drivers/net/bonding/bond_main.c

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 #include <linux/kernel.h>
  35 #include <linux/module.h>
  36 #include <linux/types.h>
  37 #include <linux/fcntl.h>
  38 #include <linux/interrupt.h>
  39 #include <linux/ptrace.h>
  40 #include <linux/ioport.h>
  41 #include <linux/in.h>
  42 #include <net/ip.h>
  43 #include <linux/ip.h>
  44 #include <linux/tcp.h>
  45 #include <linux/udp.h>
  46 #include <linux/slab.h>
  47 #include <linux/string.h>
  48 #include <linux/init.h>
  49 #include <linux/timer.h>
  50 #include <linux/socket.h>
  51 #include <linux/ctype.h>
  52 #include <linux/inet.h>
  53 #include <linux/bitops.h>
  54 #include <linux/io.h>
  55 #include <asm/dma.h>
  56 #include <linux/uaccess.h>
  57 #include <linux/errno.h>
  58 #include <linux/netdevice.h>
  59 #include <linux/inetdevice.h>
  60 #include <linux/igmp.h>
  61 #include <linux/etherdevice.h>
  62 #include <linux/skbuff.h>
  63 #include <net/sock.h>
  64 #include <linux/rtnetlink.h>
  65 #include <linux/smp.h>
  66 #include <linux/if_ether.h>
  67 #include <net/arp.h>
  68 #include <linux/mii.h>
  69 #include <linux/ethtool.h>
  70 #include <linux/if_vlan.h>
  71 #include <linux/if_bonding.h>
  72 #include <linux/jiffies.h>
  73 #include <linux/preempt.h>
  74 #include <net/route.h>
  75 #include <net/net_namespace.h>
  76 #include <net/netns/generic.h>
  77 #include <net/pkt_sched.h>
  78 #include <linux/rculist.h>
  79 #include <net/flow_dissector.h>
  80 #include <net/switchdev.h>
  81 #include <net/bonding.h>
  82 #include <net/bond_3ad.h>
  83 #include <net/bond_alb.h>
  84
  85 #include "bonding_priv.h"
  86
  87 /*---------------------------- Module parameters ----------------------------*/
  88
  89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  90
  91 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  92 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  93 static int num_peer_notif = 1;
  94 static int miimon;
  95 static int updelay;
  96 static int downdelay;
  97 static int use_carrier  = 1;
  98 static char *mode;
  99 static char *primary;
 100 static char *primary_reselect;
 101 static char *lacp_rate;
 102 static int min_links;
 103 static char *ad_select;
 104 static char *xmit_hash_policy;
 105 static int arp_interval;
 106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 107 static char *arp_validate;
 108 static char *arp_all_targets;
 109 static char *fail_over_mac;
 110 static int all_slaves_active;
 111 static struct bond_params bonding_defaults;
 112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 113 static int packets_per_slave = 1;
 114 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
 115
 116 module_param(max_bonds, int, 0);
 117 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 118 module_param(tx_queues, int, 0);
 119 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 120 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 121 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 122                                "failover event (alias of num_unsol_na)");
 123 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 124 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 125                                "failover event (alias of num_grat_arp)");
 126 module_param(miimon, int, 0);
 127 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 128 module_param(updelay, int, 0);
 129 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 130 module_param(downdelay, int, 0);
 131 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 132                             "in milliseconds");
 133 module_param(use_carrier, int, 0);
 134 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 135                               "0 for off, 1 for on (default)");
 136 module_param(mode, charp, 0);
 137 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 138                        "1 for active-backup, 2 for balance-xor, "
 139                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 140                        "6 for balance-alb");
 141 module_param(primary, charp, 0);
 142 MODULE_PARM_DESC(primary, "Primary network device to use");
 143 module_param(primary_reselect, charp, 0);
 144 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 145                                    "once it comes up; "
 146                                    "0 for always (default), "
 147                                    "1 for only if speed of primary is "
 148                                    "better, "
 149                                    "2 for only on active slave "
 150                                    "failure");
 151 module_param(lacp_rate, charp, 0);
 152 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 153                             "0 for slow, 1 for fast");
 154 module_param(ad_select, charp, 0);
 155 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 156                             "0 for stable (default), 1 for bandwidth, "
 157                             "2 for count");
 158 module_param(min_links, int, 0);
 159 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 160
 161 module_param(xmit_hash_policy, charp, 0);
 162 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 163                                    "0 for layer 2 (default), 1 for layer 3+4, "
 164                                    "2 for layer 2+3, 3 for encap layer 2+3, "
 165                                    "4 for encap layer 3+4");
 166 module_param(arp_interval, int, 0);
 167 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 168 module_param_array(arp_ip_target, charp, NULL, 0);
 169 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 170 module_param(arp_validate, charp, 0);
