 for always.");
 184 module_param(resend_igmp, int, 0);
 185 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 186                               "link failure");
 187 module_param(packets_per_slave, int, 0);
 188 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
 189                                     "mode; 0 for a random slave, 1 packet per "
 190                                     "slave (default), >1 packets per slave.");
 191 module_param(lp_interval, uint, 0);
 192 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
 193                               "the bonding driver sends learning packets to "
 194                               "each slaves peer switch. The default is 1.");
 195
 196 /*----------------------------- Global variables ----------------------------*/
 197
 198 #ifdef CONFIG_NET_POLL_CONTROLLER
 199 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 200 #endif
 201
 202 int bond_net_id __read_mostly;
 203
 204 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 205 static int arp_ip_count;
 206 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 207 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 208 static int lacp_fast;
 209
 210 /*-------------------------- Forward declarations ---------------------------*/
 211
 212 static int bond_init(struct net_device *bond_dev);
 213 static void bond_uninit(struct net_device *bond_dev);
 214 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
 215                                                 struct rtnl_link_stats64 *stats);
 216 static void bond_slave_arr_handler(struct work_struct *work);
 217
 218 /*---------------------------- General routines -----------------------------*/
 219
 220 const char *bond_mode_name(int mode)
 221 {
 222         static const char *names[] = {
 223                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 224                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 225                 [BOND_MODE_XOR] = "load balancing (xor)",
 226                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 227                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 228                 [BOND_MODE_TLB] = "transmit load balancing",
 229                 [BOND_MODE_ALB] = "adaptive load balancing",
 230         };
 231
 232         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 233                 return "unknown";
 234
 235         return names[mode];
 236 }
 237
 238 /*---------------------------------- VLAN -----------------------------------*/
 239
 240 /**
 241  * bond_dev_queue_xmit - Prepare skb for xmit.
 242  *
 243  * @bond: bond device that got this skb for tx.
 244  * @skb: hw accel VLAN tagged skb to transmit
 245  * @slave_dev: slave that is supposed to xmit this skbuff
 246  */
 247 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 248                         struct net_device *slave_dev)
 249 {
 250         skb->dev = slave_dev;
 251
 252         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 253                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 254         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 255
 256         if (unlikely(netpoll_tx_running(bond->dev)))
 257                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 258         else
 259                 dev_queue_xmit(skb);
 260 }
 261
 262 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 263  * We don't protect the slave list iteration with a lock because:
 264  * a. This operation is performed in IOCTL context,
 265  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 266  * c. Holding a lock with BH disabled while directly calling a base driver
 267  *    entry point is generally a BAD idea.
 268  *
 269  * The design of synchronization/protection for this operation in the 8021q
 270  * module is good for one or more VLAN devices over a single physical device
 271  * and cannot be extended for a teaming solution like bonding, so there is a
 272  * potential race condition here where a net device from the vlan group might
 273  * be referenced (either by a base driver or the 8021q code) while it is being
 274  * removed from the system. However, it turns out we're not making matters
 275  * worse, and if it works for regular VLAN usage it will work here too.
 276 */
 277
 278 /**
 279  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 280  * @bond_dev: bonding net device that got called
 281  * @vid: vlan id being added
 282  */
 283 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 284                                 __be16 proto, u16 vid)
 285 {
 286         struct bonding *bond = netdev_priv(bond_dev);
 287         struct slave *slave, *rollback_slave;
 288         struct list_head *iter;
 289         int res;
 290
 291         bond_for_each_slave(bond, slave, iter) {
 292                 res = vlan_vid_add(slave->dev, proto, vid);
 293                 if (res)
 294                         goto unwind;
 295         }
 296
 297         return 0;
 298
 299 unwind:
 300         /* unwind to the slave that failed */
 301         bond_for_each_slave(bond, rollback_slave, iter) {
 302                 if (rollback_slave == slave)
 303                         break;
 304
 305                 vlan_vid_del(rollback_slave->dev, proto, vid);
 306         }
 307
 308         return res;
 309 }
 310
 311 /**
 312  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 313  * @bond_dev: bonding net device that got called
 314  * @vid: vlan id being removed
 315  */
 316 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 317                                  __be16 proto, u16 vid)
 318 {
 319         struct bonding *bond = netdev_priv(bond_dev);
 320         struct list_head *iter;
 321         struct slave *slave;
 322
 323         bond_for_each_slave(bond, slave, iter)
 324                 vlan_vid_del(slave->dev, proto, vid);
 325
 326         if (bond_is_lb(bond))
 327                 bond_alb_clear_vlan(bond, vid);
 328
 329         return 0;
 330 }
 331
 332 /*------------------------------- Link status -------------------------------*/
 333
 334 /* Set the carrier state for the master according to the state of its
 335  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 336  * do special 802.3ad magic.
 337  *
 338  * Returns zero if carrier state does not change, nonzero if it does.
 339  */
 340 int bond_set_carrier(struct bonding *bond)
 341 {
 342         struct list_head *iter;
 343         struct slave *slave;
 344
 345         if (!bond_has_slaves(bond))
 346                 goto down;
 347
 348         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 349                 return bond_3ad_set_carrier(bond);
 350
 351         bond_for_each_slave(bond, slave, iter) {
 352                 if (slave->link == BOND_LINK_UP) {
 353                         if (!netif_carrier_ok(bond->dev)) {
 354                                 netif_carrier_on(bond->dev);
 355                                 return 1;
 356                         }
 357                         return 0;
 358                 }
 359         }
 360
 361 down:
 362         if (netif_carrier_ok(bond->dev)) {
 363                 netif_carrier_off(bond->dev);
 364                 return 1;
 365         }
 366         return 0;
 367 }
 368
 369 /* Get link speed and duplex from the slave's base driver
 370  * using ethtool. If for some reason the call fails or the
 371  * values are invalid, set speed and duplex to -1,
 372  * and return.
 373  */
 374 static void bond_update_speed_duplex(struct slave *slave)
 375 {
 376         struct net_device *slave_dev = slave->dev;
 377         struct ethtool_cmd ecmd;
 378         u32 slave_speed;
 379         int res;
 380
 381         slave->speed = SPEED_UNKNOWN;
 382         slave->duplex = DUPLEX_UNKNOWN;
 383
 384         res = __ethtool_get_settings(slave_dev, &ecmd);
 385         if (res < 0)
 386                 return;
 387
 388         slave_speed = ethtool_cmd_speed(&ecmd);
 389         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 390                 return;
 391
 392         switch (ecmd.duplex) {
 393         case DUPLEX_FULL:
 394         case DUPLEX_HALF:
 395                 break;
 396         default:
 397                 return;
 398         }
 399
 400         slave->speed = slave_speed;
 401         slave->duplex = ecmd.duplex;
 402
 403         return;
 404 }
 405
 406 const char *bond_slave_link_status(s8 link)
 407 {
 408         switch (link) {
 409         case BOND_LINK_UP:
 410                 return "up";
 411         case BOND_LINK_FAIL:
 412                 return "going down";
 413         case BOND_LINK_DOWN:
 414                 return "down";
 415         case BOND_LINK_BACK:
 416                 return "going back";
 417         default:
 418                 return "unknown";
 419         }
 420 }
 421
 422 /* if <dev> supports MII link status reporting, check its link status.
 423  *
 424  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 425  * depending upon the setting of the use_carrier parameter.
 426  *
 427  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 428  * can't tell and just pretend it is), or 0, meaning that the link is
 429  * down.
 430  *
 431  * If reporting is non-zero, instead of faking link up, return -1 if
 432  * both ETHTOOL and MII ioctls fail (meaning the device does not
 433  * support them).  If use_carrier is set, return whatever it says.
 434  * It'd be nice if there was a good way to tell if a driver supports
 435  * netif_carrier, but there really isn't.
 436  */
 437 static int bond_check_dev_link(struct bonding *bond,
 438                                struct net_device *slave_dev, int reporting)
 439 {
 440         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 441         int (*ioctl)(struct net_device *, struct ifreq *, int);
 442         struct ifreq ifr;
 443         struct mii_ioctl_data *mii;
 444
 445         if (!reporting && !netif_running(slave_dev))
 446                 return 0;
 447
 448         if (bond->params.use_carrier)
 449                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 450
 451         /* Try to get link status using Ethtool first. */
 452         if (slave_dev->ethtool_ops->get_link)
 453                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 454                         BMSR_LSTATUS : 0;
 455
 456         /* Ethtool can't be used, fallback to MII ioctls. */
 457         ioctl = slave_ops->ndo_do_ioctl;
 458         if (ioctl) {
 459                 /* TODO: set pointer to correct ioctl on a per team member
 460                  *       bases to make this more efficient. that is, once
 461                  *       we determine the correct ioctl, we will always
 462                  *       call it and not the others for that team
 463                  *       member.
 464                  */
 465
 466                 /* We cannot assume that SIOCGMIIPHY will also read a
 467                  * register; not all network drivers (e.g., e100)
 468                  * support that.
 469                  */
 470
 471                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 472                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 473                 mii = if_mii(&ifr);
 474                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 475                         mii->reg_num = MII_BMSR;
 476                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 477                                 return mii->val_out & BMSR_LSTATUS;
 478                 }
 479         }
 480
 481         /* If reporting, report that either there's no dev->do_ioctl,
 482          * or both SIOCGMIIREG and get_link failed (meaning that we
 483          * cannot report link status).  If not reporting, pretend
 484          * we're ok.
 485          */
 486         return reporting ? -1 : BMSR_LSTATUS;
 487 }
 488
 489 /*----------------------------- Multicast list ------------------------------*/
 490
 491 /* Push the promiscuity flag down to appropriate slaves */
 492 static int bond_set_promiscuity(struct bonding *bond, int inc)
 493 {
 494         struct list_head *iter;
 495         int err = 0;
 496
 497         if (bond_uses_primary(bond)) {
 498                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 499
 500                 if (curr_active)
 501                         err = dev_set_promiscuity(curr_active->dev, inc);
 502         } else {
 503                 struct slave *slave;
 504
 505                 bond_for_each_slave(bond, slave, iter) {
 506                         err = dev_set_promiscuity(slave->dev, inc);
 507                         if (err)
 508                                 return err;
 509                 }
 510         }
 511         return err;
 512 }
 513
 514 /* Push the allmulti flag down to all slaves */
 515 static int bond_set_allmulti(struct bonding *bond, int inc)
 516 {
 517         struct list_head *iter;
 518         int err = 0;
 519
 520         if (bond_uses_primary(bond)) {
 521                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 522
 523                 if (curr_active)
 524                         err = dev_set_allmulti(curr_active->dev, inc);
 525         } else {
 526                 struct slave *slave;
 527
 528                 bond_for_each_slave(bond, slave, iter) {
 529                         err = dev_set_allmulti(slave->dev, inc);
 530                         if (err)
 531                                 return err;
 532                 }
 533         }
 534         return err;
 535 }
 536
 537 /* Retrieve the list of registered multicast addresses for the bonding
 538  * device and retransmit an IGMP JOIN request to the current active
 539  * slave.
 540  */
 541 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 542 {
 543         struct bonding *bond = container_of(work, struct bonding,
 544                                             mcast_work.work);
 545
 546         if (!rtnl_trylock()) {
 547                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 548                 return;
 549         }
 550         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 551
 552         if (bond->igmp_retrans > 1) {
 553                 bond->igmp_retrans--;
 554                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 555         }
 556         rtnl_unlock();
 557 }
 558
 559 /* Flush bond's hardware addresses from slave */
 560 static void bond_hw_addr_flush(struct net_device *bond_dev,
 561                                struct net_device *slave_dev)
 562 {
 563         struct bonding *bond = netdev_priv(bond_dev);
 564
 565         dev_uc_unsync(slave_dev, bond_dev);
 566         dev_mc_unsync(slave_dev, bond_dev);
 567
 568         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 569                 /* del lacpdu mc addr from mc list */
 570                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 571
 572                 dev_mc_del(slave_dev, lacpdu_multicast);
 573         }
 574 }
 575
 576 /*--------------------------- Active slave change ---------------------------*/
 577
 578 /* Update the hardware address list and promisc/allmulti for the new and
 579  * old active slaves (if any).  Modes that are not using primary keep all
 580  * slaves up date at all times; only the modes that use primary need to call
 581  * this function to swap these settings during a failover.
 582  */
 583 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 584                               struct slave *old_active)
 585 {
 586         if (old_active) {
 587                 if (bond->dev->flags & IFF_PROMISC)
 588                         dev_set_promiscuity(old_active->dev, -1);
 589
 590                 if (bond->dev->flags & IFF_ALLMULTI)
 591                         dev_set_allmulti(old_active->dev, -1);
 592
 593                 bond_hw_addr_flush(bond->dev, old_active->dev);
 594         }
 595
 596         if (new_active) {
 597                 /* FIXME: Signal errors upstream. */
 598                 if (bond->dev->flags & IFF_PROMISC)
 599                         dev_set_promiscuity(new_active->dev, 1);
 600
 601                 if (bond->dev->flags & IFF_ALLMULTI)
 602                         dev_set_allmulti(new_active->dev, 1);
 603
 604                 netif_addr_lock_bh(bond->dev);
 605                 dev_uc_sync(new_active->dev, bond->dev);
 606                 dev_mc_sync(new_active->dev, bond->dev);
 607                 netif_addr_unlock_bh(bond->dev);
 608         }
 609 }
 610
 611 /**
 612  * bond_set_dev_addr - clone slave's address to bond
 613  * @bond_dev: bond net device
 614  * @slave_dev: slave net device
 615  *
 616  * Should be called with RTNL held.
 617  */
 618 static void bond_set_dev_addr(struct net_device *bond_dev,
 619                               struct net_device *slave_dev)
 620 {
 621         netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 622                    bond_dev, slave_dev, slave_dev->addr_len);
 623         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 624         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 625         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 626 }
 627
 628 static struct slave *bond_get_old_active(struct bonding *bond,
 629                                          struct slave *new_active)
 630 {
 631         struct slave *slave;
 632         struct list_head *iter;
 633
 634         bond_for_each_slave(bond, slave, iter) {
 635                 if (slave == new_active)
 636                         continue;
 637
 638                 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
 639                         return slave;
 640         }
 641
 642         return NULL;
 643 }
 644
 645 /* bond_do_fail_over_mac
 646  *
 647  * Perform special MAC address swapping for fail_over_mac settings
 648  *
 649  * Called with RTNL
 650  */
 651 static void bond_do_fail_over_mac(struct bonding *bond,
 652                                   struct slave *new_active,
 653                                   struct slave *old_active)
 654 {
 655         u8 tmp_mac[ETH_ALEN];
 656         struct sockaddr saddr;
 657         int rv;
 658
 659         switch (bond->params.fail_over_mac) {
 660         case BOND_FOM_ACTIVE:
 661                 if (new_active)
 662                         bond_set_dev_addr(bond->dev, new_active->dev);
 663                 break;
 664         case BOND_FOM_FOLLOW:
 665                 /* if new_active && old_active, swap them
 666                  * if just old_active, do nothing (going to no active slave)
 667                  * if just new_active, set new_active to bond's MAC
 668                  */
 669                 if (!new_active)
 670                         return;
 671
 672                 if (!old_active)
 673                         old_active = bond_get_old_active(bond, new_active);
 674
 675                 if (old_active) {
 676                         ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
 677                         ether_addr_copy(saddr.sa_data,
 678                                         old_active->dev->dev_addr);
 679                         saddr.sa_family = new_active->dev->type;
 680                 } else {
 681                         ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
 682                         saddr.sa_family = bond->dev->type;
 683                 }
 684
 685                 rv = dev_set_mac_address(new_active->dev, &saddr);
 686                 if (rv) {
 687                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 688                                    -rv, new_active->dev->name);
 689                         goto out;
 690                 }
 691
 692                 if (!old_active)
 693                         goto out;
 694
 695                 ether_addr_copy(saddr.sa_data, tmp_mac);
 696                 saddr.sa_family = old_active->dev->type;
 697
 698                 rv = dev_set_mac_address(old_active->dev, &saddr);
 699                 if (rv)
 700                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 701                                    -rv, new_active->dev->name);
 702 out:
 703                 break;
 704         default:
 705                 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
 706                            bond->params.fail_over_mac);
 707                 break;
 708         }
 709
 710 }
 711
 712 static bool bond_should_change_active(struct bonding *bond)
 713 {
 714         struct slave *prim = rtnl_dereference(bond->primary_slave);
 715         struct slave *curr = rtnl_dereference(bond->curr_active_slave);
 716
 717         if (!prim || !curr || curr->link != BOND_LINK_UP)
 718                 return true;
 719         if (bond->force_primary) {
 720                 bond->force_primary = false;
 721                 return true;
 722         }
 723         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 724             (prim->speed < curr->speed ||
 725              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 726                 return false;
 727         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 728                 return false;
 729         return true;
 730 }
 731
 732 /**
 733  * find_best_interface - select the best available slave to be the active one
 734  * @bond: our bonding struct
 735  */
 736 static struct slave *bond_find_best_slave(struct bonding *bond)
 737 {
 738         struct slave *slave, *bestslave = NULL, *primary;
 739         struct list_head *iter;
 740         int mintime = bond->params.updelay;
 741
 742         primary = rtnl_dereference(bond->primary_slave);
 743         if (primary && primary->link == BOND_LINK_UP &&
 744             bond_should_change_active(bond))
 745                 return primary;
 746
 747         bond_for_each_slave(bond, slave, iter) {
 748                 if (slave->link == BOND_LINK_UP)
 749                         return slave;
 750                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
 751                     slave->delay < mintime) {
 752                         mintime = slave->delay;
 753                         bestslave = slave;
 754                 }
 755         }
 756
 757         return bestslave;
 758 }
 759
 760 static bool bond_should_notify_peers(struct bonding *bond)
 761 {
 762         struct slave *slave;
 763
 764         rcu_read_lock();
 765         slave = rcu_dereference(bond->curr_active_slave);
 766         rcu_read_unlock();
 767
 768         netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
 769                    slave ? slave->dev->name : "NULL");
 770
 771         if (!slave || !bond->send_peer_notif ||
 772             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 773                 return false;
 774
 775         return true;
 776 }
 777
 778 /**
 779  * change_active_interface - change the active slave into the specified one
 780  * @bond: our bonding struct
 781  * @new: the new slave to make the active one
 782  *
 783  * Set the new slave to the bond's settings and unset them on the old
 784  * curr_active_slave.
 785  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 786  *
 787  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 788  * because it is apparently the best available slave we have, even though its
 789  * updelay hasn't timed out yet.
 790  *
 791  * Caller must hold RTNL.
 792  */
 793 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 794 {
 795         struct slave *old_active;
 796
 797         ASSERT_RTNL();
 798
 799         old_active = rtnl_dereference(bond->curr_active_slave);
 800
 801         if (old_active == new_active)
 802                 return;
 803
 804         if (new_active) {
 805                 new_active->last_link_up = jiffies;
 806
 807                 if (new_active->link == BOND_LINK_BACK) {
 808                         if (bond_uses_primary(bond)) {
 809                                 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
 810                                             new_active->dev->name,
 811                                             (bond->params.updelay - new_active->delay) * bond->params.miimon);
 812                         }
 813
 814                         new_active->delay = 0;
 815                         bond_set_slave_link_state(new_active, BOND_LINK_UP);
 816
 817                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 818                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 819
 820                         if (bond_is_lb(bond))
 821                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 822                 } else {
 823                         if (bond_uses_primary(bond)) {
 824                                 netdev_info(bond->dev, "making interface %s the new active one\n",
 825                                             new_active->dev->name);
 826                         }
 827                 }
 828         }
 829
 830         if (bond_uses_primary(bond))
 831                 bond_hw_addr_swap(bond, new_active, old_active);
 832
 833         if (bond_is_lb(bond)) {
 834                 bond_alb_handle_active_change(bond, new_active);
 835                 if (old_active)
 836                         bond_set_slave_inactive_flags(old_active,
 837                                                       BOND_SLAVE_NOTIFY_NOW);
 838                 if (new_active)
 839                         bond_set_slave_active_flags(new_active,
 840                                                     BOND_SLAVE_NOTIFY_NOW);
 841         } else {
 842                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 843         }
 844
 845         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
 846                 if (old_active)
 847                         bond_set_slave_inactive_flags(old_active,
 848                                                       BOND_SLAVE_NOTIFY_NOW);
 849
 850                 if (new_active) {
 851                         bool should_notify_peers = false;
 852
 853                         bond_set_slave_active_flags(new_active,
 854                                                     BOND_SLAVE_NOTIFY_NOW);
 855
 856                         if (bond->params.fail_over_mac)
 857                                 bond_do_fail_over_mac(bond, new_active,
 858                                                       old_active);
 859
 860                         if (netif_running(bond->dev)) {
 861                                 bond->send_peer_notif =
 862                                         bond->params.num_peer_notif;
 863                                 should_notify_peers =
 864                                         bond_should_notify_peers(bond);
 865                         }
 866
 867                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 868                         if (should_notify_peers)
 869                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 870                                                          bond->dev);
 871                 }
 872         }
 873
 874         /* resend IGMP joins since active slave has changed or
 875          * all were sent on curr_active_slave.
 876          * resend only if bond is brought up with the affected
 877          * bonding modes and the retransmission is enabled
 878          */
 879         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 880             ((bond_uses_primary(bond) && new_active) ||
 881              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
 882                 bond->igmp_retrans = bond->params.resend_igmp;
 883                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 884         }
 885 }
 886
 887 /**
 888  * bond_select_active_slave - select a new active slave, if needed
 889  * @bond: our bonding struct
 890  *
 891  * This functions should be called when one of the following occurs:
 892  * - The old curr_active_slave has been released or lost its link.
 893  * - The primary_slave has got its link back.
 894  * - A slave has got its link back and there's no old curr_active_slave.
 895  *
 896  * Caller must hold RTNL.
 897  */
 898 void bond_select_active_slave(struct bonding *bond)
 899 {
 900         struct slave *best_slave;
 901         int rv;
 902
 903         ASSERT_RTNL();
 904
 905         best_slave = bond_find_best_slave(bond);
 906         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
 907                 bond_change_active_slave(bond, best_slave);
 908                 rv = bond_set_carrier(bond);
 909                 if (!rv)
 910                         return;
 911
 912                 if (netif_carrier_ok(bond->dev)) {
 913                         netdev_info(bond->dev, "first active interface up!\n");
 914                 } else {
 915                         netdev_info(bond->dev, "now running without any active interface!\n");
 916                 }
 917         }
 918 }
 919
 920 #ifdef CONFIG_NET_POLL_CONTROLLER
 921 static inline int slave_enable_netpoll(struct slave *slave)
 922 {
 923         struct netpoll *np;
 924         int err = 0;
 925
 926         np = kzalloc(sizeof(*np), GFP_KERNEL);
 927         err = -ENOMEM;
 928         if (!np)
 929                 goto out;
 930
 931         err = __netpoll_setup(np, slave->dev);
 932         if (err) {
 933                 kfree(np);
 934                 goto out;
 935         }
 936         slave->np = np;
 937 out:
 938         return err;
 939 }
 940 static inline void slave_disable_netpoll(struct slave *slave)
 941 {
 942         struct netpoll *np = slave->np;
 943
 944         if (!np)
 945                 return;
 946
 947         slave->np = NULL;
 948         __netpoll_free_async(np);
 949 }
 950
 951 static void bond_poll_controller(struct net_device *bond_dev)
 952 {
 953         struct bonding *bond = netdev_priv(bond_dev);
 954         struct slave *slave = NULL;
 955         struct list_head *iter;
 956         struct ad_info ad_info;
 957         struct netpoll_info *ni;
 958         const struct net_device_ops *ops;
 959
 960         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 961                 if (bond_3ad_get_active_agg_info(bond, &ad_info))
 962                         return;
 963
 964         rcu_read_lock_bh();
 965         bond_for_each_slave_rcu(bond, slave, iter) {
 966                 ops = slave->dev->netdev_ops;
 967                 if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller)
 968                         continue;
 969
 970                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 971                         struct aggregator *agg =
 972                             SLAVE_AD_INFO(slave)->port.aggregator;
 973
 974                         if (agg &&
 975                             agg->aggregator_identifier != ad_info.aggregator_id)
 976                                 continue;
 977                 }
 978
 979                 ni = rcu_dereference_bh(slave->dev->npinfo);
 980                 if (down_trylock(&ni->dev_lock))
 981                         continue;
 982                 ops->ndo_poll_controller(slave->dev);
 983                 up(&ni->dev_lock);
 984         }
 985         rcu_read_unlock_bh();
 986 }
 987
 988 static void bond_netpoll_cleanup(struct net_device *bond_dev)
 989 {
 990         struct bonding *bond = netdev_priv(bond_dev);
 991         struct list_head *iter;
 992         struct slave *slave;
 993
 994         bond_for_each_slave(bond, slave, iter)
 995                 if (bond_slave_is_up(slave))
 996                         slave_disable_netpoll(slave);
 997 }
 998
 999 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1000 {
1001         struct bonding *bond = netdev_priv(dev);
1002         struct list_head *iter;
1003         struct slave *slave;
1004         int err = 0;
1005
1006         bond_for_each_slave(bond, slave, iter) {
1007                 err = slave_enable_netpoll(slave);
1008                 if (err) {
1009                         bond_netpoll_cleanup(dev);
1010                         break;
1011                 }
1012         }
1013         return err;
1014 }
1015 #else
1016 static inline int slave_enable_netpoll(struct slave *slave)
1017 {
1018         return 0;
1019 }
1020 static inline void slave_disable_netpoll(struct slave *slave)
1021 {
1022 }
1023 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1024 {
1025 }
1026 #endif
1027
1028 /*---------------------------------- IOCTL ----------------------------------*/
1029
1030 static netdev_features_t bond_fix_features(struct net_device *dev,
1031                                            netdev_features_t features)
1032 {
1033         struct bonding *bond = netdev_priv(dev);
1034         struct list_head *iter;
1035         netdev_features_t mask;
1036         struct slave *slave;
1037
1038         /* If any slave has the offload feature flag set,
1039          * set the offload flag on the bond.
1040          */
1041         mask = features | NETIF_F_HW_SWITCH_OFFLOAD;
1042
1043         features &= ~NETIF_F_ONE_FOR_ALL;
1044         features |= NETIF_F_ALL_FOR_ALL;
1045
1046         bond_for_each_slave(bond, slave, iter) {
1047                 features = netdev_increment_features(features,
1048                                                      slave->dev->features,
1049                                                      mask);
1050         }
1051         features = netdev_add_tso_features(features, mask);
1052
1053         return features;
1054 }
1055
1056 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1057                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1058                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1059
1060 #define BOND_ENC_FEATURES       (NETIF_F_ALL_CSUM | NETIF_F_SG | NETIF_F_RXCSUM |\
1061                                  NETIF_F_TSO)
1062
1063 static void bond_compute_features(struct bonding *bond)
1064 {
1065         unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1066                                         IFF_XMIT_DST_RELEASE_PERM;
1067         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1068         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1069         struct net_device *bond_dev = bond->dev;
1070         struct list_head *iter;
1071         struct slave *slave;
1072         unsigned short max_hard_header_len = ETH_HLEN;
1073         unsigned int gso_max_size = GSO_MAX_SIZE;
1074         u16 gso_max_segs = GSO_MAX_SEGS;
1075
1076         if (!bond_has_slaves(bond))
1077                 goto done;
1078         vlan_features &= NETIF_F_ALL_FOR_ALL;
1079
1080         bond_for_each_slave(bond, slave, iter) {
1081                 vlan_features = netdev_increment_features(vlan_features,
1082                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1083
1084                 enc_features = netdev_increment_features(enc_features,
1085                                                          slave->dev->hw_enc_features,
1086                                                          BOND_ENC_FEATURES);
1087                 dst_release_flag &= slave->dev->priv_flags;
1088                 if (slave->dev->hard_header_len > max_hard_header_len)
1089                         max_hard_header_len = slave->dev->hard_header_len;
1090
1091                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1092                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1093         }
1094
1095 done:
1096         bond_dev->vlan_features = vlan_features;
1097         bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
1098         bond_dev->hard_header_len = max_hard_header_len;
1099         bond_dev->gso_max_segs = gso_max_segs;
1100         netif_set_gso_max_size(bond_dev, gso_max_size);
1101
1102         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1103         if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1104             dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1105                 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1106
1107         netdev_change_features(bond_dev);
1108 }
1109
1110 static void bond_setup_by_slave(struct net_device *bond_dev,
1111                                 struct net_device *slave_dev)
1112 {
1113         bond_dev->header_ops        = slave_dev->header_ops;
1114
1115         bond_dev->type              = slave_dev->type;
1116         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1117         bond_dev->addr_len          = slave_dev->addr_len;
1118
1119         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1120                 slave_dev->addr_len);
1121 }
1122
1123 /* On bonding slaves other than the currently active slave, suppress
1124  * duplicates except for alb non-mcast/bcast.
1125  */
1126 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1127                                             struct slave *slave,
1128                                             struct bonding *bond)
1129 {
1130         if (bond_is_slave_inactive(slave)) {
1131                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1132                     skb->pkt_type != PACKET_BROADCAST &&
1133                     skb->pkt_type != PACKET_MULTICAST)
1134                         return false;
1135                 return true;
1136         }
1137         return false;
1138 }
1139
1140 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1141 {
1142         struct sk_buff *skb = *pskb;
1143         struct slave *slave;
1144         struct bonding *bond;
1145         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1146                           struct slave *);
1147         int ret = RX_HANDLER_ANOTHER;
1148
1149         skb = skb_share_check(skb, GFP_ATOMIC);
1150         if (unlikely(!skb))
1151                 return RX_HANDLER_CONSUMED;
1152
1153         *pskb = skb;
1154
1155         slave = bond_slave_get_rcu(skb->dev);
1156         bond = slave->bond;
1157
1158         recv_probe = ACCESS_ONCE(bond->recv_probe);
1159         if (recv_probe) {
1160                 ret = recv_probe(skb, bond, slave);
1161                 if (ret == RX_HANDLER_CONSUMED) {
1162                         consume_skb(skb);
1163                         return ret;
1164                 }
1165         }
1166
1167         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1168                 return RX_HANDLER_EXACT;
1169         }
1170
1171         skb->dev = bond->dev;
1172
1173         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1174             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1175             skb->pkt_type == PACKET_HOST) {
1176
1177                 if (unlikely(skb_cow_head(skb,
1178                                           skb->data - skb_mac_header(skb)))) {
1179                         kfree_skb(skb);
1180                         return RX_HANDLER_CONSUMED;
1181                 }
1182                 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1183         }
1184
1185         return ret;
1186 }
1187
1188 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1189                                       struct net_device *slave_dev,
1190                                       struct slave *slave)
1191 {
1192         int err;
1193
1194         err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1195         if (err)
1196                 return err;
1197         slave_dev->flags |= IFF_SLAVE;
1198         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1199         return 0;
1200 }
1201
1202 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1203                                   struct net_device *slave_dev)
1204 {
1205         netdev_upper_dev_unlink(slave_dev, bond_dev);
1206         slave_dev->flags &= ~IFF_SLAVE;
1207         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1208 }
1209
1210 static struct slave *bond_alloc_slave(struct bonding *bond)
1211 {
1212         struct slave *slave = NULL;
1213
1214         slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1215         if (!slave)
1216                 return NULL;
1217
1218         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1219                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1220                                                GFP_KERNEL);
1221                 if (!SLAVE_AD_INFO(slave)) {
1222                         kfree(slave);
1223                         return NULL;
1224                 }
1225         }
1226         return slave;
1227 }
1228
1229 static void bond_free_slave(struct slave *slave)
1230 {
1231         struct bonding *bond = bond_get_bond_by_slave(slave);
1232
1233         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1234                 kfree(SLAVE_AD_INFO(slave));
1235
1236         kfree(slave);
1237 }
1238
1239 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1240 {
1241         info->bond_mode = BOND_MODE(bond);
1242         info->miimon = bond->params.miimon;
1243         info->num_slaves = bond->slave_cnt;
1244 }
1245
1246 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1247 {
1248         strcpy(info->slave_name, slave->dev->name);
1249         info->link = slave->link;
1250         info->state = bond_slave_state(slave);
1251         info->link_failure_count = slave->link_failure_count;
1252 }
1253
1254 static void bond_netdev_notify(struct net_device *dev,
1255                                struct netdev_bonding_info *info)
1256 {
1257         rtnl_lock();
1258         netdev_bonding_info_change(dev, info);
1259         rtnl_unlock();
1260 }
1261
1262 static void bond_netdev_notify_work(struct work_struct *_work)
1263 {
1264         struct netdev_notify_work *w =
1265                 container_of(_work, struct netdev_notify_work, work.work);
1266
1267         bond_netdev_notify(w->dev, &w->bonding_info);
1268         dev_put(w->dev);
1269         kfree(w);
1270 }
1271
1272 void bond_queue_slave_event(struct slave *slave)
1273 {
1274         struct bonding *bond = slave->bond;
1275         struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1276
1277         if (!nnw)
1278                 return;
1279
1280         dev_hold(slave->dev);
1281         nnw->dev = slave->dev;
1282         bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1283         bond_fill_ifbond(bond, &nnw->bonding_info.master);
1284         INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1285
1286         queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1287 }
1288
1289 /* enslave device <slave> to bond device <master> */
1290 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1291 {
1292         struct bonding *bond = netdev_priv(bond_dev);
1293         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1294         struct slave *new_slave = NULL, *prev_slave;
1295         struct sockaddr addr;
1296         int link_reporting;
1297         int res = 0, i;
1298
1299         if (!bond->params.use_carrier &&
1300             slave_dev->ethtool_ops->get_link == NULL &&
1301             slave_ops->ndo_do_ioctl == NULL) {
1302                 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1303                             slave_dev->name);
1304         }
1305
1306         /* already enslaved */
1307         if (slave_dev->flags & IFF_SLAVE) {
1308                 netdev_dbg(bond_dev, "Error: Device was already enslaved\n");
1309                 return -EBUSY;
1310         }
1311
1312         if (bond_dev == slave_dev) {
1313                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1314                 return -EPERM;
1315         }
1316
1317         /* vlan challenged mutual exclusion */
1318         /* no need to lock since we're protected by rtnl_lock */
1319         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1320                 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1321                            slave_dev->name);
1322                 if (vlan_uses_dev(bond_dev)) {
1323                         netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1324                                    slave_dev->name, bond_dev->name);
1325                         return -EPERM;
1326                 } else {
1327                         netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1328                                     slave_dev->name, slave_dev->name,
1329                                     bond_dev->name);
1330                 }
1331         } else {
1332                 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1333                            slave_dev->name);
1334         }
1335
1336         /* Old ifenslave binaries are no longer supported.  These can
1337          * be identified with moderate accuracy by the state of the slave:
1338          * the current ifenslave will set the interface down prior to
1339          * enslaving it; the old ifenslave will not.
1340          */
1341         if ((slave_dev->flags & IFF_UP)) {
1342                 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1343                            slave_dev->name);
1344                 res = -EPERM;
1345                 goto err_undo_flags;
1346         }
1347
1348         /* set bonding device ether type by slave - bonding netdevices are
1349          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1350          * there is a need to override some of the type dependent attribs/funcs.