 171 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 172                                "0 for none (default), 1 for active, "
 173                                "2 for backup, 3 for all");
 174 module_param(arp_all_targets, charp, 0);
 175 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 176 module_param(fail_over_mac, charp, 0);
 177 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 178                                 "the same MAC; 0 for none (default), "
 179                                 "1 for active, 2 for follow");
 180 module_param(all_slaves_active, int, 0);
 181 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
 182                                      "by setting active flag for all slaves; "
 183                                      "0 for never (default), 1 for always.");
 184 module_param(resend_igmp, int, 0);
 185 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 186                               "link failure");
 187 module_param(packets_per_slave, int, 0);
 188 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
 189                                     "mode; 0 for a random slave, 1 packet per "
 190                                     "slave (default), >1 packets per slave.");
 191 module_param(lp_interval, uint, 0);
 192 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
 193                               "the bonding driver sends learning packets to "
 194                               "each slaves peer switch. The default is 1.");
 195
 196 /*----------------------------- Global variables ----------------------------*/
 197
 198 #ifdef CONFIG_NET_POLL_CONTROLLER
 199 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 200 #endif
 201
 202 int bond_net_id __read_mostly;
 203
 204 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 205 static int arp_ip_count;
 206 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 207 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 208 static int lacp_fast;
 209
 210 /*-------------------------- Forward declarations ---------------------------*/
 211
 212 static int bond_init(struct net_device *bond_dev);
 213 static void bond_uninit(struct net_device *bond_dev);
 214 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
 215                                                 struct rtnl_link_stats64 *stats);
 216 static void bond_slave_arr_handler(struct work_struct *work);
 217
 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 struct slave *bond_choose_primary_or_current(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 || prim->link != BOND_LINK_UP) {
 718                 if (!curr || curr->link != BOND_LINK_UP)
 719                         return NULL;
 720                 return curr;
 721         }
 722
 723         if (bond->force_primary) {
 724                 bond->force_primary = false;
 725                 return prim;
 726         }
 727
 728         if (!curr || curr->link != BOND_LINK_UP)
 729                 return prim;
 730
 731         /* At this point, prim and curr are both up */
 732         switch (bond->params.primary_reselect) {
 733         case BOND_PRI_RESELECT_ALWAYS:
 734                 return prim;
 735         case BOND_PRI_RESELECT_BETTER:
 736                 if (prim->speed < curr->speed)
 737                         return curr;
 738                 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
 739                         return curr;
 740                 return prim;
 741         case BOND_PRI_RESELECT_FAILURE:
 742                 return curr;
 743         default:
 744                 netdev_err(bond->dev, "impossible primary_reselect %d\n",
 745                            bond->params.primary_reselect);
 746                 return curr;
 747         }
 748 }
 749
 750 /**
 751  * bond_find_best_slave - select the best available slave to be the active one
 752  * @bond: our bonding struct
 753  */
 754 static struct slave *bond_find_best_slave(struct bonding *bond)
 755 {
 756         struct slave *slave, *bestslave = NULL;
 757         struct list_head *iter;
 758         int mintime = bond->params.updelay;
 759
 760         slave = bond_choose_primary_or_current(bond);
 761         if (slave)
 762                 return slave;
 763
 764         bond_for_each_slave(bond, slave, iter) {
 765                 if (slave->link == BOND_LINK_UP)
 766                         return slave;
 767                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
 768                     slave->delay < mintime) {
 769                         mintime = slave->delay;
 770                         bestslave = slave;
 771                 }
 772         }
 773
 774         return bestslave;
 775 }
 776
 777 static bool bond_should_notify_peers(struct bonding *bond)
 778 {
 779         struct slave *slave;
 780
 781         rcu_read_lock();
 782         slave = rcu_dereference(bond->curr_active_slave);
 783         rcu_read_unlock();
 784
 785         netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
 786                    slave ? slave->dev->name : "NULL");
 787
 788         if (!slave || !bond->send_peer_notif ||
 789             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 790                 return false;
 791
 792         return true;
 793 }
 794
 795 /**
 796  * change_active_interface - change the active slave into the specified one
 797  * @bond: our bonding struct
 798  * @new: the new slave to make the active one
 799  *
 800  * Set the new slave to the bond's settings and unset them on the old
 801  * curr_active_slave.
 802  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 803  *
 804  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 805  * because it is apparently the best available slave we have, even though its
 806  * updelay hasn't timed out yet.
 807  *
 808  * Caller must hold RTNL.