1351          *
1352          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1353          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1354          */
1355         if (!bond_has_slaves(bond)) {
1356                 if (bond_dev->type != slave_dev->type) {
1357                         netdev_dbg(bond_dev, "change device type from %d to %d\n",
1358                                    bond_dev->type, slave_dev->type);
1359
1360                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1361                                                        bond_dev);
1362                         res = notifier_to_errno(res);
1363                         if (res) {
1364                                 netdev_err(bond_dev, "refused to change device type\n");
1365                                 res = -EBUSY;
1366                                 goto err_undo_flags;
1367                         }
1368
1369                         /* Flush unicast and multicast addresses */
1370                         dev_uc_flush(bond_dev);
1371                         dev_mc_flush(bond_dev);
1372
1373                         if (slave_dev->type != ARPHRD_ETHER)
1374                                 bond_setup_by_slave(bond_dev, slave_dev);
1375                         else {
1376                                 ether_setup(bond_dev);
1377                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1378                         }
1379
1380                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1381                                                  bond_dev);
1382                 }
1383         } else if (bond_dev->type != slave_dev->type) {
1384                 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1385                            slave_dev->name, slave_dev->type, bond_dev->type);
1386                 res = -EINVAL;
1387                 goto err_undo_flags;
1388         }
1389
1390         if (slave_ops->ndo_set_mac_address == NULL) {
1391                 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1392                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1393                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1394                         if (!bond_has_slaves(bond)) {
1395                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1396                                 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1397                         } else {
1398                                 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");
1399                                 res = -EOPNOTSUPP;
1400                                 goto err_undo_flags;
1401                         }
1402                 }
1403         }
1404
1405         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1406
1407         /* If this is the first slave, then we need to set the master's hardware
1408          * address to be the same as the slave's.
1409          */
1410         if (!bond_has_slaves(bond) &&
1411             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1412                 bond_set_dev_addr(bond->dev, slave_dev);
1413
1414         new_slave = bond_alloc_slave(bond);
1415         if (!new_slave) {
1416                 res = -ENOMEM;
1417                 goto err_undo_flags;
1418         }
1419
1420         new_slave->bond = bond;
1421         new_slave->dev = slave_dev;
1422         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1423          * is set via sysfs or module option if desired.
1424          */
1425         new_slave->queue_id = 0;
1426
1427         /* Save slave's original mtu and then set it to match the bond */
1428         new_slave->original_mtu = slave_dev->mtu;
1429         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1430         if (res) {
1431                 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1432                 goto err_free;
1433         }
1434
1435         /* Save slave's original ("permanent") mac address for modes
1436          * that need it, and for restoring it upon release, and then
1437          * set it to the master's address
1438          */
1439         ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1440
1441         if (!bond->params.fail_over_mac ||
1442             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1443                 /* Set slave to master's mac address.  The application already
1444                  * set the master's mac address to that of the first slave
1445                  */
1446                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1447                 addr.sa_family = slave_dev->type;
1448                 res = dev_set_mac_address(slave_dev, &addr);
1449                 if (res) {
1450                         netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1451                         goto err_restore_mtu;
1452                 }
1453         }
1454
1455         /* open the slave since the application closed it */
1456         res = dev_open(slave_dev);
1457         if (res) {
1458                 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1459                 goto err_restore_mac;
1460         }
1461
1462         slave_dev->priv_flags |= IFF_BONDING;
1463         /* initialize slave stats */
1464         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1465
1466         if (bond_is_lb(bond)) {
1467                 /* bond_alb_init_slave() must be called before all other stages since
1468                  * it might fail and we do not want to have to undo everything
1469                  */
1470                 res = bond_alb_init_slave(bond, new_slave);
1471                 if (res)
1472                         goto err_close;
1473         }
1474
1475         /* If the mode uses primary, then the following is handled by
1476          * bond_change_active_slave().
1477          */
1478         if (!bond_uses_primary(bond)) {
1479                 /* set promiscuity level to new slave */
1480                 if (bond_dev->flags & IFF_PROMISC) {
1481                         res = dev_set_promiscuity(slave_dev, 1);
1482                         if (res)
1483                                 goto err_close;
1484                 }
1485
1486                 /* set allmulti level to new slave */
1487                 if (bond_dev->flags & IFF_ALLMULTI) {
1488                         res = dev_set_allmulti(slave_dev, 1);
1489                         if (res)
1490                                 goto err_close;
1491                 }
1492
1493                 netif_addr_lock_bh(bond_dev);
1494
1495                 dev_mc_sync_multiple(slave_dev, bond_dev);
1496                 dev_uc_sync_multiple(slave_dev, bond_dev);
1497
1498                 netif_addr_unlock_bh(bond_dev);
1499         }
1500
1501         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1502                 /* add lacpdu mc addr to mc list */
1503                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1504
1505                 dev_mc_add(slave_dev, lacpdu_multicast);
1506         }
1507
1508         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1509         if (res) {
1510                 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1511                            slave_dev->name);
1512                 goto err_close;
1513         }
1514
1515         prev_slave = bond_last_slave(bond);
1516
1517         new_slave->delay = 0;
1518         new_slave->link_failure_count = 0;
1519
1520         bond_update_speed_duplex(new_slave);
1521
1522         new_slave->last_rx = jiffies -
1523                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1524         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1525                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1526
1527         if (bond->params.miimon && !bond->params.use_carrier) {
1528                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1529
1530                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1531                         /* miimon is set but a bonded network driver
1532                          * does not support ETHTOOL/MII and
1533                          * arp_interval is not set.  Note: if
1534                          * use_carrier is enabled, we will never go
1535                          * here (because netif_carrier is always
1536                          * supported); thus, we don't need to change
1537                          * the messages for netif_carrier.
1538                          */
1539                         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",
1540                                     slave_dev->name);
1541                 } else if (link_reporting == -1) {
1542                         /* unable get link status using mii/ethtool */
1543                         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",
1544                                     slave_dev->name);
1545                 }
1546         }
1547
1548         /* check for initial state */
1549         if (bond->params.miimon) {
1550                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1551                         if (bond->params.updelay) {
1552                                 bond_set_slave_link_state(new_slave,
1553                                                           BOND_LINK_BACK);
1554                                 new_slave->delay = bond->params.updelay;
1555                         } else {
1556                                 bond_set_slave_link_state(new_slave,
1557                                                           BOND_LINK_UP);
1558                         }
1559                 } else {
1560                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN);
1561                 }
1562         } else if (bond->params.arp_interval) {
1563                 bond_set_slave_link_state(new_slave,
1564                                           (netif_carrier_ok(slave_dev) ?
1565                                           BOND_LINK_UP : BOND_LINK_DOWN));
1566         } else {
1567                 bond_set_slave_link_state(new_slave, BOND_LINK_UP);
1568         }
1569
1570         if (new_slave->link != BOND_LINK_DOWN)
1571                 new_slave->last_link_up = jiffies;
1572         netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1573                    new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1574                    (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1575
1576         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1577                 /* if there is a primary slave, remember it */
1578                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1579                         rcu_assign_pointer(bond->primary_slave, new_slave);
1580                         bond->force_primary = true;
1581                 }
1582         }
1583
1584         switch (BOND_MODE(bond)) {
1585         case BOND_MODE_ACTIVEBACKUP:
1586                 bond_set_slave_inactive_flags(new_slave,
1587                                               BOND_SLAVE_NOTIFY_NOW);
1588                 break;
1589         case BOND_MODE_8023AD:
1590                 /* in 802.3ad mode, the internal mechanism
1591                  * will activate the slaves in the selected
1592                  * aggregator
1593                  */
1594                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1595                 /* if this is the first slave */
1596                 if (!prev_slave) {
1597                         SLAVE_AD_INFO(new_slave)->id = 1;
1598                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1599                          * can be called only after the mac address of the bond is set
1600                          */
1601                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1602                 } else {
1603                         SLAVE_AD_INFO(new_slave)->id =
1604                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1605                 }
1606
1607                 bond_3ad_bind_slave(new_slave);
1608                 break;
1609         case BOND_MODE_TLB:
1610         case BOND_MODE_ALB:
1611                 bond_set_active_slave(new_slave);
1612                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1613                 break;
1614         default:
1615                 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1616
1617                 /* always active in trunk mode */
1618                 bond_set_active_slave(new_slave);
1619
1620                 /* In trunking mode there is little meaning to curr_active_slave
1621                  * anyway (it holds no special properties of the bond device),
1622                  * so we can change it without calling change_active_interface()
1623                  */
1624                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1625                     new_slave->link == BOND_LINK_UP)
1626                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1627
1628                 break;
1629         } /* switch(bond_mode) */
1630
1631 #ifdef CONFIG_NET_POLL_CONTROLLER
1632         slave_dev->npinfo = bond->dev->npinfo;
1633         if (slave_dev->npinfo) {
1634                 if (slave_enable_netpoll(new_slave)) {
1635                         netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1636                         res = -EBUSY;
1637                         goto err_detach;
1638                 }
1639         }
1640 #endif
1641
1642         if (!(bond_dev->features & NETIF_F_LRO))
1643                 dev_disable_lro(slave_dev);
1644
1645         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1646                                          new_slave);
1647         if (res) {
1648                 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1649                 goto err_detach;
1650         }
1651
1652         res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1653         if (res) {
1654                 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1655                 goto err_unregister;
1656         }
1657
1658         res = bond_sysfs_slave_add(new_slave);
1659         if (res) {
1660                 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1661                 goto err_upper_unlink;
1662         }
1663
1664         bond->slave_cnt++;
1665         bond_compute_features(bond);
1666         bond_set_carrier(bond);
1667
1668         if (bond_uses_primary(bond)) {
1669                 block_netpoll_tx();
1670                 bond_select_active_slave(bond);
1671                 unblock_netpoll_tx();
1672         }
1673
1674         if (bond_mode_uses_xmit_hash(bond))
1675                 bond_update_slave_arr(bond, NULL);
1676
1677         netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1678                     slave_dev->name,
1679                     bond_is_active_slave(new_slave) ? "an active" : "a backup",
1680                     new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1681
1682         /* enslave is successful */
1683         bond_queue_slave_event(new_slave);
1684         return 0;
1685
1686 /* Undo stages on error */
1687 err_upper_unlink:
1688         bond_upper_dev_unlink(bond_dev, slave_dev);
1689
1690 err_unregister:
1691         netdev_rx_handler_unregister(slave_dev);
1692
1693 err_detach:
1694         if (!bond_uses_primary(bond))
1695                 bond_hw_addr_flush(bond_dev, slave_dev);
1696
1697         vlan_vids_del_by_dev(slave_dev, bond_dev);
1698         if (rcu_access_pointer(bond->primary_slave) == new_slave)
1699                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1700         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1701                 block_netpoll_tx();
1702                 bond_change_active_slave(bond, NULL);
1703                 bond_select_active_slave(bond);
1704                 unblock_netpoll_tx();
1705         }
1706         /* either primary_slave or curr_active_slave might've changed */
1707         synchronize_rcu();
1708         slave_disable_netpoll(new_slave);
1709
1710 err_close:
1711         slave_dev->priv_flags &= ~IFF_BONDING;
1712         dev_close(slave_dev);
1713
1714 err_restore_mac:
1715         if (!bond->params.fail_over_mac ||
1716             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1717                 /* XXX TODO - fom follow mode needs to change master's
1718                  * MAC if this slave's MAC is in use by the bond, or at
1719                  * least print a warning.
1720                  */
1721                 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1722                 addr.sa_family = slave_dev->type;
1723                 dev_set_mac_address(slave_dev, &addr);
1724         }
1725
1726 err_restore_mtu:
1727         dev_set_mtu(slave_dev, new_slave->original_mtu);
1728
1729 err_free:
1730         bond_free_slave(new_slave);
1731
1732 err_undo_flags:
1733         /* Enslave of first slave has failed and we need to fix master's mac */
1734         if (!bond_has_slaves(bond) &&
1735             ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
1736                 eth_hw_addr_random(bond_dev);
1737
1738         return res;
1739 }
1740
1741 /* Try to release the slave device <slave> from the bond device <master>
1742  * It is legal to access curr_active_slave without a lock because all the function
1743  * is RTNL-locked. If "all" is true it means that the function is being called
1744  * while destroying a bond interface and all slaves are being released.
1745  *
1746  * The rules for slave state should be:
1747  *   for Active/Backup:
1748  *     Active stays on all backups go down
1749  *   for Bonded connections:
1750  *     The first up interface should be left on and all others downed.
1751  */
1752 static int __bond_release_one(struct net_device *bond_dev,
1753                               struct net_device *slave_dev,
1754                               bool all)
1755 {
1756         struct bonding *bond = netdev_priv(bond_dev);
1757         struct slave *slave, *oldcurrent;
1758         struct sockaddr addr;
1759         int old_flags = bond_dev->flags;
1760         netdev_features_t old_features = bond_dev->features;
1761
1762         /* slave is not a slave or master is not master of this slave */
1763         if (!(slave_dev->flags & IFF_SLAVE) ||
1764             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1765                 netdev_dbg(bond_dev, "cannot release %s\n",
1766                            slave_dev->name);
1767                 return -EINVAL;
1768         }
1769
1770         block_netpoll_tx();
1771
1772         slave = bond_get_slave_by_dev(bond, slave_dev);
1773         if (!slave) {
1774                 /* not a slave of this bond */
1775                 netdev_info(bond_dev, "%s not enslaved\n",
1776                             slave_dev->name);
1777                 unblock_netpoll_tx();
1778                 return -EINVAL;
1779         }
1780
1781         bond_sysfs_slave_del(slave);
1782
1783         /* recompute stats just before removing the slave */
1784         bond_get_stats(bond->dev, &bond->bond_stats);
1785
1786         bond_upper_dev_unlink(bond_dev, slave_dev);
1787         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1788          * for this slave anymore.
1789          */
1790         netdev_rx_handler_unregister(slave_dev);
1791
1792         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1793                 bond_3ad_unbind_slave(slave);
1794
1795         if (bond_mode_uses_xmit_hash(bond))
1796                 bond_update_slave_arr(bond, slave);
1797
1798         netdev_info(bond_dev, "Releasing %s interface %s\n",
1799                     bond_is_active_slave(slave) ? "active" : "backup",
1800                     slave_dev->name);
1801
1802         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1803
1804         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1805
1806         if (!all && (!bond->params.fail_over_mac ||
1807                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1808                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1809                     bond_has_slaves(bond))
1810                         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",
1811                                     slave_dev->name, slave->perm_hwaddr,
1812                                     bond_dev->name, slave_dev->name);
1813         }
1814
1815         if (rtnl_dereference(bond->primary_slave) == slave)
1816                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1817
1818         if (oldcurrent == slave)
1819                 bond_change_active_slave(bond, NULL);
1820
1821         if (bond_is_lb(bond)) {
1822                 /* Must be called only after the slave has been
1823                  * detached from the list and the curr_active_slave
1824                  * has been cleared (if our_slave == old_current),
1825                  * but before a new active slave is selected.
1826                  */
1827                 bond_alb_deinit_slave(bond, slave);
1828         }
1829
1830         if (all) {
1831                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1832         } else if (oldcurrent == slave) {
1833                 /* Note that we hold RTNL over this sequence, so there
1834                  * is no concern that another slave add/remove event
1835                  * will interfere.
1836                  */
1837                 bond_select_active_slave(bond);
1838         }
1839
1840         if (!bond_has_slaves(bond)) {
1841                 bond_set_carrier(bond);
1842                 eth_hw_addr_random(bond_dev);
1843         }
1844
1845         unblock_netpoll_tx();
1846         synchronize_rcu();
1847         bond->slave_cnt--;
1848
1849         if (!bond_has_slaves(bond)) {
1850                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1851                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1852         }
1853
1854         bond_compute_features(bond);
1855         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1856             (old_features & NETIF_F_VLAN_CHALLENGED))
1857                 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1858                             slave_dev->name, bond_dev->name);
1859
1860         vlan_vids_del_by_dev(slave_dev, bond_dev);
1861
1862         /* If the mode uses primary, then this case was handled above by
1863          * bond_change_active_slave(..., NULL)
1864          */
1865         if (!bond_uses_primary(bond)) {
1866                 /* unset promiscuity level from slave
1867                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1868                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1869                  * value of that flag before that change, as that was the value
1870                  * when this slave was attached, so we cache at the start of the
1871                  * function and use it here. Same goes for ALLMULTI below
1872                  */
1873                 if (old_flags & IFF_PROMISC)
1874                         dev_set_promiscuity(slave_dev, -1);
1875
1876                 /* unset allmulti level from slave */
1877                 if (old_flags & IFF_ALLMULTI)
1878                         dev_set_allmulti(slave_dev, -1);
1879
1880                 bond_hw_addr_flush(bond_dev, slave_dev);
1881         }
1882
1883         slave_disable_netpoll(slave);
1884
1885         /* close slave before restoring its mac address */
1886         dev_close(slave_dev);
1887
1888         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1889             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1890                 /* restore original ("permanent") mac address */
1891                 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1892                 addr.sa_family = slave_dev->type;
1893                 dev_set_mac_address(slave_dev, &addr);
1894         }
1895
1896         dev_set_mtu(slave_dev, slave->original_mtu);
1897
1898         slave_dev->priv_flags &= ~IFF_BONDING;
1899
1900         bond_free_slave(slave);
1901
1902         return 0;
1903 }
1904
1905 /* A wrapper used because of ndo_del_link */
1906 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1907 {
1908         return __bond_release_one(bond_dev, slave_dev, false);
1909 }
1910
1911 /* First release a slave and then destroy the bond if no more slaves are left.
1912  * Must be under rtnl_lock when this function is called.