 809  */
 810 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 811 {
 812         struct slave *old_active;
 813
 814         ASSERT_RTNL();
 815
 816         old_active = rtnl_dereference(bond->curr_active_slave);
 817
 818         if (old_active == new_active)
 819                 return;
 820
 821         if (new_active) {
 822                 new_active->last_link_up = jiffies;
 823
 824                 if (new_active->link == BOND_LINK_BACK) {
 825                         if (bond_uses_primary(bond)) {
 826                                 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
 827                                             new_active->dev->name,
 828                                             (bond->params.updelay - new_active->delay) * bond->params.miimon);
 829                         }
 830
 831                         new_active->delay = 0;
 832                         bond_set_slave_link_state(new_active, BOND_LINK_UP);
 833
 834                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 835                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 836
 837                         if (bond_is_lb(bond))
 838                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 839                 } else {
 840                         if (bond_uses_primary(bond)) {
 841                                 netdev_info(bond->dev, "making interface %s the new active one\n",
 842                                             new_active->dev->name);
 843                         }
 844                 }
 845         }
 846
 847         if (bond_uses_primary(bond))
 848                 bond_hw_addr_swap(bond, new_active, old_active);
 849
 850         if (bond_is_lb(bond)) {
 851                 bond_alb_handle_active_change(bond, new_active);
 852                 if (old_active)
 853                         bond_set_slave_inactive_flags(old_active,
 854                                                       BOND_SLAVE_NOTIFY_NOW);
 855                 if (new_active)
 856                         bond_set_slave_active_flags(new_active,
 857                                                     BOND_SLAVE_NOTIFY_NOW);
 858         } else {
 859                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 860         }
 861
 862         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
 863                 if (old_active)
 864                         bond_set_slave_inactive_flags(old_active,
 865                                                       BOND_SLAVE_NOTIFY_NOW);
 866
 867                 if (new_active) {
 868                         bool should_notify_peers = false;
 869
 870                         bond_set_slave_active_flags(new_active,
 871                                                     BOND_SLAVE_NOTIFY_NOW);
 872
 873                         if (bond->params.fail_over_mac)
 874                                 bond_do_fail_over_mac(bond, new_active,
 875                                                       old_active);
 876
 877                         if (netif_running(bond->dev)) {
 878                                 bond->send_peer_notif =
 879                                         bond->params.num_peer_notif;
 880                                 should_notify_peers =
 881                                         bond_should_notify_peers(bond);
 882                         }
 883
 884                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 885                         if (should_notify_peers)
 886                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 887                                                          bond->dev);
 888                 }
 889         }
 890
 891         /* resend IGMP joins since active slave has changed or
 892          * all were sent on curr_active_slave.
 893          * resend only if bond is brought up with the affected
 894          * bonding modes and the retransmission is enabled
 895          */
 896         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 897             ((bond_uses_primary(bond) && new_active) ||
 898              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
 899                 bond->igmp_retrans = bond->params.resend_igmp;
 900                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 901         }
 902 }
 903
 904 /**
 905  * bond_select_active_slave - select a new active slave, if needed
 906  * @bond: our bonding struct
 907  *
 908  * This functions should be called when one of the following occurs:
 909  * - The old curr_active_slave has been released or lost its link.
 910  * - The primary_slave has got its link back.
 911  * - A slave has got its link back and there's no old curr_active_slave.
 912  *
 913  * Caller must hold RTNL.
 914  */
 915 void bond_select_active_slave(struct bonding *bond)
 916 {
 917         struct slave *best_slave;
 918         int rv;
 919
 920         ASSERT_RTNL();
 921
 922         best_slave = bond_find_best_slave(bond);
 923         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
 924                 bond_change_active_slave(bond, best_slave);
 925                 rv = bond_set_carrier(bond);
 926                 if (!rv)
 927                         return;
 928
 929                 if (netif_carrier_ok(bond->dev)) {
 930                         netdev_info(bond->dev, "first active interface up!\n");
 931                 } else {
 932                         netdev_info(bond->dev, "now running without any active interface!\n");
 933                 }
 934         }
 935 }
 936
 937 #ifdef CONFIG_NET_POLL_CONTROLLER
 938 static inline int slave_enable_netpoll(struct slave *slave)
 939 {
 940         struct netpoll *np;
 941         int err = 0;
 942
 943         np = kzalloc(sizeof(*np), GFP_KERNEL);
 944         err = -ENOMEM;
 945         if (!np)
 946                 goto out;
 947
 948         err = __netpoll_setup(np, slave->dev);
 949         if (err) {
 950                 kfree(np);
 951                 goto out;
 952         }
 953         slave->np = np;
 954 out:
 955         return err;
 956 }
 957 static inline void slave_disable_netpoll(struct slave *slave)
 958 {
 959         struct netpoll *np = slave->np;
 960
 961         if (!np)
 962                 return;
 963
 964         slave->np = NULL;
 965         __netpoll_free_async(np);
 966 }
 967
 968 static void bond_poll_controller(struct net_device *bond_dev)
 969 {
 970         struct bonding *bond = netdev_priv(bond_dev);
 971         struct slave *slave = NULL;
 972         struct list_head *iter;
 973         struct ad_info ad_info;