1913  */
1914 static int  bond_release_and_destroy(struct net_device *bond_dev,
1915                                      struct net_device *slave_dev)
1916 {
1917         struct bonding *bond = netdev_priv(bond_dev);
1918         int ret;
1919
1920         ret = bond_release(bond_dev, slave_dev);
1921         if (ret == 0 && !bond_has_slaves(bond)) {
1922                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1923                 netdev_info(bond_dev, "Destroying bond %s\n",
1924                             bond_dev->name);
1925                 bond_remove_proc_entry(bond);
1926                 unregister_netdevice(bond_dev);
1927         }
1928         return ret;
1929 }
1930
1931 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1932 {
1933         struct bonding *bond = netdev_priv(bond_dev);
1934         bond_fill_ifbond(bond, info);
1935         return 0;
1936 }
1937
1938 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1939 {
1940         struct bonding *bond = netdev_priv(bond_dev);
1941         struct list_head *iter;
1942         int i = 0, res = -ENODEV;
1943         struct slave *slave;
1944
1945         bond_for_each_slave(bond, slave, iter) {
1946                 if (i++ == (int)info->slave_id) {
1947                         res = 0;
1948                         bond_fill_ifslave(slave, info);
1949                         break;
1950                 }
1951         }
1952
1953         return res;
1954 }
1955
1956 /*-------------------------------- Monitoring -------------------------------*/
1957
1958 /* called with rcu_read_lock() */
1959 static int bond_miimon_inspect(struct bonding *bond)
1960 {
1961         int link_state, commit = 0;
1962         struct list_head *iter;
1963         struct slave *slave;
1964         bool ignore_updelay;
1965
1966         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
1967
1968         bond_for_each_slave_rcu(bond, slave, iter) {
1969                 slave->new_link = BOND_LINK_NOCHANGE;
1970
1971                 link_state = bond_check_dev_link(bond, slave->dev, 0);
1972
1973                 switch (slave->link) {
1974                 case BOND_LINK_UP:
1975                         if (link_state)
1976                                 continue;
1977
1978                         bond_set_slave_link_state(slave, BOND_LINK_FAIL);
1979                         slave->delay = bond->params.downdelay;
1980                         if (slave->delay) {
1981                                 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
1982                                             (BOND_MODE(bond) ==
1983                                              BOND_MODE_ACTIVEBACKUP) ?
1984                                              (bond_is_active_slave(slave) ?
1985                                               "active " : "backup ") : "",
1986                                             slave->dev->name,
1987                                             bond->params.downdelay * bond->params.miimon);
1988                         }
1989                         /*FALLTHRU*/
1990                 case BOND_LINK_FAIL:
1991                         if (link_state) {
1992                                 /* recovered before downdelay expired */
1993                                 bond_set_slave_link_state(slave, BOND_LINK_UP);
1994                                 slave->last_link_up = jiffies;
1995                                 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
1996                                             (bond->params.downdelay - slave->delay) *
1997                                             bond->params.miimon,
1998                                             slave->dev->name);
1999                                 continue;
2000                         }
2001
2002                         if (slave->delay <= 0) {
2003                                 slave->new_link = BOND_LINK_DOWN;
2004                                 commit++;
2005                                 continue;
2006                         }
2007
2008                         slave->delay--;
2009                         break;
2010
2011                 case BOND_LINK_DOWN:
2012                         if (!link_state)
2013                                 continue;
2014
2015                         bond_set_slave_link_state(slave, BOND_LINK_BACK);
2016                         slave->delay = bond->params.updelay;
2017
2018                         if (slave->delay) {
2019                                 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2020                                             slave->dev->name,
2021                                             ignore_updelay ? 0 :
2022                                             bond->params.updelay *
2023                                             bond->params.miimon);
2024                         }
2025                         /*FALLTHRU*/
2026                 case BOND_LINK_BACK:
2027                         if (!link_state) {
2028                                 bond_set_slave_link_state(slave,
2029                                                           BOND_LINK_DOWN);
2030                                 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2031                                             (bond->params.updelay - slave->delay) *
2032                                             bond->params.miimon,
2033                                             slave->dev->name);
2034
2035                                 continue;
2036                         }
2037
2038                         if (ignore_updelay)
2039                                 slave->delay = 0;
2040
2041                         if (slave->delay <= 0) {
2042                                 slave->new_link = BOND_LINK_UP;
2043                                 commit++;
2044                                 ignore_updelay = false;
2045                                 continue;
2046                         }
2047
2048                         slave->delay--;
2049                         break;
2050                 }
2051         }
2052
2053         return commit;
2054 }
2055
2056 static void bond_miimon_commit(struct bonding *bond)
2057 {
2058         struct list_head *iter;
2059         struct slave *slave, *primary;
2060
2061         bond_for_each_slave(bond, slave, iter) {
2062                 switch (slave->new_link) {
2063                 case BOND_LINK_NOCHANGE:
2064                         continue;
2065
2066                 case BOND_LINK_UP:
2067                         bond_set_slave_link_state(slave, BOND_LINK_UP);
2068                         slave->last_link_up = jiffies;
2069
2070                         primary = rtnl_dereference(bond->primary_slave);
2071                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2072                                 /* prevent it from being the active one */
2073                                 bond_set_backup_slave(slave);
2074                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2075                                 /* make it immediately active */
2076                                 bond_set_active_slave(slave);
2077                         } else if (slave != primary) {
2078                                 /* prevent it from being the active one */
2079                                 bond_set_backup_slave(slave);
2080                         }
2081
2082                         netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2083                                     slave->dev->name,
2084                                     slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2085                                     slave->duplex ? "full" : "half");
2086
2087                         /* notify ad that the link status has changed */
2088                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2089                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2090
2091                         if (bond_is_lb(bond))
2092                                 bond_alb_handle_link_change(bond, slave,
2093                                                             BOND_LINK_UP);
2094
2095                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2096                                 bond_update_slave_arr(bond, NULL);
2097
2098                         if (!bond->curr_active_slave || slave == primary)
2099                                 goto do_failover;
2100
2101                         continue;
2102
2103                 case BOND_LINK_DOWN:
2104                         if (slave->link_failure_count < UINT_MAX)
2105                                 slave->link_failure_count++;
2106
2107                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2108
2109                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2110                             BOND_MODE(bond) == BOND_MODE_8023AD)
2111                                 bond_set_slave_inactive_flags(slave,
2112                                                               BOND_SLAVE_NOTIFY_NOW);
2113
2114                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2115                                     slave->dev->name);
2116
2117                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2118                                 bond_3ad_handle_link_change(slave,
2119                                                             BOND_LINK_DOWN);
2120
2121                         if (bond_is_lb(bond))
2122                                 bond_alb_handle_link_change(bond, slave,
2123                                                             BOND_LINK_DOWN);
2124
2125                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2126                                 bond_update_slave_arr(bond, NULL);
2127
2128                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2129                                 goto do_failover;
2130
2131                         continue;
2132
2133                 default:
2134                         netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2135                                    slave->new_link, slave->dev->name);
2136                         slave->new_link = BOND_LINK_NOCHANGE;
2137
2138                         continue;
2139                 }
2140
2141 do_failover:
2142                 block_netpoll_tx();
2143                 bond_select_active_slave(bond);
2144                 unblock_netpoll_tx();
2145         }
2146
2147         bond_set_carrier(bond);
2148 }
2149
2150 /* bond_mii_monitor
2151  *
2152  * Really a wrapper that splits the mii monitor into two phases: an
2153  * inspection, then (if inspection indicates something needs to be done)
2154  * an acquisition of appropriate locks followed by a commit phase to
2155  * implement whatever link state changes are indicated.
2156  */
2157 static void bond_mii_monitor(struct work_struct *work)
2158 {
2159         struct bonding *bond = container_of(work, struct bonding,
2160                                             mii_work.work);
2161         bool should_notify_peers = false;
2162         unsigned long delay;
2163
2164         delay = msecs_to_jiffies(bond->params.miimon);
2165
2166         if (!bond_has_slaves(bond))
2167                 goto re_arm;
2168
2169         rcu_read_lock();
2170
2171         should_notify_peers = bond_should_notify_peers(bond);
2172
2173         if (bond_miimon_inspect(bond)) {
2174                 rcu_read_unlock();
2175
2176                 /* Race avoidance with bond_close cancel of workqueue */
2177                 if (!rtnl_trylock()) {
2178                         delay = 1;
2179                         should_notify_peers = false;
2180                         goto re_arm;
2181                 }
2182
2183                 bond_miimon_commit(bond);
2184
2185                 rtnl_unlock();  /* might sleep, hold no other locks */
2186         } else
2187                 rcu_read_unlock();
2188
2189 re_arm:
2190         if (bond->params.miimon)
2191                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2192
2193         if (should_notify_peers) {
2194                 if (!rtnl_trylock())
2195                         return;
2196                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2197                 rtnl_unlock();
2198         }
2199 }
2200
2201 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2202 {
2203         struct net_device *upper;
2204         struct list_head *iter;
2205         bool ret = false;
2206
2207         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2208                 return true;
2209
2210         rcu_read_lock();
2211         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2212                 if (ip == bond_confirm_addr(upper, 0, ip)) {
2213                         ret = true;
2214                         break;
2215                 }
2216         }
2217         rcu_read_unlock();
2218
2219         return ret;
2220 }
2221
2222 /* We go to the (large) trouble of VLAN tagging ARP frames because
2223  * switches in VLAN mode (especially if ports are configured as
2224  * "native" to a VLAN) might not pass non-tagged frames.
2225  */
2226 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2227                           __be32 dest_ip, __be32 src_ip,
2228                           struct bond_vlan_tag *tags)
2229 {
2230         struct sk_buff *skb;
2231         struct bond_vlan_tag *outer_tag = tags;
2232
2233         netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2234                    arp_op, slave_dev->name, &dest_ip, &src_ip);
2235
2236         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2237                          NULL, slave_dev->dev_addr, NULL);
2238
2239         if (!skb) {
2240                 net_err_ratelimited("ARP packet allocation failed\n");
2241                 return;
2242         }
2243
2244         if (!tags || tags->vlan_proto == VLAN_N_VID)
2245                 goto xmit;
2246
2247         tags++;
2248
2249         /* Go through all the tags backwards and add them to the packet */
2250         while (tags->vlan_proto != VLAN_N_VID) {
2251                 if (!tags->vlan_id) {
2252                         tags++;
2253                         continue;
2254                 }
2255
2256                 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2257                            ntohs(outer_tag->vlan_proto), tags->vlan_id);
2258                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2259                                                 tags->vlan_id);
2260                 if (!skb) {
2261                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2262                         return;
2263                 }
2264
2265                 tags++;
2266         }
2267         /* Set the outer tag */
2268         if (outer_tag->vlan_id) {
2269                 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2270                            ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2271                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2272                                        outer_tag->vlan_id);
2273         }
2274
2275 xmit:
2276         arp_xmit(skb);
2277 }
2278
2279 /* Validate the device path between the @start_dev and the @end_dev.
2280  * The path is valid if the @end_dev is reachable through device
2281  * stacking.
2282  * When the path is validated, collect any vlan information in the
2283  * path.
2284  */
2285 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2286                                               struct net_device *end_dev,
2287                                               int level)
2288 {
2289         struct bond_vlan_tag *tags;
2290         struct net_device *upper;
2291         struct list_head  *iter;
2292
2293         if (start_dev == end_dev) {
2294                 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2295                 if (!tags)
2296                         return ERR_PTR(-ENOMEM);
2297                 tags[level].vlan_proto = VLAN_N_VID;
2298                 return tags;
2299         }
2300
2301         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2302                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2303                 if (IS_ERR_OR_NULL(tags)) {
2304                         if (IS_ERR(tags))
2305                                 return tags;
2306                         continue;
2307                 }
2308                 if (is_vlan_dev(upper)) {
2309                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2310                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2311                 }
2312
2313                 return tags;
2314         }
2315
2316         return NULL;
2317 }
2318
2319 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2320 {
2321         struct rtable *rt;
2322         struct bond_vlan_tag *tags;
2323         __be32 *targets = bond->params.arp_targets, addr;
2324         int i;
2325
2326         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2327                 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2328                 tags = NULL;
2329
2330                 /* Find out through which dev should the packet go */
2331                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2332                                      RTO_ONLINK, 0);
2333                 if (IS_ERR(rt)) {
2334                         /* there's no route to target - try to send arp
2335                          * probe to generate any traffic (arp_validate=0)
2336                          */
2337                         if (bond->params.arp_validate)
2338                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2339                                                      bond->dev->name,
2340                                                      &targets[i]);
2341                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2342                                       0, tags);
2343                         continue;
2344                 }
2345
2346                 /* bond device itself */
2347                 if (rt->dst.dev == bond->dev)
2348                         goto found;
2349
2350                 rcu_read_lock();
2351                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2352                 rcu_read_unlock();
2353
2354                 if (!IS_ERR_OR_NULL(tags))
2355                         goto found;
2356
2357                 /* Not our device - skip */
2358                 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2359                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2360
2361                 ip_rt_put(rt);
2362                 continue;
2363
2364 found:
2365                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2366                 ip_rt_put(rt);
2367                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2368                               addr, tags);
2369                 kfree(tags);
2370         }
2371 }
2372
2373 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2374 {
2375         int i;
2376
2377         if (!sip || !bond_has_this_ip(bond, tip)) {
2378                 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2379                            &sip, &tip);
2380                 return;
2381         }
2382
2383         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2384         if (i == -1) {
2385                 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2386                            &sip);
2387                 return;
2388         }
2389         slave->last_rx = jiffies;
2390         slave->target_last_arp_rx[i] = jiffies;
2391 }
2392
2393 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2394                  struct slave *slave)
2395 {
2396         struct arphdr *arp = (struct arphdr *)skb->data;
2397         struct slave *curr_active_slave;
2398         unsigned char *arp_ptr;
2399         __be32 sip, tip;
2400         int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2401
2402         if (!slave_do_arp_validate(bond, slave)) {
2403                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2404                     !slave_do_arp_validate_only(bond))
2405                         slave->last_rx = jiffies;
2406                 return RX_HANDLER_ANOTHER;
2407         } else if (!is_arp) {
2408                 return RX_HANDLER_ANOTHER;
2409         }
2410
2411         alen = arp_hdr_len(bond->dev);
2412
2413         netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2414                    skb->dev->name);
2415
2416         if (alen > skb_headlen(skb)) {
2417                 arp = kmalloc(alen, GFP_ATOMIC);
2418                 if (!arp)
2419                         goto out_unlock;
2420                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2421                         goto out_unlock;
2422         }
2423
2424         if (arp->ar_hln != bond->dev->addr_len ||
2425             skb->pkt_type == PACKET_OTHERHOST ||
2426             skb->pkt_type == PACKET_LOOPBACK ||
2427             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2428             arp->ar_pro != htons(ETH_P_IP) ||
2429             arp->ar_pln != 4)
2430                 goto out_unlock;
2431
2432         arp_ptr = (unsigned char *)(arp + 1);
2433         arp_ptr += bond->dev->addr_len;
2434         memcpy(&sip, arp_ptr, 4);
2435         arp_ptr += 4 + bond->dev->addr_len;
2436         memcpy(&tip, arp_ptr, 4);
2437
2438         netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2439                    slave->dev->name, bond_slave_state(slave),
2440                      bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2441                      &sip, &tip);
2442
2443         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2444
2445         /* Backup slaves won't see the ARP reply, but do come through
2446          * here for each ARP probe (so we swap the sip/tip to validate
2447          * the probe).  In a "redundant switch, common router" type of
2448          * configuration, the ARP probe will (hopefully) travel from
2449          * the active, through one switch, the router, then the other
2450          * switch before reaching the backup.
2451          *
2452          * We 'trust' the arp requests if there is an active slave and
2453          * it received valid arp reply(s) after it became active. This
2454          * is done to avoid endless looping when we can't reach the
2455          * arp_ip_target and fool ourselves with our own arp requests.
2456          */
2457
2458         if (bond_is_active_slave(slave))
2459                 bond_validate_arp(bond, slave, sip, tip);
2460         else if (curr_active_slave &&
2461                  time_after(slave_last_rx(bond, curr_active_slave),
2462                             curr_active_slave->last_link_up))
2463                 bond_validate_arp(bond, slave, tip, sip);
2464
2465 out_unlock:
2466         if (arp != (struct arphdr *)skb->data)
2467                 kfree(arp);
2468         return RX_HANDLER_ANOTHER;
2469 }
2470
2471 /* function to verify if we're in the arp_interval timeslice, returns true if
2472  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2473  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2474  */
2475 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2476                                   int mod)
2477 {
2478         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2479
2480         return time_in_range(jiffies,
2481                              last_act - delta_in_ticks,
2482                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2483 }
2484
2485 /* This function is called regularly to monitor each slave's link
2486  * ensuring that traffic is being sent and received when arp monitoring
2487  * is used in load-balancing mode. if the adapter has been dormant, then an
2488  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2489  * arp monitoring in active backup mode.
2490  */
2491 static void bond_loadbalance_arp_mon(struct work_struct *work)
2492 {
2493         struct bonding *bond = container_of(work, struct bonding,
2494                                             arp_work.work);
2495         struct slave *slave, *oldcurrent;
2496         struct list_head *iter;
2497         int do_failover = 0, slave_state_changed = 0;
2498
2499         if (!bond_has_slaves(bond))
2500                 goto re_arm;
2501
2502         rcu_read_lock();
2503
2504         oldcurrent = rcu_dereference(bond->curr_active_slave);
2505         /* see if any of the previous devices are up now (i.e. they have
2506          * xmt and rcv traffic). the curr_active_slave does not come into
2507          * the picture unless it is null. also, slave->last_link_up is not
2508          * needed here because we send an arp on each slave and give a slave
2509          * as long as it needs to get the tx/rx within the delta.
2510          * TODO: what about up/down delay in arp mode? it wasn't here before
2511          *       so it can wait
2512          */
2513         bond_for_each_slave_rcu(bond, slave, iter) {
2514                 unsigned long trans_start = dev_trans_start(slave->dev);
2515
2516                 if (slave->link != BOND_LINK_UP) {
2517                         if (bond_time_in_interval(bond, trans_start, 1) &&
2518                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2519
2520                                 slave->link  = BOND_LINK_UP;
2521                                 slave_state_changed = 1;
2522
2523                                 /* primary_slave has no meaning in round-robin
2524                                  * mode. the window of a slave being up and
2525                                  * curr_active_slave being null after enslaving
2526                                  * is closed.
2527                                  */
2528                                 if (!oldcurrent) {
2529                                         netdev_info(bond->dev, "link status definitely up for interface %s\n",
2530                                                     slave->dev->name);
2531                                         do_failover = 1;
2532                                 } else {
2533                                         netdev_info(bond->dev, "interface %s is now up\n",
2534                                                     slave->dev->name);
2535                                 }
2536                         }
2537                 } else {
2538                         /* slave->link == BOND_LINK_UP */
2539
2540                         /* not all switches will respond to an arp request
2541                          * when the source ip is 0, so don't take the link down
2542                          * if we don't know our ip yet
2543                          */
2544                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2545                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2546
2547                                 slave->link  = BOND_LINK_DOWN;
2548                                 slave_state_changed = 1;
2549
2550                                 if (slave->link_failure_count < UINT_MAX)
2551                                         slave->link_failure_count++;
2552
2553                                 netdev_info(bond->dev, "interface %s is now down\n",
2554                                             slave->dev->name);
2555
2556                                 if (slave == oldcurrent)
2557                                         do_failover = 1;
2558                         }
2559                 }
2560
2561                 /* note: if switch is in round-robin mode, all links
2562                  * must tx arp to ensure all links rx an arp - otherwise
2563                  * links may oscillate or not come up at all; if switch is
2564                  * in something like xor mode, there is nothing we can
2565                  * do - all replies will be rx'ed on same link causing slaves
2566                  * to be unstable during low/no traffic periods
2567                  */
2568                 if (bond_slave_is_up(slave))
2569                         bond_arp_send_all(bond, slave);
2570         }
2571
2572         rcu_read_unlock();
2573
2574         if (do_failover || slave_state_changed) {
2575                 if (!rtnl_trylock())
2576                         goto re_arm;
2577
2578                 if (slave_state_changed) {
2579                         bond_slave_state_change(bond);
2580                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2581                                 bond_update_slave_arr(bond, NULL);
2582                 }
2583                 if (do_failover) {
2584                         block_netpoll_tx();
2585                         bond_select_active_slave(bond);
2586                         unblock_netpoll_tx();
2587                 }
2588                 rtnl_unlock();
2589         }
2590
2591 re_arm:
2592         if (bond->params.arp_interval)
2593                 queue_delayed_work(bond->wq, &bond->arp_work,
2594                                    msecs_to_jiffies(bond->params.arp_interval));
2595 }
2596
2597 /* Called to inspect slaves for active-backup mode ARP monitor link state
2598  * changes.  Sets new_link in slaves to specify what action should take
2599  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2600  * to link states must be committed.
2601  *
2602  * Called with rcu_read_lock held.
2603  */
2604 static int bond_ab_arp_inspect(struct bonding *bond)
2605 {
2606         unsigned long trans_start, last_rx;
2607         struct list_head *iter;
2608         struct slave *slave;
2609         int commit = 0;
2610
2611         bond_for_each_slave_rcu(bond, slave, iter) {
2612                 slave->new_link = BOND_LINK_NOCHANGE;
2613                 last_rx = slave_last_rx(bond, slave);
2614
2615                 if (slave->link != BOND_LINK_UP) {
2616                         if (bond_time_in_interval(bond, last_rx, 1)) {
2617                                 slave->new_link = BOND_LINK_UP;
2618                                 commit++;
2619                         }
2620                         continue;
2621                 }
2622
2623                 /* Give slaves 2*delta after being enslaved or made
2624                  * active.  This avoids bouncing, as the last receive
2625                  * times need a full ARP monitor cycle to be updated.
2626                  */
2627                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2628                         continue;
2629
2630                 /* Backup slave is down if:
2631                  * - No current_arp_slave AND
2632                  * - more than 3*delta since last receive AND
2633                  * - the bond has an IP address
2634                  *
2635                  * Note: a non-null current_arp_slave indicates
2636                  * the curr_active_slave went down and we are
2637                  * searching for a new one; under this condition
2638                  * we only take the curr_active_slave down - this
2639                  * gives each slave a chance to tx/rx traffic
2640                  * before being taken out
2641                  */
2642                 if (!bond_is_active_slave(slave) &&
2643                     !rcu_access_pointer(bond->current_arp_slave) &&
2644                     !bond_time_in_interval(bond, last_rx, 3)) {
2645                         slave->new_link = BOND_LINK_DOWN;
2646                         commit++;
2647                 }
2648
2649                 /* Active slave is down if:
2650                  * - more than 2*delta since transmitting OR
2651                  * - (more than 2*delta since receive AND
2652                  *    the bond has an IP address)
2653                  */
2654                 trans_start = dev_trans_start(slave->dev);
2655                 if (bond_is_active_slave(slave) &&
2656                     (!bond_time_in_interval(bond, trans_start, 2) ||
2657                      !bond_time_in_interval(bond, last_rx, 2))) {
2658                         slave->new_link = BOND_LINK_DOWN;
2659                         commit++;
2660                 }
2661         }
2662
2663         return commit;
2664 }
2665
2666 /* Called to commit link state changes noted by inspection step of
2667  * active-backup mode ARP monitor.
2668  *
2669  * Called with RTNL hold.
2670  */
2671 static void bond_ab_arp_commit(struct bonding *bond)
2672 {
2673         unsigned long trans_start;
2674         struct list_head *iter;
2675         struct slave *slave;
2676
2677         bond_for_each_slave(bond, slave, iter) {
2678                 switch (slave->new_link) {
2679                 case BOND_LINK_NOCHANGE:
2680                         continue;
2681
2682                 case BOND_LINK_UP:
2683                         trans_start = dev_trans_start(slave->dev);
2684                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
2685                             (!rtnl_dereference(bond->curr_active_slave) &&
2686                              bond_time_in_interval(bond, trans_start, 1))) {
2687                                 struct slave *current_arp_slave;
2688
2689                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2690                                 bond_set_slave_link_state(slave, BOND_LINK_UP);
2691                                 if (current_arp_slave) {
2692                                         bond_set_slave_inactive_flags(
2693                                                 current_arp_slave,
2694                                                 BOND_SLAVE_NOTIFY_NOW);
2695                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2696                                 }
2697
2698                                 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2699                                             slave->dev->name);
2700
2701                                 if (!rtnl_dereference(bond->curr_active_slave) ||
2702                                     slave == rtnl_dereference(bond->primary_slave))
2703                                         goto do_failover;
2704
2705                         }
2706
2707                         continue;
2708
2709                 case BOND_LINK_DOWN:
2710                         if (slave->link_failure_count < UINT_MAX)
2711                                 slave->link_failure_count++;
2712
2713                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2714                         bond_set_slave_inactive_flags(slave,
2715                                                       BOND_SLAVE_NOTIFY_NOW);
2716
2717                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2718                                     slave->dev->name);
2719
2720                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
2721                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2722                                 goto do_failover;
2723                         }
2724
2725                         continue;
2726
2727                 default:
2728                         netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2729                                    slave->new_link, slave->dev->name);
2730                         continue;
2731                 }
2732
2733 do_failover:
2734                 block_netpoll_tx();
2735                 bond_select_active_slave(bond);
2736                 unblock_netpoll_tx();
2737         }
2738
2739         bond_set_carrier(bond);
2740 }
2741
2742 /* Send ARP probes for active-backup mode ARP monitor.
2743  *
2744  * Called with rcu_read_lock held.
2745  */
2746 static bool bond_ab_arp_probe(struct bonding *bond)
2747 {
2748         struct slave *slave, *before = NULL, *new_slave = NULL,
2749                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2750                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2751         struct list_head *iter;
2752         bool found = false;
2753         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2754
2755         if (curr_arp_slave && curr_active_slave)
2756                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2757                             curr_arp_slave->dev->name,
2758                             curr_active_slave->dev->name);
2759
2760         if (curr_active_slave) {
2761                 bond_arp_send_all(bond, curr_active_slave);
2762                 return should_notify_rtnl;
2763         }
2764
2765         /* if we don't have a curr_active_slave, search for the next available
2766          * backup slave from the current_arp_slave and make it the candidate
2767          * for becoming the curr_active_slave
2768          */
2769
2770         if (!curr_arp_slave) {
2771                 curr_arp_slave = bond_first_slave_rcu(bond);
2772                 if (!curr_arp_slave)
2773                         return should_notify_rtnl;
2774         }
2775
2776         bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2777
2778         bond_for_each_slave_rcu(bond, slave, iter) {
2779                 if (!found && !before && bond_slave_is_up(slave))
2780                         before = slave;
2781
2782                 if (found && !new_slave && bond_slave_is_up(slave))
2783                         new_slave = slave;
2784                 /* if the link state is up at this point, we
2785                  * mark it down - this can happen if we have
2786                  * simultaneous link failures and
2787                  * reselect_active_interface doesn't make this
2788                  * one the current slave so it is still marked
2789                  * up when it is actually down
2790                  */
2791                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2792                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2793                         if (slave->link_failure_count < UINT_MAX)
2794                                 slave->link_failure_count++;
2795
2796                         bond_set_slave_inactive_flags(slave,
2797                                                       BOND_SLAVE_NOTIFY_LATER);
2798
2799                         netdev_info(bond->dev, "backup interface %s is now down\n",
2800                                     slave->dev->name);
2801                 }
2802                 if (slave == curr_arp_slave)
2803                         found = true;
2804         }
2805
2806         if (!new_slave && before)
2807                 new_slave = before;
2808
2809         if (!new_slave)
2810                 goto check_state;
2811
2812         bond_set_slave_link_state(new_slave, BOND_LINK_BACK);
2813         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2814         bond_arp_send_all(bond, new_slave);
2815         new_slave->last_link_up = jiffies;
2816         rcu_assign_pointer(bond->current_arp_slave, new_slave);
2817
2818 check_state:
2819         bond_for_each_slave_rcu(bond, slave, iter) {
2820                 if (slave->should_notify) {
2821                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2822                         break;
2823                 }
2824         }
2825         return should_notify_rtnl;
2826 }
2827
2828 static void bond_activebackup_arp_mon(struct work_struct *work)
2829 {
2830         struct bonding *bond = container_of(work, struct bonding,
2831                                             arp_work.work);
2832         bool should_notify_peers = false;
2833         bool should_notify_rtnl = false;
2834         int delta_in_ticks;
2835
2836         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2837
2838         if (!bond_has_slaves(bond))
2839                 goto re_arm;
2840
2841         rcu_read_lock();
2842
2843         should_notify_peers = bond_should_notify_peers(bond);
2844
2845         if (bond_ab_arp_inspect(bond)) {
2846                 rcu_read_unlock();
2847
2848                 /* Race avoidance with bond_close flush of workqueue */
2849                 if (!rtnl_trylock()) {
2850                         delta_in_ticks = 1;
2851                         should_notify_peers = false;
2852                         goto re_arm;
2853                 }
2854
2855                 bond_ab_arp_commit(bond);
2856
2857                 rtnl_unlock();
2858                 rcu_read_lock();
2859         }
2860
2861         should_notify_rtnl = bond_ab_arp_probe(bond);
2862         rcu_read_unlock();
2863
2864 re_arm:
2865         if (bond->params.arp_interval)
2866                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2867
2868         if (should_notify_peers || should_notify_rtnl) {
2869                 if (!rtnl_trylock())
2870                         return;
2871
2872                 if (should_notify_peers)
2873                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2874                                                  bond->dev);
2875                 if (should_notify_rtnl)
2876                         bond_slave_state_notify(bond);
2877
2878                 rtnl_unlock();
2879         }
2880 }
2881
2882 /*-------------------------- netdev event handling --------------------------*/
2883
2884 /* Change device name */
2885 static int bond_event_changename(struct bonding *bond)
2886 {
2887         bond_remove_proc_entry(bond);
2888         bond_create_proc_entry(bond);
2889
2890         bond_debug_reregister(bond);
2891
2892         return NOTIFY_DONE;
2893 }
2894
2895 static int bond_master_netdev_event(unsigned long event,
2896                                     struct net_device *bond_dev)
2897 {
2898         struct bonding *event_bond = netdev_priv(bond_dev);
2899
2900         switch (event) {
2901         case NETDEV_CHANGENAME:
2902                 return bond_event_changename(event_bond);
2903         case NETDEV_UNREGISTER:
2904                 bond_remove_proc_entry(event_bond);
2905                 break;
2906         case NETDEV_REGISTER:
2907                 bond_create_proc_entry(event_bond);
2908                 break;
2909         case NETDEV_NOTIFY_PEERS:
2910                 if (event_bond->send_peer_notif)
2911                         event_bond->send_peer_notif--;
2912                 break;
2913         default:
2914                 break;
2915         }
2916
2917         return NOTIFY_DONE;
2918 }
2919
2920 static int bond_slave_netdev_event(unsigned long event,
2921                                    struct net_device *slave_dev)
2922 {
2923         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
2924         struct bonding *bond;
2925         struct net_device *bond_dev;
2926         u32 old_speed;
2927         u8 old_duplex;
2928
2929         /* A netdev event can be generated while enslaving a device
2930          * before netdev_rx_handler_register is called in which case
2931          * slave will be NULL
2932          */
2933         if (!slave)
2934                 return NOTIFY_DONE;
2935         bond_dev = slave->bond->dev;
2936         bond = slave->bond;
2937         primary = rtnl_dereference(bond->primary_slave);
2938
2939         switch (event) {
2940         case NETDEV_UNREGISTER:
2941                 if (bond_dev->type != ARPHRD_ETHER)
2942                         bond_release_and_destroy(bond_dev, slave_dev);
2943                 else
2944                         bond_release(bond_dev, slave_dev);
2945                 break;
2946         case NETDEV_UP:
2947         case NETDEV_CHANGE:
2948                 old_speed = slave->speed;
2949                 old_duplex = slave->duplex;
2950
2951                 bond_update_speed_duplex(slave);
2952
2953                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2954                         if (old_speed != slave->speed)
2955                                 bond_3ad_adapter_speed_changed(slave);
2956                         if (old_duplex != slave->duplex)
2957                                 bond_3ad_adapter_duplex_changed(slave);
2958                 }
2959                 /* Fallthrough */
2960         case NETDEV_DOWN:
2961                 /* Refresh slave-array if applicable!
2962                  * If the setup does not use miimon or arpmon (mode-specific!),
2963                  * then these events will not cause the slave-array to be
2964                  * refreshed. This will cause xmit to use a slave that is not
2965                  * usable. Avoid such situation by refeshing the array at these
2966                  * events. If these (miimon/arpmon) parameters are configured
2967                  * then array gets refreshed twice and that should be fine!
2968                  */
2969                 if (bond_mode_uses_xmit_hash(bond))
2970                         bond_update_slave_arr(bond, NULL);
2971                 break;
2972         case NETDEV_CHANGEMTU:
2973                 /* TODO: Should slaves be allowed to
2974                  * independently alter their MTU?  For
2975                  * an active-backup bond, slaves need
2976                  * not be the same type of device, so
2977                  * MTUs may vary.  For other modes,
2978                  * slaves arguably should have the
2979                  * same MTUs. To do this, we'd need to
2980                  * take over the slave's change_mtu
2981                  * function for the duration of their
2982                  * servitude.
2983                  */
2984                 break;
2985         case NETDEV_CHANGENAME:
2986                 /* we don't care if we don't have primary set */
2987                 if (!bond_uses_primary(bond) ||
2988                     !bond->params.primary[0])
2989                         break;
2990
2991                 if (slave == primary) {
2992                         /* slave's name changed - he's no longer primary */
2993                         RCU_INIT_POINTER(bond->primary_slave, NULL);
2994                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2995                         /* we have a new primary slave */
2996                         rcu_assign_pointer(bond->primary_slave, slave);
2997                 } else { /* we didn't change primary - exit */
2998                         break;
2999                 }
3000
3001                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3002                             primary ? slave_dev->name : "none");
3003
3004                 block_netpoll_tx();
3005                 bond_select_active_slave(bond);
3006                 unblock_netpoll_tx();
3007                 break;
3008         case NETDEV_FEAT_CHANGE:
3009                 bond_compute_features(bond);
3010                 break;
3011         case NETDEV_RESEND_IGMP:
3012                 /* Propagate to master device */
3013                 call_netdevice_notifiers(event, slave->bond->dev);
3014                 break;
3015         default:
3016                 break;
3017         }
3018
3019         return NOTIFY_DONE;
3020 }
3021
3022 /* bond_netdev_event: handle netdev notifier chain events.
3023  *
3024  * This function receives events for the netdev chain.  The caller (an
3025  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3026  * locks for us to safely manipulate the slave devices (RTNL lock,
3027  * dev_probe_lock).