 974         struct netpoll_info *ni;
 975         const struct net_device_ops *ops;
 976
 977         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 978                 if (bond_3ad_get_active_agg_info(bond, &ad_info))
 979                         return;
 980
 981         rcu_read_lock_bh();
 982         bond_for_each_slave_rcu(bond, slave, iter) {
 983                 ops = slave->dev->netdev_ops;
 984                 if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller)
 985                         continue;
 986
 987                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 988                         struct aggregator *agg =
 989                             SLAVE_AD_INFO(slave)->port.aggregator;
 990
 991                         if (agg &&
 992                             agg->aggregator_identifier != ad_info.aggregator_id)
 993                                 continue;
 994                 }
 995
 996                 ni = rcu_dereference_bh(slave->dev->npinfo);
 997                 if (down_trylock(&ni->dev_lock))
 998                         continue;
 999                 ops->ndo_poll_controller(slave->dev);
1000                 up(&ni->dev_lock);
1001         }
1002         rcu_read_unlock_bh();
1003 }
1004
1005 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1006 {
1007         struct bonding *bond = netdev_priv(bond_dev);
1008         struct list_head *iter;
1009         struct slave *slave;
1010
1011         bond_for_each_slave(bond, slave, iter)
1012                 if (bond_slave_is_up(slave))
1013                         slave_disable_netpoll(slave);
1014 }
1015
1016 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1017 {
1018         struct bonding *bond = netdev_priv(dev);
1019         struct list_head *iter;
1020         struct slave *slave;
1021         int err = 0;
1022
1023         bond_for_each_slave(bond, slave, iter) {
1024                 err = slave_enable_netpoll(slave);
1025                 if (err) {
1026                         bond_netpoll_cleanup(dev);
1027                         break;
1028                 }
1029         }
1030         return err;
1031 }
1032 #else
1033 static inline int slave_enable_netpoll(struct slave *slave)
1034 {
1035         return 0;
1036 }
1037 static inline void slave_disable_netpoll(struct slave *slave)
1038 {
1039 }
1040 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1041 {
1042 }
1043 #endif
1044
1045 /*---------------------------------- IOCTL ----------------------------------*/
1046
1047 static netdev_features_t bond_fix_features(struct net_device *dev,
1048                                            netdev_features_t features)
1049 {
1050         struct bonding *bond = netdev_priv(dev);
1051         struct list_head *iter;
1052         netdev_features_t mask;
1053         struct slave *slave;
1054
1055         mask = features;
1056
1057         features &= ~NETIF_F_ONE_FOR_ALL;
1058         features |= NETIF_F_ALL_FOR_ALL;
1059
1060         bond_for_each_slave(bond, slave, iter) {
1061                 features = netdev_increment_features(features,
1062                                                      slave->dev->features,
1063                                                      mask);
1064         }
1065         features = netdev_add_tso_features(features, mask);
1066
1067         return features;
1068 }
1069
1070 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1071                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1072                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1073
1074 #define BOND_ENC_FEATURES       (NETIF_F_ALL_CSUM | NETIF_F_SG | NETIF_F_RXCSUM |\
1075                                  NETIF_F_TSO)
1076
1077 static void bond_compute_features(struct bonding *bond)
1078 {
1079         unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1080                                         IFF_XMIT_DST_RELEASE_PERM;
1081         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1082         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1083         struct net_device *bond_dev = bond->dev;
1084         struct list_head *iter;
1085         struct slave *slave;
1086         unsigned short max_hard_header_len = ETH_HLEN;
1087         unsigned int gso_max_size = GSO_MAX_SIZE;
1088         u16 gso_max_segs = GSO_MAX_SEGS;
1089
1090         if (!bond_has_slaves(bond))
1091                 goto done;
1092         vlan_features &= NETIF_F_ALL_FOR_ALL;
1093
1094         bond_for_each_slave(bond, slave, iter) {
1095                 vlan_features = netdev_increment_features(vlan_features,
1096                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1097
1098                 enc_features = netdev_increment_features(enc_features,
1099                                                          slave->dev->hw_enc_features,
1100                                                          BOND_ENC_FEATURES);
1101                 dst_release_flag &= slave->dev->priv_flags;
1102                 if (slave->dev->hard_header_len > max_hard_header_len)
1103                         max_hard_header_len = slave->dev->hard_header_len;
1104
1105                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1106                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1107         }
1108
1109 done:
1110         bond_dev->vlan_features = vlan_features;
1111         bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
1112         bond_dev->hard_header_len = max_hard_header_len;
1113         bond_dev->gso_max_segs = gso_max_segs;
1114         netif_set_gso_max_size(bond_dev, gso_max_size);
1115
1116         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1117         if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1118             dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1119                 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1120
1121         netdev_change_features(bond_dev);
1122 }
1123
1124 static void bond_setup_by_slave(struct net_device *bond_dev,
1125                                 struct net_device *slave_dev)
1126 {
1127         bond_dev->header_ops        = slave_dev->header_ops;
1128
1129         bond_dev->type              = slave_dev->type;
1130         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1131         bond_dev->addr_len          = slave_dev->addr_len;
1132
1133         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1134                 slave_dev->addr_len);
1135 }
1136
1137 /* On bonding slaves other than the currently active slave, suppress
1138  * duplicates except for alb non-mcast/bcast.