3028  */
3029 static int bond_netdev_event(struct notifier_block *this,
3030                              unsigned long event, void *ptr)
3031 {
3032         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3033
3034         netdev_dbg(event_dev, "event: %lx\n", event);
3035
3036         if (!(event_dev->priv_flags & IFF_BONDING))
3037                 return NOTIFY_DONE;
3038
3039         if (event_dev->flags & IFF_MASTER) {
3040                 netdev_dbg(event_dev, "IFF_MASTER\n");
3041                 return bond_master_netdev_event(event, event_dev);
3042         }
3043
3044         if (event_dev->flags & IFF_SLAVE) {
3045                 netdev_dbg(event_dev, "IFF_SLAVE\n");
3046                 return bond_slave_netdev_event(event, event_dev);
3047         }
3048
3049         return NOTIFY_DONE;
3050 }
3051
3052 static struct notifier_block bond_netdev_notifier = {
3053         .notifier_call = bond_netdev_event,
3054 };
3055
3056 /*---------------------------- Hashing Policies -----------------------------*/
3057
3058 /* L2 hash helper */
3059 static inline u32 bond_eth_hash(struct sk_buff *skb)
3060 {
3061         struct ethhdr *ep, hdr_tmp;
3062
3063         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3064         if (ep)
3065                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3066         return 0;
3067 }
3068
3069 /* Extract the appropriate headers based on bond's xmit policy */
3070 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3071                               struct flow_keys *fk)
3072 {
3073         const struct ipv6hdr *iph6;
3074         const struct iphdr *iph;
3075         int noff, proto = -1;
3076
3077         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3078                 return skb_flow_dissect(skb, fk);
3079
3080         fk->ports = 0;
3081         noff = skb_network_offset(skb);
3082         if (skb->protocol == htons(ETH_P_IP)) {
3083                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3084                         return false;
3085                 iph = ip_hdr(skb);
3086                 fk->src = iph->saddr;
3087                 fk->dst = iph->daddr;
3088                 noff += iph->ihl << 2;
3089                 if (!ip_is_fragment(iph))
3090                         proto = iph->protocol;
3091         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3092                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3093                         return false;
3094                 iph6 = ipv6_hdr(skb);
3095                 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
3096                 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
3097                 noff += sizeof(*iph6);
3098                 proto = iph6->nexthdr;
3099         } else {
3100                 return false;
3101         }
3102         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3103                 fk->ports = skb_flow_get_ports(skb, noff, proto);
3104
3105         return true;
3106 }
3107
3108 /**
3109  * bond_xmit_hash - generate a hash value based on the xmit policy
3110  * @bond: bonding device
3111  * @skb: buffer to use for headers
3112  *
3113  * This function will extract the necessary headers from the skb buffer and use
3114  * them to generate a hash based on the xmit_policy set in the bonding device
3115  */
3116 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3117 {
3118         struct flow_keys flow;
3119         u32 hash;
3120
3121         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3122             !bond_flow_dissect(bond, skb, &flow))
3123                 return bond_eth_hash(skb);
3124
3125         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3126             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3127                 hash = bond_eth_hash(skb);
3128         else
3129                 hash = (__force u32)flow.ports;
3130         hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
3131         hash ^= (hash >> 16);
3132         hash ^= (hash >> 8);
3133
3134         return hash;
3135 }
3136
3137 /*-------------------------- Device entry points ----------------------------*/
3138
3139 static void bond_work_init_all(struct bonding *bond)
3140 {
3141         INIT_DELAYED_WORK(&bond->mcast_work,
3142                           bond_resend_igmp_join_requests_delayed);
3143         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3144         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3145         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3146                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3147         else
3148                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3149         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3150         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3151 }
3152
3153 static void bond_work_cancel_all(struct bonding *bond)
3154 {
3155         cancel_delayed_work_sync(&bond->mii_work);
3156         cancel_delayed_work_sync(&bond->arp_work);
3157         cancel_delayed_work_sync(&bond->alb_work);
3158         cancel_delayed_work_sync(&bond->ad_work);
3159         cancel_delayed_work_sync(&bond->mcast_work);
3160         cancel_delayed_work_sync(&bond->slave_arr_work);
3161 }
3162
3163 static int bond_open(struct net_device *bond_dev)
3164 {
3165         struct bonding *bond = netdev_priv(bond_dev);
3166         struct list_head *iter;
3167         struct slave *slave;
3168
3169         /* reset slave->backup and slave->inactive */
3170         if (bond_has_slaves(bond)) {
3171                 bond_for_each_slave(bond, slave, iter) {
3172                         if (bond_uses_primary(bond) &&
3173                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3174                                 bond_set_slave_inactive_flags(slave,
3175                                                               BOND_SLAVE_NOTIFY_NOW);
3176                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3177                                 bond_set_slave_active_flags(slave,
3178                                                             BOND_SLAVE_NOTIFY_NOW);
3179                         }
3180                 }
3181         }
3182
3183         bond_work_init_all(bond);
3184
3185         if (bond_is_lb(bond)) {
3186                 /* bond_alb_initialize must be called before the timer
3187                  * is started.
3188                  */
3189                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3190                         return -ENOMEM;
3191                 if (bond->params.tlb_dynamic_lb)
3192                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3193         }
3194
3195         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3196                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3197
3198         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3199                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3200                 bond->recv_probe = bond_arp_rcv;
3201         }
3202
3203         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3204                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3205                 /* register to receive LACPDUs */
3206                 bond->recv_probe = bond_3ad_lacpdu_recv;
3207                 bond_3ad_initiate_agg_selection(bond, 1);
3208         }
3209
3210         if (bond_mode_uses_xmit_hash(bond))
3211                 bond_update_slave_arr(bond, NULL);
3212
3213         return 0;
3214 }
3215
3216 static int bond_close(struct net_device *bond_dev)
3217 {
3218         struct bonding *bond = netdev_priv(bond_dev);
3219
3220         bond_work_cancel_all(bond);
3221         bond->send_peer_notif = 0;
3222         if (bond_is_lb(bond))
3223                 bond_alb_deinitialize(bond);
3224         bond->recv_probe = NULL;
3225
3226         return 0;
3227 }
3228
3229 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3230                                                 struct rtnl_link_stats64 *stats)
3231 {
3232         struct bonding *bond = netdev_priv(bond_dev);
3233         struct rtnl_link_stats64 temp;
3234         struct list_head *iter;
3235         struct slave *slave;
3236
3237         memcpy(stats, &bond->bond_stats, sizeof(*stats));
3238
3239         bond_for_each_slave(bond, slave, iter) {
3240                 const struct rtnl_link_stats64 *sstats =
3241                         dev_get_stats(slave->dev, &temp);
3242                 struct rtnl_link_stats64 *pstats = &slave->slave_stats;
3243
3244                 stats->rx_packets +=  sstats->rx_packets - pstats->rx_packets;
3245                 stats->rx_bytes += sstats->rx_bytes - pstats->rx_bytes;
3246                 stats->rx_errors += sstats->rx_errors - pstats->rx_errors;
3247                 stats->rx_dropped += sstats->rx_dropped - pstats->rx_dropped;
3248
3249                 stats->tx_packets += sstats->tx_packets - pstats->tx_packets;;
3250                 stats->tx_bytes += sstats->tx_bytes - pstats->tx_bytes;
3251                 stats->tx_errors += sstats->tx_errors - pstats->tx_errors;
3252                 stats->tx_dropped += sstats->tx_dropped - pstats->tx_dropped;
3253
3254                 stats->multicast += sstats->multicast - pstats->multicast;
3255                 stats->collisions += sstats->collisions - pstats->collisions;
3256
3257                 stats->rx_length_errors += sstats->rx_length_errors - pstats->rx_length_errors;
3258                 stats->rx_over_errors += sstats->rx_over_errors - pstats->rx_over_errors;
3259                 stats->rx_crc_errors += sstats->rx_crc_errors - pstats->rx_crc_errors;
3260                 stats->rx_frame_errors += sstats->rx_frame_errors - pstats->rx_frame_errors;
3261                 stats->rx_fifo_errors += sstats->rx_fifo_errors - pstats->rx_fifo_errors;
3262                 stats->rx_missed_errors += sstats->rx_missed_errors - pstats->rx_missed_errors;
3263
3264                 stats->tx_aborted_errors += sstats->tx_aborted_errors - pstats->tx_aborted_errors;
3265                 stats->tx_carrier_errors += sstats->tx_carrier_errors - pstats->tx_carrier_errors;
3266                 stats->tx_fifo_errors += sstats->tx_fifo_errors - pstats->tx_fifo_errors;
3267                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors - pstats->tx_heartbeat_errors;
3268                 stats->tx_window_errors += sstats->tx_window_errors - pstats->tx_window_errors;
3269
3270                 /* save off the slave stats for the next run */
3271                 memcpy(pstats, sstats, sizeof(*sstats));
3272         }
3273         memcpy(&bond->bond_stats, stats, sizeof(*stats));
3274
3275         return stats;
3276 }
3277
3278 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3279 {
3280         struct bonding *bond = netdev_priv(bond_dev);
3281         struct net_device *slave_dev = NULL;
3282         struct ifbond k_binfo;
3283         struct ifbond __user *u_binfo = NULL;
3284         struct ifslave k_sinfo;
3285         struct ifslave __user *u_sinfo = NULL;
3286         struct mii_ioctl_data *mii = NULL;
3287         struct bond_opt_value newval;
3288         struct net *net;
3289         int res = 0;
3290
3291         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3292
3293         switch (cmd) {
3294         case SIOCGMIIPHY:
3295                 mii = if_mii(ifr);
3296                 if (!mii)
3297                         return -EINVAL;
3298
3299                 mii->phy_id = 0;
3300                 /* Fall Through */
3301         case SIOCGMIIREG:
3302                 /* We do this again just in case we were called by SIOCGMIIREG
3303                  * instead of SIOCGMIIPHY.
3304                  */
3305                 mii = if_mii(ifr);
3306                 if (!mii)
3307                         return -EINVAL;
3308
3309                 if (mii->reg_num == 1) {
3310                         mii->val_out = 0;
3311                         if (netif_carrier_ok(bond->dev))
3312                                 mii->val_out = BMSR_LSTATUS;
3313                 }
3314
3315                 return 0;
3316         case BOND_INFO_QUERY_OLD:
3317         case SIOCBONDINFOQUERY:
3318                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3319
3320                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3321                         return -EFAULT;
3322
3323                 res = bond_info_query(bond_dev, &k_binfo);
3324                 if (res == 0 &&
3325                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3326                         return -EFAULT;
3327
3328                 return res;
3329         case BOND_SLAVE_INFO_QUERY_OLD:
3330         case SIOCBONDSLAVEINFOQUERY:
3331                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3332
3333                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3334                         return -EFAULT;
3335
3336                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3337                 if (res == 0 &&
3338                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3339                         return -EFAULT;
3340
3341                 return res;
3342         default:
3343                 break;
3344         }
3345
3346         net = dev_net(bond_dev);
3347
3348         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3349                 return -EPERM;
3350
3351         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3352
3353         netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3354
3355         if (!slave_dev)
3356                 return -ENODEV;
3357
3358         netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3359         switch (cmd) {
3360         case BOND_ENSLAVE_OLD:
3361         case SIOCBONDENSLAVE:
3362                 res = bond_enslave(bond_dev, slave_dev);
3363                 break;
3364         case BOND_RELEASE_OLD:
3365         case SIOCBONDRELEASE:
3366                 res = bond_release(bond_dev, slave_dev);
3367                 break;
3368         case BOND_SETHWADDR_OLD:
3369         case SIOCBONDSETHWADDR:
3370                 bond_set_dev_addr(bond_dev, slave_dev);
3371                 res = 0;
3372                 break;
3373         case BOND_CHANGE_ACTIVE_OLD:
3374         case SIOCBONDCHANGEACTIVE:
3375                 bond_opt_initstr(&newval, slave_dev->name);
3376                 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3377                 break;
3378         default:
3379                 res = -EOPNOTSUPP;
3380         }
3381
3382         return res;
3383 }
3384
3385 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3386 {
3387         struct bonding *bond = netdev_priv(bond_dev);
3388
3389         if (change & IFF_PROMISC)
3390                 bond_set_promiscuity(bond,
3391                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3392
3393         if (change & IFF_ALLMULTI)
3394                 bond_set_allmulti(bond,
3395                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3396 }
3397
3398 static void bond_set_rx_mode(struct net_device *bond_dev)
3399 {
3400         struct bonding *bond = netdev_priv(bond_dev);
3401         struct list_head *iter;
3402         struct slave *slave;
3403
3404         rcu_read_lock();
3405         if (bond_uses_primary(bond)) {
3406                 slave = rcu_dereference(bond->curr_active_slave);
3407                 if (slave) {
3408                         dev_uc_sync(slave->dev, bond_dev);
3409                         dev_mc_sync(slave->dev, bond_dev);
3410                 }
3411         } else {
3412                 bond_for_each_slave_rcu(bond, slave, iter) {
3413                         dev_uc_sync_multiple(slave->dev, bond_dev);
3414                         dev_mc_sync_multiple(slave->dev, bond_dev);
3415                 }
3416         }
3417         rcu_read_unlock();
3418 }
3419
3420 static int bond_neigh_init(struct neighbour *n)
3421 {
3422         struct bonding *bond = netdev_priv(n->dev);
3423         const struct net_device_ops *slave_ops;
3424         struct neigh_parms parms;
3425         struct slave *slave;
3426         int ret;
3427
3428         slave = bond_first_slave(bond);
3429         if (!slave)
3430                 return 0;
3431         slave_ops = slave->dev->netdev_ops;
3432         if (!slave_ops->ndo_neigh_setup)
3433                 return 0;
3434
3435         parms.neigh_setup = NULL;
3436         parms.neigh_cleanup = NULL;
3437         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3438         if (ret)
3439                 return ret;
3440
3441         /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3442          * after the last slave has been detached.  Assumes that all slaves
3443          * utilize the same neigh_cleanup (true at this writing as only user
3444          * is ipoib).
3445          */
3446         n->parms->neigh_cleanup = parms.neigh_cleanup;
3447
3448         if (!parms.neigh_setup)
3449                 return 0;
3450
3451         return parms.neigh_setup(n);
3452 }
3453
3454 /* The bonding ndo_neigh_setup is called at init time beofre any
3455  * slave exists. So we must declare proxy setup function which will
3456  * be used at run time to resolve the actual slave neigh param setup.
3457  *
3458  * It's also called by master devices (such as vlans) to setup their
3459  * underlying devices. In that case - do nothing, we're already set up from
3460  * our init.
3461  */
3462 static int bond_neigh_setup(struct net_device *dev,
3463                             struct neigh_parms *parms)
3464 {
3465         /* modify only our neigh_parms */
3466         if (parms->dev == dev)
3467                 parms->neigh_setup = bond_neigh_init;
3468
3469         return 0;
3470 }
3471
3472 /* Change the MTU of all of a master's slaves to match the master */
3473 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3474 {
3475         struct bonding *bond = netdev_priv(bond_dev);
3476         struct slave *slave, *rollback_slave;
3477         struct list_head *iter;
3478         int res = 0;
3479
3480         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3481
3482         bond_for_each_slave(bond, slave, iter) {
3483                 netdev_dbg(bond_dev, "s %p c_m %p\n",
3484                            slave, slave->dev->netdev_ops->ndo_change_mtu);
3485
3486                 res = dev_set_mtu(slave->dev, new_mtu);
3487
3488                 if (res) {
3489                         /* If we failed to set the slave's mtu to the new value
3490                          * we must abort the operation even in ACTIVE_BACKUP
3491                          * mode, because if we allow the backup slaves to have
3492                          * different mtu values than the active slave we'll
3493                          * need to change their mtu when doing a failover. That
3494                          * means changing their mtu from timer context, which
3495                          * is probably not a good idea.
3496                          */
3497                         netdev_dbg(bond_dev, "err %d %s\n", res,
3498                                    slave->dev->name);
3499                         goto unwind;
3500                 }
3501         }
3502
3503         bond_dev->mtu = new_mtu;
3504
3505         return 0;
3506
3507 unwind:
3508         /* unwind from head to the slave that failed */
3509         bond_for_each_slave(bond, rollback_slave, iter) {
3510                 int tmp_res;
3511
3512                 if (rollback_slave == slave)
3513                         break;
3514
3515                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3516                 if (tmp_res) {
3517                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3518                                    tmp_res, rollback_slave->dev->name);
3519                 }
3520         }
3521
3522         return res;
3523 }
3524
3525 /* Change HW address
3526  *
3527  * Note that many devices must be down to change the HW address, and
3528  * downing the master releases all slaves.  We can make bonds full of
3529  * bonding devices to test this, however.
3530  */
3531 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3532 {
3533         struct bonding *bond = netdev_priv(bond_dev);
3534         struct slave *slave, *rollback_slave;
3535         struct sockaddr *sa = addr, tmp_sa;
3536         struct list_head *iter;
3537         int res = 0;
3538
3539         if (BOND_MODE(bond) == BOND_MODE_ALB)
3540                 return bond_alb_set_mac_address(bond_dev, addr);
3541
3542
3543         netdev_dbg(bond_dev, "bond=%p\n", bond);
3544
3545         /* If fail_over_mac is enabled, do nothing and return success.
3546          * Returning an error causes ifenslave to fail.
3547          */
3548         if (bond->params.fail_over_mac &&
3549             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3550                 return 0;
3551
3552         if (!is_valid_ether_addr(sa->sa_data))
3553                 return -EADDRNOTAVAIL;
3554
3555         bond_for_each_slave(bond, slave, iter) {
3556                 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3557                 res = dev_set_mac_address(slave->dev, addr);
3558                 if (res) {
3559                         /* TODO: consider downing the slave
3560                          * and retry ?
3561                          * User should expect communications
3562                          * breakage anyway until ARP finish
3563                          * updating, so...
3564                          */
3565                         netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3566                         goto unwind;
3567                 }
3568         }
3569
3570         /* success */
3571         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3572         return 0;
3573
3574 unwind:
3575         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3576         tmp_sa.sa_family = bond_dev->type;
3577
3578         /* unwind from head to the slave that failed */
3579         bond_for_each_slave(bond, rollback_slave, iter) {
3580                 int tmp_res;
3581
3582                 if (rollback_slave == slave)
3583                         break;
3584
3585                 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3586                 if (tmp_res) {
3587                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3588                                    tmp_res, rollback_slave->dev->name);
3589                 }
3590         }
3591
3592         return res;
3593 }
3594
3595 /**
3596  * bond_xmit_slave_id - transmit skb through slave with slave_id
3597  * @bond: bonding device that is transmitting
3598  * @skb: buffer to transmit
3599  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3600  *
3601  * This function tries to transmit through slave with slave_id but in case
3602  * it fails, it tries to find the first available slave for transmission.
3603  * The skb is consumed in all cases, thus the function is void.
3604  */
3605 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3606 {
3607         struct list_head *iter;
3608         struct slave *slave;
3609         int i = slave_id;
3610
3611         /* Here we start from the slave with slave_id */
3612         bond_for_each_slave_rcu(bond, slave, iter) {
3613                 if (--i < 0) {
3614                         if (bond_slave_can_tx(slave)) {
3615                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3616                                 return;
3617                         }
3618                 }
3619         }
3620
3621         /* Here we start from the first slave up to slave_id */
3622         i = slave_id;
3623         bond_for_each_slave_rcu(bond, slave, iter) {
3624                 if (--i < 0)
3625                         break;
3626                 if (bond_slave_can_tx(slave)) {
3627                         bond_dev_queue_xmit(bond, skb, slave->dev);
3628                         return;
3629                 }
3630         }
3631         /* no slave that can tx has been found */
3632         bond_tx_drop(bond->dev, skb);
3633 }
3634
3635 /**
3636  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3637  * @bond: bonding device to use
3638  *
3639  * Based on the value of the bonding device's packets_per_slave parameter
3640  * this function generates a slave id, which is usually used as the next
3641  * slave to transmit through.
3642  */
3643 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3644 {
3645         u32 slave_id;
3646         struct reciprocal_value reciprocal_packets_per_slave;
3647         int packets_per_slave = bond->params.packets_per_slave;
3648
3649         switch (packets_per_slave) {
3650         case 0:
3651                 slave_id = prandom_u32();
3652                 break;
3653         case 1:
3654                 slave_id = bond->rr_tx_counter;
3655                 break;
3656         default:
3657                 reciprocal_packets_per_slave =
3658                         bond->params.reciprocal_packets_per_slave;
3659                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3660                                              reciprocal_packets_per_slave);
3661                 break;
3662         }
3663         bond->rr_tx_counter++;
3664
3665         return slave_id;
3666 }
3667
3668 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3669 {
3670         struct bonding *bond = netdev_priv(bond_dev);
3671         struct iphdr *iph = ip_hdr(skb);
3672         struct slave *slave;
3673         u32 slave_id;
3674
3675         /* Start with the curr_active_slave that joined the bond as the
3676          * default for sending IGMP traffic.  For failover purposes one
3677          * needs to maintain some consistency for the interface that will
3678          * send the join/membership reports.  The curr_active_slave found
3679          * will send all of this type of traffic.
3680          */
3681         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3682                 slave = rcu_dereference(bond->curr_active_slave);
3683                 if (slave)
3684                         bond_dev_queue_xmit(bond, skb, slave->dev);
3685                 else
3686                         bond_xmit_slave_id(bond, skb, 0);
3687         } else {
3688                 int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3689
3690                 if (likely(slave_cnt)) {
3691                         slave_id = bond_rr_gen_slave_id(bond);
3692                         bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3693                 } else {
3694                         bond_tx_drop(bond_dev, skb);
3695                 }
3696         }
3697
3698         return NETDEV_TX_OK;
3699 }
3700
3701 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3702  * the bond has a usable interface.
3703  */
3704 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3705 {
3706         struct bonding *bond = netdev_priv(bond_dev);
3707         struct slave *slave;
3708
3709         slave = rcu_dereference(bond->curr_active_slave);
3710         if (slave)
3711                 bond_dev_queue_xmit(bond, skb, slave->dev);
3712         else
3713                 bond_tx_drop(bond_dev, skb);
3714
3715         return NETDEV_TX_OK;
3716 }
3717
3718 /* Use this to update slave_array when (a) it's not appropriate to update
3719  * slave_array right away (note that update_slave_array() may sleep)
3720  * and / or (b) RTNL is not held.
3721  */
3722 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3723 {
3724         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3725 }
3726
3727 /* Slave array work handler. Holds only RTNL */
3728 static void bond_slave_arr_handler(struct work_struct *work)
3729 {
3730         struct bonding *bond = container_of(work, struct bonding,
3731                                             slave_arr_work.work);
3732         int ret;
3733
3734         if (!rtnl_trylock())
3735                 goto err;
3736
3737         ret = bond_update_slave_arr(bond, NULL);
3738         rtnl_unlock();
3739         if (ret) {
3740                 pr_warn_ratelimited("Failed to update slave array from WT\n");
3741                 goto err;
3742         }
3743         return;
3744
3745 err:
3746         bond_slave_arr_work_rearm(bond, 1);
3747 }
3748
3749 /* Build the usable slaves array in control path for modes that use xmit-hash
3750  * to determine the slave interface -
3751  * (a) BOND_MODE_8023AD
3752  * (b) BOND_MODE_XOR
3753  * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3754  *
3755  * The caller is expected to hold RTNL only and NO other lock!
3756  */
3757 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3758 {
3759         struct slave *slave;
3760         struct list_head *iter;
3761         struct bond_up_slave *new_arr, *old_arr;
3762         int slaves_in_agg;
3763         int agg_id = 0;
3764         int ret = 0;
3765
3766 #ifdef CONFIG_LOCKDEP
3767         WARN_ON(lockdep_is_held(&bond->mode_lock));
3768 #endif
3769
3770         new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3771                           GFP_KERNEL);
3772         if (!new_arr) {
3773                 ret = -ENOMEM;
3774                 pr_err("Failed to build slave-array.\n");
3775                 goto out;
3776         }
3777         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3778                 struct ad_info ad_info;
3779
3780                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3781                         pr_debug("bond_3ad_get_active_agg_info failed\n");
3782                         kfree_rcu(new_arr, rcu);
3783                         /* No active aggragator means it's not safe to use
3784                          * the previous array.
3785                          */
3786                         old_arr = rtnl_dereference(bond->slave_arr);
3787                         if (old_arr) {
3788                                 RCU_INIT_POINTER(bond->slave_arr, NULL);
3789                                 kfree_rcu(old_arr, rcu);
3790                         }
3791                         goto out;
3792                 }
3793                 slaves_in_agg = ad_info.ports;
3794                 agg_id = ad_info.aggregator_id;
3795         }
3796         bond_for_each_slave(bond, slave, iter) {
3797                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3798                         struct aggregator *agg;
3799
3800                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
3801                         if (!agg || agg->aggregator_identifier != agg_id)
3802                                 continue;
3803                 }
3804                 if (!bond_slave_can_tx(slave))
3805                         continue;
3806                 if (skipslave == slave)
3807                         continue;
3808                 new_arr->arr[new_arr->count++] = slave;
3809         }
3810
3811         old_arr = rtnl_dereference(bond->slave_arr);
3812         rcu_assign_pointer(bond->slave_arr, new_arr);
3813         if (old_arr)
3814                 kfree_rcu(old_arr, rcu);
3815 out:
3816         if (ret != 0 && skipslave) {
3817                 int idx;
3818
3819                 /* Rare situation where caller has asked to skip a specific
3820                  * slave but allocation failed (most likely!). BTW this is
3821                  * only possible when the call is initiated from
3822                  * __bond_release_one(). In this situation; overwrite the
3823                  * skipslave entry in the array with the last entry from the
3824                  * array to avoid a situation where the xmit path may choose
3825                  * this to-be-skipped slave to send a packet out.
3826                  */
3827                 old_arr = rtnl_dereference(bond->slave_arr);
3828                 for (idx = 0; idx < old_arr->count; idx++) {
3829                         if (skipslave == old_arr->arr[idx]) {
3830                                 old_arr->arr[idx] =
3831                                     old_arr->arr[old_arr->count-1];
3832                                 old_arr->count--;
3833                                 break;
3834                         }
3835                 }
3836         }
3837         return ret;
3838 }
3839
3840 /* Use this Xmit function for 3AD as well as XOR modes. The current
3841  * usable slave array is formed in the control path. The xmit function
3842  * just calculates hash and sends the packet out.
3843  */
3844 static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3845 {
3846         struct bonding *bond = netdev_priv(dev);
3847         struct slave *slave;
3848         struct bond_up_slave *slaves;
3849         unsigned int count;
3850
3851         slaves = rcu_dereference(bond->slave_arr);
3852         count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3853         if (likely(count)) {
3854                 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3855                 bond_dev_queue_xmit(bond, skb, slave->dev);
3856         } else {
3857                 bond_tx_drop(dev, skb);
3858         }
3859
3860         return NETDEV_TX_OK;
3861 }
3862
3863 /* in broadcast mode, we send everything to all usable interfaces. */
3864 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3865 {
3866         struct bonding *bond = netdev_priv(bond_dev);
3867         struct slave *slave = NULL;
3868         struct list_head *iter;
3869
3870         bond_for_each_slave_rcu(bond, slave, iter) {
3871                 if (bond_is_last_slave(bond, slave))
3872                         break;
3873                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3874                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3875
3876                         if (!skb2) {
3877                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3878                                                     bond_dev->name, __func__);
3879                                 continue;
3880                         }
3881                         bond_dev_queue_xmit(bond, skb2, slave->dev);
3882                 }
3883         }
3884         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3885                 bond_dev_queue_xmit(bond, skb, slave->dev);
3886         else
3887                 bond_tx_drop(bond_dev, skb);
3888
3889         return NETDEV_TX_OK;
3890 }
3891
3892 /*------------------------- Device initialization ---------------------------*/
3893
3894 /* Lookup the slave that corresponds to a qid */
3895 static inline int bond_slave_override(struct bonding *bond,
3896                                       struct sk_buff *skb)
3897 {
3898         struct slave *slave = NULL;
3899         struct list_head *iter;
3900
3901         if (!skb->queue_mapping)
3902                 return 1;
3903
3904         /* Find out if any slaves have the same mapping as this skb. */
3905         bond_for_each_slave_rcu(bond, slave, iter) {
3906                 if (slave->queue_id == skb->queue_mapping) {
3907                         if (bond_slave_is_up(slave) &&
3908                             slave->link == BOND_LINK_UP) {
3909                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3910                                 return 0;
3911                         }
3912                         /* If the slave isn't UP, use default transmit policy. */
3913                         break;
3914                 }
3915         }
3916
3917         return 1;
3918 }
3919
3920
3921 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3922                              void *accel_priv, select_queue_fallback_t fallback)
3923 {
3924         /* This helper function exists to help dev_pick_tx get the correct
3925          * destination queue.  Using a helper function skips a call to
3926          * skb_tx_hash and will put the skbs in the queue we expect on their
3927          * way down to the bonding driver.
3928          */
3929         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3930
3931         /* Save the original txq to restore before passing to the driver */
3932         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3933
3934         if (unlikely(txq >= dev->real_num_tx_queues)) {
3935                 do {
3936                         txq -= dev->real_num_tx_queues;
3937                 } while (txq >= dev->real_num_tx_queues);
3938         }
3939         return txq;
3940 }
3941
3942 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3943 {
3944         struct bonding *bond = netdev_priv(dev);
3945
3946         if (bond_should_override_tx_queue(bond) &&
3947             !bond_slave_override(bond, skb))
3948                 return NETDEV_TX_OK;
3949
3950         switch (BOND_MODE(bond)) {
3951         case BOND_MODE_ROUNDROBIN:
3952                 return bond_xmit_roundrobin(skb, dev);
3953         case BOND_MODE_ACTIVEBACKUP:
3954                 return bond_xmit_activebackup(skb, dev);
3955         case BOND_MODE_8023AD:
3956         case BOND_MODE_XOR:
3957                 return bond_3ad_xor_xmit(skb, dev);
3958         case BOND_MODE_BROADCAST:
3959                 return bond_xmit_broadcast(skb, dev);
3960         case BOND_MODE_ALB:
3961                 return bond_alb_xmit(skb, dev);
3962         case BOND_MODE_TLB:
3963                 return bond_tlb_xmit(skb, dev);
3964         default:
3965                 /* Should never happen, mode already checked */
3966                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
3967                 WARN_ON_ONCE(1);
3968                 bond_tx_drop(dev, skb);
3969                 return NETDEV_TX_OK;
3970         }
3971 }
3972
3973 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3974 {
3975         struct bonding *bond = netdev_priv(dev);
3976         netdev_tx_t ret = NETDEV_TX_OK;
3977
3978         /* If we risk deadlock from transmitting this in the
3979          * netpoll path, tell netpoll to queue the frame for later tx
3980          */
3981         if (unlikely(is_netpoll_tx_blocked(dev)))
3982                 return NETDEV_TX_BUSY;
3983
3984         rcu_read_lock();
3985         if (bond_has_slaves(bond))
3986                 ret = __bond_start_xmit(skb, dev);
3987         else
3988                 bond_tx_drop(dev, skb);
3989         rcu_read_unlock();
3990
3991         return ret;
3992 }
3993
3994 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3995                                      struct ethtool_cmd *ecmd)
3996 {
3997         struct bonding *bond = netdev_priv(bond_dev);
3998         unsigned long speed = 0;
3999         struct list_head *iter;
4000         struct slave *slave;
4001
4002         ecmd->duplex = DUPLEX_UNKNOWN;
4003         ecmd->port = PORT_OTHER;
4004
4005         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4006          * do not need to check mode.  Though link speed might not represent
4007          * the true receive or transmit bandwidth (not all modes are symmetric)
4008          * this is an accurate maximum.
4009          */
4010         bond_for_each_slave(bond, slave, iter) {
4011                 if (bond_slave_can_tx(slave)) {
4012                         if (slave->speed != SPEED_UNKNOWN)
4013                                 speed += slave->speed;
4014                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
4015                             slave->duplex != DUPLEX_UNKNOWN)
4016                                 ecmd->duplex = slave->duplex;
4017                 }
4018         }
4019         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
4020
4021         return 0;
4022 }
4023
4024 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4025                                      struct ethtool_drvinfo *drvinfo)
4026 {
4027         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4028         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4029         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4030                  BOND_ABI_VERSION);
4031 }
4032
4033 static const struct ethtool_ops bond_ethtool_ops = {
4034         .get_drvinfo            = bond_ethtool_get_drvinfo,
4035         .get_settings           = bond_ethtool_get_settings,
4036         .get_link               = ethtool_op_get_link,
4037 };
4038
4039 static const struct net_device_ops bond_netdev_ops = {
4040         .ndo_init               = bond_init,
4041         .ndo_uninit             = bond_uninit,
4042         .ndo_open               = bond_open,
4043         .ndo_stop               = bond_close,
4044         .ndo_start_xmit         = bond_start_xmit,
4045         .ndo_select_queue       = bond_select_queue,
4046         .ndo_get_stats64        = bond_get_stats,
4047         .ndo_do_ioctl           = bond_do_ioctl,
4048         .ndo_change_rx_flags    = bond_change_rx_flags,
4049         .ndo_set_rx_mode        = bond_set_rx_mode,
4050         .ndo_change_mtu         = bond_change_mtu,
4051         .ndo_set_mac_address    = bond_set_mac_address,
4052         .ndo_neigh_setup        = bond_neigh_setup,
4053         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4054         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4055 #ifdef CONFIG_NET_POLL_CONTROLLER
4056         .ndo_netpoll_setup      = bond_netpoll_setup,
4057         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4058         .ndo_poll_controller    = bond_poll_controller,
4059 #endif
4060         .ndo_add_slave          = bond_enslave,
4061         .ndo_del_slave          = bond_release,
4062         .ndo_fix_features       = bond_fix_features,
4063         .ndo_bridge_setlink     = ndo_dflt_netdev_switch_port_bridge_setlink,
4064         .ndo_bridge_dellink     = ndo_dflt_netdev_switch_port_bridge_dellink,
4065         .ndo_features_check     = passthru_features_check,
4066 };
4067
4068 static const struct device_type bond_type = {
4069         .name = "bond",
4070 };
4071
4072 static void bond_destructor(struct net_device *bond_dev)
4073 {
4074         struct bonding *bond = netdev_priv(bond_dev);
4075         if (bond->wq)
4076                 destroy_workqueue(bond->wq);
4077         free_netdev(bond_dev);
4078 }
4079
4080 void bond_setup(struct net_device *bond_dev)
4081 {
4082         struct bonding *bond = netdev_priv(bond_dev);
4083
4084         spin_lock_init(&bond->mode_lock);
4085         bond->params = bonding_defaults;
4086
4087         /* Initialize pointers */
4088         bond->dev = bond_dev;
4089
4090         /* Initialize the device entry points */
4091         ether_setup(bond_dev);
4092         bond_dev->netdev_ops = &bond_netdev_ops;
4093         bond_dev->ethtool_ops = &bond_ethtool_ops;
4094
4095         bond_dev->destructor = bond_destructor;
4096
4097         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4098
4099         /* Initialize the device options */
4100         bond_dev->tx_queue_len = 0;
4101         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4102         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT;
4103         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4104
4105         /* don't acquire bond device's netif_tx_lock when transmitting */
4106         bond_dev->features |= NETIF_F_LLTX;
4107
4108         /* By default, we declare the bond to be fully
4109          * VLAN hardware accelerated capable. Special
4110          * care is taken in the various xmit functions
4111          * when there are slaves that are not hw accel
4112          * capable
4113          */
4114
4115         /* Don't allow bond devices to change network namespaces. */
4116         bond_dev->features |= NETIF_F_NETNS_LOCAL;
4117
4118         bond_dev->hw_features = BOND_VLAN_FEATURES |
4119                                 NETIF_F_HW_VLAN_CTAG_TX |
4120                                 NETIF_F_HW_VLAN_CTAG_RX |
4121                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4122
4123         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4124         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4125         bond_dev->features |= bond_dev->hw_features;
4126 }
4127
4128 /* Destroy a bonding device.