1139  */
1140 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1141                                             struct slave *slave,
1142                                             struct bonding *bond)
1143 {
1144         if (bond_is_slave_inactive(slave)) {
1145                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1146                     skb->pkt_type != PACKET_BROADCAST &&
1147                     skb->pkt_type != PACKET_MULTICAST)
1148                         return false;
1149                 return true;
1150         }
1151         return false;
1152 }
1153
1154 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1155 {
1156         struct sk_buff *skb = *pskb;
1157         struct slave *slave;
1158         struct bonding *bond;
1159         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1160                           struct slave *);
1161         int ret = RX_HANDLER_ANOTHER;
1162
1163         skb = skb_share_check(skb, GFP_ATOMIC);
1164         if (unlikely(!skb))
1165                 return RX_HANDLER_CONSUMED;
1166
1167         *pskb = skb;
1168
1169         slave = bond_slave_get_rcu(skb->dev);
1170         bond = slave->bond;
1171
1172         recv_probe = ACCESS_ONCE(bond->recv_probe);
1173         if (recv_probe) {
1174                 ret = recv_probe(skb, bond, slave);
1175                 if (ret == RX_HANDLER_CONSUMED) {
1176                         consume_skb(skb);
1177                         return ret;
1178                 }
1179         }
1180
1181         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1182                 return RX_HANDLER_EXACT;
1183         }
1184
1185         skb->dev = bond->dev;
1186
1187         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1188             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1189             skb->pkt_type == PACKET_HOST) {
1190
1191                 if (unlikely(skb_cow_head(skb,
1192                                           skb->data - skb_mac_header(skb)))) {
1193                         kfree_skb(skb);
1194                         return RX_HANDLER_CONSUMED;
1195                 }
1196                 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1197         }
1198
1199         return ret;
1200 }
1201
1202 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1203                                       struct net_device *slave_dev,
1204                                       struct slave *slave)
1205 {
1206         int err;
1207
1208         err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1209         if (err)
1210                 return err;
1211         slave_dev->flags |= IFF_SLAVE;
1212         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1213         return 0;
1214 }
1215
1216 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1217                                   struct net_device *slave_dev)
1218 {
1219         netdev_upper_dev_unlink(slave_dev, bond_dev);
1220         slave_dev->flags &= ~IFF_SLAVE;
1221         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1222 }
1223
1224 static struct slave *bond_alloc_slave(struct bonding *bond)
1225 {
1226         struct slave *slave = NULL;
1227
1228         slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1229         if (!slave)
1230                 return NULL;
1231
1232         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1233                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1234                                                GFP_KERNEL);
1235                 if (!SLAVE_AD_INFO(slave)) {
1236                         kfree(slave);
1237                         return NULL;
1238                 }
1239         }
1240         return slave;
1241 }
1242
1243 static void bond_free_slave(struct slave *slave)
1244 {
1245         struct bonding *bond = bond_get_bond_by_slave(slave);
1246
1247         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1248                 kfree(SLAVE_AD_INFO(slave));
1249
1250         kfree(slave);
1251 }
1252
1253 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1254 {
1255         info->bond_mode = BOND_MODE(bond);
1256         info->miimon = bond->params.miimon;
1257         info->num_slaves = bond->slave_cnt;
1258 }
1259
1260 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1261 {
1262         strcpy(info->slave_name, slave->dev->name);
1263         info->link = slave->link;
1264         info->state = bond_slave_state(slave);
1265         info->link_failure_count = slave->link_failure_count;
1266 }
1267
1268 static void bond_netdev_notify(struct net_device *dev,
1269                                struct netdev_bonding_info *info)
1270 {
1271         rtnl_lock();
1272         netdev_bonding_info_change(dev, info);
1273         rtnl_unlock();
1274 }
1275
1276 static void bond_netdev_notify_work(struct work_struct *_work)
1277 {
1278         struct netdev_notify_work *w =
1279                 container_of(_work, struct netdev_notify_work, work.work);
1280
1281         bond_netdev_notify(w->dev, &w->bonding_info);
1282         dev_put(w->dev);
1283         kfree(w);
1284 }
1285
1286 void bond_queue_slave_event(struct slave *slave)
1287 {
1288         struct bonding *bond = slave->bond;
1289         struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1290
1291         if (!nnw)
1292                 return;
1293
1294         dev_hold(slave->dev);
1295         nnw->dev = slave->dev;
1296         bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1297         bond_fill_ifbond(bond, &nnw->bonding_info.master);
1298         INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1299
1300         queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1301 }
1302
1303 /* enslave device <slave> to bond device <master> */
1304 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1305 {
1306         struct bonding *bond = netdev_priv(bond_dev);
1307         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1308         struct slave *new_slave = NULL, *prev_slave;
1309         struct sockaddr addr;
1310         int link_reporting;
1311         int res = 0, i;
1312
1313         if (!bond->params.use_carrier &&
1314             slave_dev->ethtool_ops->get_link == NULL &&
1315             slave_ops->ndo_do_ioctl == NULL) {
1316                 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1317                             slave_dev->name);
1318         }
1319
1320         /* already enslaved */
1321         if (slave_dev->flags & IFF_SLAVE) {
1322                 netdev_dbg(bond_dev, "Error: Device was already enslaved\n");
1323                 return -EBUSY;
1324         }
1325
1326         if (bond_dev == slave_dev) {
1327                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1328                 return -EPERM;
1329         }
1330
1331         /* vlan challenged mutual exclusion */
1332         /* no need to lock since we're protected by rtnl_lock */
1333         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1334                 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1335                            slave_dev->name);
1336                 if (vlan_uses_dev(bond_dev)) {
1337                         netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1338                                    slave_dev->name, bond_dev->name);
1339                         return -EPERM;
1340                 } else {
1341                         netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1342                                     slave_dev->name, slave_dev->name,
1343                                     bond_dev->name);
1344                 }
1345         } else {
1346                 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1347                            slave_dev->name);
1348         }
1349
1350         /* Old ifenslave binaries are no longer supported.  These can
1351          * be identified with moderate accuracy by the state of the slave:
1352          * the current ifenslave will set the interface down prior to
1353          * enslaving it; the old ifenslave will not.