4129  * Must be under rtnl_lock when this function is called.
4130  */
4131 static void bond_uninit(struct net_device *bond_dev)
4132 {
4133         struct bonding *bond = netdev_priv(bond_dev);
4134         struct list_head *iter;
4135         struct slave *slave;
4136         struct bond_up_slave *arr;
4137
4138         bond_netpoll_cleanup(bond_dev);
4139
4140         /* Release the bonded slaves */
4141         bond_for_each_slave(bond, slave, iter)
4142                 __bond_release_one(bond_dev, slave->dev, true);
4143         netdev_info(bond_dev, "Released all slaves\n");
4144
4145         arr = rtnl_dereference(bond->slave_arr);
4146         if (arr) {
4147                 RCU_INIT_POINTER(bond->slave_arr, NULL);
4148                 kfree_rcu(arr, rcu);
4149         }
4150
4151         list_del(&bond->bond_list);
4152
4153         bond_debug_unregister(bond);
4154 }
4155
4156 /*------------------------- Module initialization ---------------------------*/
4157
4158 static int bond_check_params(struct bond_params *params)
4159 {
4160         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4161         struct bond_opt_value newval;
4162         const struct bond_opt_value *valptr;
4163         int arp_all_targets_value;
4164
4165         /* Convert string parameters. */
4166         if (mode) {
4167                 bond_opt_initstr(&newval, mode);
4168                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4169                 if (!valptr) {
4170                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4171                         return -EINVAL;
4172                 }
4173                 bond_mode = valptr->value;
4174         }
4175
4176         if (xmit_hash_policy) {
4177                 if ((bond_mode != BOND_MODE_XOR) &&
4178                     (bond_mode != BOND_MODE_8023AD) &&
4179                     (bond_mode != BOND_MODE_TLB)) {
4180                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4181                                 bond_mode_name(bond_mode));
4182                 } else {
4183                         bond_opt_initstr(&newval, xmit_hash_policy);
4184                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4185                                                 &newval);
4186                         if (!valptr) {
4187                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4188                                        xmit_hash_policy);
4189                                 return -EINVAL;
4190                         }
4191                         xmit_hashtype = valptr->value;
4192                 }
4193         }
4194
4195         if (lacp_rate) {
4196                 if (bond_mode != BOND_MODE_8023AD) {
4197                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4198                                 bond_mode_name(bond_mode));
4199                 } else {
4200                         bond_opt_initstr(&newval, lacp_rate);
4201                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4202                                                 &newval);
4203                         if (!valptr) {
4204                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4205                                        lacp_rate);
4206                                 return -EINVAL;
4207                         }
4208                         lacp_fast = valptr->value;
4209                 }
4210         }
4211
4212         if (ad_select) {
4213                 bond_opt_initstr(&newval, ad_select);
4214                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4215                                         &newval);
4216                 if (!valptr) {
4217                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4218                         return -EINVAL;
4219                 }
4220                 params->ad_select = valptr->value;
4221                 if (bond_mode != BOND_MODE_8023AD)
4222                         pr_warn("ad_select param only affects 802.3ad mode\n");
4223         } else {
4224                 params->ad_select = BOND_AD_STABLE;
4225         }
4226
4227         if (max_bonds < 0) {
4228                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4229                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4230                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4231         }
4232
4233         if (miimon < 0) {
4234                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4235                         miimon, INT_MAX);
4236                 miimon = 0;
4237         }
4238
4239         if (updelay < 0) {
4240                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4241                         updelay, INT_MAX);
4242                 updelay = 0;
4243         }
4244
4245         if (downdelay < 0) {
4246                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4247                         downdelay, INT_MAX);
4248                 downdelay = 0;
4249         }
4250
4251         if ((use_carrier != 0) && (use_carrier != 1)) {
4252                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4253                         use_carrier);
4254                 use_carrier = 1;
4255         }
4256
4257         if (num_peer_notif < 0 || num_peer_notif > 255) {
4258                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4259                         num_peer_notif);
4260                 num_peer_notif = 1;
4261         }
4262
4263         /* reset values for 802.3ad/TLB/ALB */
4264         if (!bond_mode_uses_arp(bond_mode)) {
4265                 if (!miimon) {
4266                         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");
4267                         pr_warn("Forcing miimon to 100msec\n");
4268                         miimon = BOND_DEFAULT_MIIMON;
4269                 }
4270         }
4271
4272         if (tx_queues < 1 || tx_queues > 255) {
4273                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4274                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4275                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4276         }
4277
4278         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4279                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4280                         all_slaves_active);
4281                 all_slaves_active = 0;
4282         }
4283
4284         if (resend_igmp < 0 || resend_igmp > 255) {
4285                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4286                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4287                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4288         }
4289
4290         bond_opt_initval(&newval, packets_per_slave);
4291         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4292                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4293                         packets_per_slave, USHRT_MAX);
4294                 packets_per_slave = 1;
4295         }
4296
4297         if (bond_mode == BOND_MODE_ALB) {
4298                 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",
4299                           updelay);
4300         }
4301
4302         if (!miimon) {
4303                 if (updelay || downdelay) {
4304                         /* just warn the user the up/down delay will have
4305                          * no effect since miimon is zero...
4306                          */
4307                         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",
4308                                 updelay, downdelay);
4309                 }
4310         } else {
4311                 /* don't allow arp monitoring */
4312                 if (arp_interval) {
4313                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4314                                 miimon, arp_interval);
4315                         arp_interval = 0;
4316                 }
4317
4318                 if ((updelay % miimon) != 0) {
4319                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4320                                 updelay, miimon, (updelay / miimon) * miimon);
4321                 }
4322
4323                 updelay /= miimon;
4324
4325                 if ((downdelay % miimon) != 0) {
4326                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4327                                 downdelay, miimon,
4328                                 (downdelay / miimon) * miimon);
4329                 }
4330
4331                 downdelay /= miimon;
4332         }
4333
4334         if (arp_interval < 0) {
4335                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4336                         arp_interval, INT_MAX);
4337                 arp_interval = 0;
4338         }
4339
4340         for (arp_ip_count = 0, i = 0;
4341              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4342                 __be32 ip;
4343
4344                 /* not a complete check, but good enough to catch mistakes */
4345                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4346                     !bond_is_ip_target_ok(ip)) {
4347                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4348                                 arp_ip_target[i]);
4349                         arp_interval = 0;
4350                 } else {
4351                         if (bond_get_targets_ip(arp_target, ip) == -1)
4352                                 arp_target[arp_ip_count++] = ip;
4353                         else
4354                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4355                                         &ip);
4356                 }
4357         }
4358
4359         if (arp_interval && !arp_ip_count) {
4360                 /* don't allow arping if no arp_ip_target given... */
4361                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4362                         arp_interval);
4363                 arp_interval = 0;
4364         }
4365
4366         if (arp_validate) {
4367                 if (!arp_interval) {
4368                         pr_err("arp_validate requires arp_interval\n");
4369                         return -EINVAL;
4370                 }
4371
4372                 bond_opt_initstr(&newval, arp_validate);
4373                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4374                                         &newval);
4375                 if (!valptr) {
4376                         pr_err("Error: invalid arp_validate \"%s\"\n",
4377                                arp_validate);
4378                         return -EINVAL;
4379                 }
4380                 arp_validate_value = valptr->value;
4381         } else {
4382                 arp_validate_value = 0;
4383         }
4384
4385         arp_all_targets_value = 0;
4386         if (arp_all_targets) {
4387                 bond_opt_initstr(&newval, arp_all_targets);
4388                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4389                                         &newval);
4390                 if (!valptr) {
4391                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4392                                arp_all_targets);
4393                         arp_all_targets_value = 0;
4394                 } else {
4395                         arp_all_targets_value = valptr->value;
4396                 }
4397         }
4398
4399         if (miimon) {
4400                 pr_info("MII link monitoring set to %d ms\n", miimon);
4401         } else if (arp_interval) {
4402                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4403                                           arp_validate_value);
4404                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4405                         arp_interval, valptr->string, arp_ip_count);
4406
4407                 for (i = 0; i < arp_ip_count; i++)
4408                         pr_cont(" %s", arp_ip_target[i]);
4409
4410                 pr_cont("\n");
4411
4412         } else if (max_bonds) {
4413                 /* miimon and arp_interval not set, we need one so things
4414                  * work as expected, see bonding.txt for details
4415                  */
4416                 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");
4417         }
4418
4419         if (primary && !bond_mode_uses_primary(bond_mode)) {
4420                 /* currently, using a primary only makes sense
4421                  * in active backup, TLB or ALB modes
4422                  */
4423                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4424                         primary, bond_mode_name(bond_mode));
4425                 primary = NULL;
4426         }
4427
4428         if (primary && primary_reselect) {
4429                 bond_opt_initstr(&newval, primary_reselect);
4430                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4431                                         &newval);
4432                 if (!valptr) {
4433                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4434                                primary_reselect);
4435                         return -EINVAL;
4436                 }
4437                 primary_reselect_value = valptr->value;
4438         } else {
4439                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4440         }
4441
4442         if (fail_over_mac) {
4443                 bond_opt_initstr(&newval, fail_over_mac);
4444                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4445                                         &newval);
4446                 if (!valptr) {
4447                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4448                                fail_over_mac);
4449                         return -EINVAL;
4450                 }
4451                 fail_over_mac_value = valptr->value;
4452                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4453                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4454         } else {
4455                 fail_over_mac_value = BOND_FOM_NONE;
4456         }
4457
4458         if (lp_interval == 0) {
4459                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4460                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4461                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4462         }
4463
4464         /* fill params struct with the proper values */
4465         params->mode = bond_mode;
4466         params->xmit_policy = xmit_hashtype;
4467         params->miimon = miimon;
4468         params->num_peer_notif = num_peer_notif;
4469         params->arp_interval = arp_interval;
4470         params->arp_validate = arp_validate_value;
4471         params->arp_all_targets = arp_all_targets_value;
4472         params->updelay = updelay;
4473         params->downdelay = downdelay;
4474         params->use_carrier = use_carrier;
4475         params->lacp_fast = lacp_fast;
4476         params->primary[0] = 0;
4477         params->primary_reselect = primary_reselect_value;
4478         params->fail_over_mac = fail_over_mac_value;
4479         params->tx_queues = tx_queues;
4480         params->all_slaves_active = all_slaves_active;
4481         params->resend_igmp = resend_igmp;
4482         params->min_links = min_links;
4483         params->lp_interval = lp_interval;
4484         params->packets_per_slave = packets_per_slave;
4485         params->tlb_dynamic_lb = 1; /* Default value */
4486         if (packets_per_slave > 0) {
4487                 params->reciprocal_packets_per_slave =
4488                         reciprocal_value(packets_per_slave);
4489         } else {
4490                 /* reciprocal_packets_per_slave is unused if
4491                  * packets_per_slave is 0 or 1, just initialize it
4492                  */
4493                 params->reciprocal_packets_per_slave =
4494                         (struct reciprocal_value) { 0 };
4495         }
4496
4497         if (primary) {
4498                 strncpy(params->primary, primary, IFNAMSIZ);
4499                 params->primary[IFNAMSIZ - 1] = 0;
4500         }
4501
4502         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4503
4504         return 0;
4505 }
4506
4507 static struct lock_class_key bonding_netdev_xmit_lock_key;
4508 static struct lock_class_key bonding_netdev_addr_lock_key;
4509 static struct lock_class_key bonding_tx_busylock_key;
4510
4511 static void bond_set_lockdep_class_one(struct net_device *dev,
4512                                        struct netdev_queue *txq,
4513                                        void *_unused)
4514 {
4515         lockdep_set_class(&txq->_xmit_lock,
4516                           &bonding_netdev_xmit_lock_key);
4517 }
4518
4519 static void bond_set_lockdep_class(struct net_device *dev)
4520 {
4521         lockdep_set_class(&dev->addr_list_lock,
4522                           &bonding_netdev_addr_lock_key);
4523         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4524         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4525 }
4526
4527 /* Called from registration process */
4528 static int bond_init(struct net_device *bond_dev)
4529 {
4530         struct bonding *bond = netdev_priv(bond_dev);
4531         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4532
4533         netdev_dbg(bond_dev, "Begin bond_init\n");
4534
4535         bond->wq = create_singlethread_workqueue(bond_dev->name);
4536         if (!bond->wq)
4537                 return -ENOMEM;
4538
4539         bond_set_lockdep_class(bond_dev);
4540
4541         list_add_tail(&bond->bond_list, &bn->dev_list);
4542
4543         bond_prepare_sysfs_group(bond);
4544
4545         bond_debug_register(bond);
4546
4547         /* Ensure valid dev_addr */
4548         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4549             bond_dev->addr_assign_type == NET_ADDR_PERM)
4550                 eth_hw_addr_random(bond_dev);
4551
4552         return 0;
4553 }
4554
4555 unsigned int bond_get_num_tx_queues(void)
4556 {
4557         return tx_queues;
4558 }
4559
4560 /* Create a new bond based on the specified name and bonding parameters.
4561  * If name is NULL, obtain a suitable "bond%d" name for us.
4562  * Caller must NOT hold rtnl_lock; we need to release it here before we
4563  * set up our sysfs entries.
4564  */
4565 int bond_create(struct net *net, const char *name)
4566 {
4567         struct net_device *bond_dev;
4568         struct bonding *bond;
4569         struct alb_bond_info *bond_info;
4570         int res;
4571
4572         rtnl_lock();
4573
4574         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4575                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
4576                                    bond_setup, tx_queues);
4577         if (!bond_dev) {
4578                 pr_err("%s: eek! can't alloc netdev!\n", name);
4579                 rtnl_unlock();
4580                 return -ENOMEM;
4581         }
4582
4583         /*
4584          * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4585          * It is set to 0 by default which is wrong.
4586          */
4587         bond = netdev_priv(bond_dev);
4588         bond_info = &(BOND_ALB_INFO(bond));
4589         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4590
4591         dev_net_set(bond_dev, net);
4592         bond_dev->rtnl_link_ops = &bond_link_ops;
4593
4594         res = register_netdevice(bond_dev);
4595
4596         netif_carrier_off(bond_dev);
4597
4598         rtnl_unlock();
4599         if (res < 0)
4600                 bond_destructor(bond_dev);
4601         return res;
4602 }
4603
4604 static int __net_init bond_net_init(struct net *net)
4605 {
4606         struct bond_net *bn = net_generic(net, bond_net_id);
4607
4608         bn->net = net;
4609         INIT_LIST_HEAD(&bn->dev_list);
4610
4611         bond_create_proc_dir(bn);
4612         bond_create_sysfs(bn);
4613
4614         return 0;
4615 }
4616
4617 static void __net_exit bond_net_exit(struct net *net)
4618 {
4619         struct bond_net *bn = net_generic(net, bond_net_id);
4620         struct bonding *bond, *tmp_bond;
4621         LIST_HEAD(list);
4622
4623         bond_destroy_sysfs(bn);
4624
4625         /* Kill off any bonds created after unregistering bond rtnl ops */
4626         rtnl_lock();
4627         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4628                 unregister_netdevice_queue(bond->dev, &list);
4629         unregister_netdevice_many(&list);
4630         rtnl_unlock();
4631
4632         bond_destroy_proc_dir(bn);
4633 }
4634
4635 static struct pernet_operations bond_net_ops = {
4636         .init = bond_net_init,
4637         .exit = bond_net_exit,
4638         .id   = &bond_net_id,
4639         .size = sizeof(struct bond_net),
4640 };
4641
4642 static int __init bonding_init(void)
4643 {
4644         int i;
4645         int res;
4646
4647         pr_info("%s", bond_version);
4648
4649         res = bond_check_params(&bonding_defaults);
4650         if (res)
4651                 goto out;
4652
4653         res = register_pernet_subsys(&bond_net_ops);
4654         if (res)
4655                 goto out;
4656
4657         res = bond_netlink_init();
4658         if (res)
4659                 goto err_link;
4660
4661         bond_create_debugfs();
4662
4663         for (i = 0; i < max_bonds; i++) {
4664                 res = bond_create(&init_net, NULL);
4665                 if (res)
4666                         goto err;
4667         }
4668
4669         register_netdevice_notifier(&bond_netdev_notifier);
4670 out:
4671         return res;
4672 err:
4673         bond_destroy_debugfs();
4674         bond_netlink_fini();
4675 err_link:
4676         unregister_pernet_subsys(&bond_net_ops);
4677         goto out;
4678
4679 }
4680
4681 static void __exit bonding_exit(void)
4682 {
4683         unregister_netdevice_notifier(&bond_netdev_notifier);
4684
4685         bond_destroy_debugfs();
4686
4687         bond_netlink_fini();
4688         unregister_pernet_subsys(&bond_net_ops);
4689
4690 #ifdef CONFIG_NET_POLL_CONTROLLER
4691         /* Make sure we don't have an imbalance on our netpoll blocking */
4692         WARN_ON(atomic_read(&netpoll_block_tx));
4693 #endif
4694 }
4695
4696 module_init(bonding_init);
4697 module_exit(bonding_exit);
4698 MODULE_LICENSE("GPL");
4699 MODULE_VERSION(DRV_VERSION);
4700 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4701 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");