1354          */
1355         if ((slave_dev->flags & IFF_UP)) {
1356                 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1357                            slave_dev->name);
1358                 res = -EPERM;
1359                 goto err_undo_flags;
1360         }
1361
1362         /* set bonding device ether type by slave - bonding netdevices are
1363          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1364          * there is a need to override some of the type dependent attribs/funcs.
1365          *
1366          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1367          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1368          */
1369         if (!bond_has_slaves(bond)) {
1370                 if (bond_dev->type != slave_dev->type) {
1371                         netdev_dbg(bond_dev, "change device type from %d to %d\n",
1372                                    bond_dev->type, slave_dev->type);
1373
1374                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1375                                                        bond_dev);
1376                         res = notifier_to_errno(res);
1377                         if (res) {
1378                                 netdev_err(bond_dev, "refused to change device type\n");
1379                                 res = -EBUSY;
1380                                 goto err_undo_flags;
1381                         }
1382
1383                         /* Flush unicast and multicast addresses */
1384                         dev_uc_flush(bond_dev);
1385                         dev_mc_flush(bond_dev);
1386
1387                         if (slave_dev->type != ARPHRD_ETHER)
1388                                 bond_setup_by_slave(bond_dev, slave_dev);
1389                         else {
1390                                 ether_setup(bond_dev);
1391                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1392                         }
1393
1394                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1395                                                  bond_dev);
1396                 }
1397         } else if (bond_dev->type != slave_dev->type) {
1398                 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1399                            slave_dev->name, slave_dev->type, bond_dev->type);
1400                 res = -EINVAL;
1401                 goto err_undo_flags;
1402         }
1403
1404         if (slave_ops->ndo_set_mac_address == NULL) {
1405                 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1406                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1407                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1408                         if (!bond_has_slaves(bond)) {
1409                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1410                                 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1411                         } else {
1412                                 netdev_err(bond_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
1413                                 res = -EOPNOTSUPP;
1414                                 goto err_undo_flags;
1415                         }
1416                 }
1417         }
1418
1419         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1420
1421         /* If this is the first slave, then we need to set the master's hardware
1422          * address to be the same as the slave's.
1423          */
1424         if (!bond_has_slaves(bond) &&
1425             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1426                 bond_set_dev_addr(bond->dev, slave_dev);
1427
1428         new_slave = bond_alloc_slave(bond);
1429         if (!new_slave) {
1430                 res = -ENOMEM;
1431                 goto err_undo_flags;
1432         }
1433
1434         new_slave->bond = bond;
1435         new_slave->dev = slave_dev;
1436         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1437          * is set via sysfs or module option if desired.
1438          */
1439         new_slave->queue_id = 0;
1440
1441         /* Save slave's original mtu and then set it to match the bond */
1442         new_slave->original_mtu = slave_dev->mtu;
1443         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1444         if (res) {
1445                 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1446                 goto err_free;
1447         }
1448
1449         /* Save slave's original ("permanent") mac address for modes
1450          * that need it, and for restoring it upon release, and then
1451          * set it to the master's address
1452          */
1453         ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1454
1455         if (!bond->params.fail_over_mac ||
1456             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1457                 /* Set slave to master's mac address.  The application already
1458                  * set the master's mac address to that of the first slave
1459                  */
1460                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1461                 addr.sa_family = slave_dev->type;
1462                 res = dev_set_mac_address(slave_dev, &addr);
1463                 if (res) {
1464                         netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1465                         goto err_restore_mtu;
1466                 }
1467         }
1468
1469         /* open the slave since the application closed it */
1470         res = dev_open(slave_dev);
1471         if (res) {
1472                 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1473                 goto err_restore_mac;
1474         }
1475
1476         slave_dev->priv_flags |= IFF_BONDING;
1477         /* initialize slave stats */
1478         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1479
1480         if (bond_is_lb(bond)) {
1481                 /* bond_alb_init_slave() must be called before all other stages since
1482                  * it might fail and we do not want to have to undo everything
1483                  */
1484                 res = bond_alb_init_slave(bond, new_slave);
1485                 if (res)
1486                         goto err_close;
1487         }
1488
1489         /* If the mode uses primary, then the following is handled by
1490          * bond_change_active_slave().
1491          */
1492         if (!bond_uses_primary(bond)) {
1493                 /* set promiscuity level to new slave */
1494                 if (bond_dev->flags & IFF_PROMISC) {
1495                         res = dev_set_promiscuity(slave_dev, 1);
1496                         if (res)
1497                                 goto err_close;
1498                 }
1499
1500                 /* set allmulti level to new slave */
1501                 if (bond_dev->flags & IFF_ALLMULTI) {
1502                         res = dev_set_allmulti(slave_dev, 1);
1503                         if (res)
1504                                 goto err_close;
1505                 }
1506
1507                 netif_addr_lock_bh(bond_dev);
1508
1509                 dev_mc_sync_multiple(slave_dev, bond_dev);
1510                 dev_uc_sync_multiple(slave_dev, bond_dev);
1511
1512                 netif_addr_unlock_bh(bond_dev);
1513         }
1514
1515         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1516                 /* add lacpdu mc addr to mc list */
1517                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1518
1519                 dev_mc_add(slave_dev, lacpdu_multicast);
1520         }
1521
1522         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1523         if (res) {
1524                 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1525                            slave_dev->name);
1526                 goto err_close;
1527         }
1528
1529         prev_slave = bond_last_slave(bond);
1530
1531         new_slave->delay = 0;
1532         new_slave->link_failure_count = 0;
1533
1534         bond_update_speed_duplex(new_slave);
1535
1536         new_slave->last_rx = jiffies -
1537                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1538         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1539                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1540
1541         if (bond->params.miimon && !bond->params.use_carrier) {
1542                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1543
1544                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1545                         /* miimon is set but a bonded network driver
1546                          * does not support ETHTOOL/MII and
1547                          * arp_interval is not set.  Note: if
1548                          * use_carrier is enabled, we will never go
1549                          * here (because netif_carrier is always
1550                          * supported); thus, we don't need to change
1551                          * the messages for netif_carrier.
1552                          */
1553                         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",
1554                                     slave_dev->name);
1555                 } else if (link_reporting == -1) {
1556                         /* unable get link status using mii/ethtool */
1557                         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",
1558                                     slave_dev->name);
1559                 }
1560         }
1561
1562         /* check for initial state */
1563         if (bond->params.miimon) {
1564                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1565                         if (bond->params.updelay) {
1566                                 bond_set_slave_link_state(new_slave,
1567                                                           BOND_LINK_BACK);
1568                                 new_slave->delay = bond->params.updelay;
1569                         } else {
1570                                 bond_set_slave_link_state(new_slave,
1571                                                           BOND_LINK_UP);
1572                         }
1573                 } else {
1574                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN);
1575                 }
1576         } else if (bond->params.arp_interval) {
1577                 bond_set_slave_link_state(new_slave,
1578                                           (netif_carrier_ok(slave_dev) ?
1579                                           BOND_LINK_UP : BOND_LINK_DOWN));
1580         } else {
1581                 bond_set_slave_link_state(new_slave, BOND_LINK_UP);
1582         }
1583
1584         if (new_slave->link != BOND_LINK_DOWN)
1585                 new_slave->last_link_up = jiffies;
1586         netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1587                    new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1588                    (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1589
1590         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1591                 /* if there is a primary slave, remember it */
1592                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1593                         rcu_assign_pointer(bond->primary_slave, new_slave);
1594                         bond->force_primary = true;
1595                 }
1596         }
1597
1598         switch (BOND_MODE(bond)) {
1599         case BOND_MODE_ACTIVEBACKUP:
1600                 bond_set_slave_inactive_flags(new_slave,
1601                                               BOND_SLAVE_NOTIFY_NOW);
1602                 break;
1603         case BOND_MODE_8023AD:
1604                 /* in 802.3ad mode, the internal mechanism
1605                  * will activate the slaves in the selected
1606                  * aggregator
1607                  */
1608                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1609                 /* if this is the first slave */
1610                 if (!prev_slave) {
1611                         SLAVE_AD_INFO(new_slave)->id = 1;
1612                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1613                          * can be called only after the mac address of the bond is set
1614                          */
1615                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1616                 } else {
1617                         SLAVE_AD_INFO(new_slave)->id =
1618                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1619                 }
1620
1621                 bond_3ad_bind_slave(new_slave);
1622                 break;
1623         case BOND_MODE_TLB:
1624         case BOND_MODE_ALB:
1625                 bond_set_active_slave(new_slave);
1626                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1627                 break;
1628         default:
1629                 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1630
1631                 /* always active in trunk mode */
1632                 bond_set_active_slave(new_slave);
1633
1634                 /* In trunking mode there is little meaning to curr_active_slave
1635                  * anyway (it holds no special properties of the bond device),
1636                  * so we can change it without calling change_active_interface()
1637                  */
1638                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1639                     new_slave->link == BOND_LINK_UP)
1640                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1641
1642                 break;
1643         } /* switch(bond_mode) */
1644
1645 #ifdef CONFIG_NET_POLL_CONTROLLER
1646         slave_dev->npinfo = bond->dev->npinfo;
1647         if (slave_dev->npinfo) {
1648                 if (slave_enable_netpoll(new_slave)) {
1649                         netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1650                         res = -EBUSY;
1651                         goto err_detach;
1652                 }
1653         }
1654 #endif
1655
1656         if (!(bond_dev->features & NETIF_F_LRO))
1657                 dev_disable_lro(slave_dev);
1658
1659         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1660                                          new_slave);
1661         if (res) {
1662                 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1663                 goto err_detach;
1664         }
1665
1666         res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1667         if (res) {
1668                 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1669                 goto err_unregister;
1670         }
1671
1672         res = bond_sysfs_slave_add(new_slave);
1673         if (res) {
1674                 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1675                 goto err_upper_unlink;
1676         }
1677
1678         bond->slave_cnt++;
1679         bond_compute_features(bond);
1680         bond_set_carrier(bond);
1681
1682         if (bond_uses_primary(bond)) {
1683                 block_netpoll_tx();
1684                 bond_select_active_slave(bond);
1685                 unblock_netpoll_tx();
1686         }
1687
1688         if (bond_mode_uses_xmit_hash(bond))
1689                 bond_update_slave_arr(bond, NULL);
1690
1691         netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1692                     slave_dev->name,
1693                     bond_is_active_slave(new_slave) ? "an active" : "a backup",
1694                     new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1695
1696         /* enslave is successful */
1697         bond_queue_slave_event(new_slave);
1698         return 0;
1699
1700 /* Undo stages on error */
1701 err_upper_unlink:
1702         bond_upper_dev_unlink(bond_dev, slave_dev);
1703
1704 err_unregister:
1705         netdev_rx_handler_unregister(slave_dev);
1706
1707 err_detach:
1708         if (!bond_uses_primary(bond))
1709                 bond_hw_addr_flush(bond_dev, slave_dev);
1710
1711         vlan_vids_del_by_dev(slave_dev, bond_dev);
1712         if (rcu_access_pointer(bond->primary_slave) == new_slave)
1713                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1714         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1715                 block_netpoll_tx();
1716                 bond_change_active_slave(bond, NULL);
1717                 bond_select_active_slave(bond);
1718                 unblock_netpoll_tx();
1719         }
1720         /* either primary_slave or curr_active_slave might've changed */
1721         synchronize_rcu();
1722         slave_disable_netpoll(new_slave);
1723
1724 err_close:
1725         slave_dev->priv_flags &= ~IFF_BONDING;
1726         dev_close(slave_dev);
1727
1728 err_restore_mac:
1729         if (!bond->params.fail_over_mac ||
1730             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1731                 /* XXX TODO - fom follow mode needs to change master's
1732                  * MAC if this slave's MAC is in use by the bond, or at
1733                  * least print a warning.
1734                  */
1735                 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1736                 addr.sa_family = slave_dev->type;
1737                 dev_set_mac_address(slave_dev, &addr);
1738         }
1739
1740 err_restore_mtu:
1741         dev_set_mtu(slave_dev, new_slave->original_mtu);
1742
1743 err_free:
1744         bond_free_slave(new_slave);
1745
1746 err_undo_flags:
1747         /* Enslave of first slave has failed and we need to fix master's mac */
1748         if (!bond_has_slaves(bond)) {
1749                 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1750                                             slave_dev->dev_addr))
1751                         eth_hw_addr_random(bond_dev);
1752                 if (bond_dev->type != ARPHRD_ETHER) {
1753                         ether_setup(bond_dev);
1754                         bond_dev->flags |= IFF_MASTER;
1755                         bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1756                 }
1757         }
1758
1759         return res;
1760 }
1761
1762 /* Try to release the slave device <slave> from the bond device <master>
1763  * It is legal to access curr_active_slave without a lock because all the function
1764  * is RTNL-locked. If "all" is true it means that the function is being called
1765  * while destroying a bond interface and all slaves are being released.
1766  *
1767  * The rules for slave state should be:
1768  *   for Active/Backup:
1769  *     Active stays on all backups go down
1770  *   for Bonded connections:
1771  *     The first up interface should be left on and all others downed.
1772  */
1773 static int __bond_release_one(struct net_device *bond_dev,
1774                               struct net_device *slave_dev,
1775                               bool all)
1776 {
1777         struct bonding *bond = netdev_priv(bond_dev);
1778         struct slave *slave, *oldcurrent;
1779         struct sockaddr addr;
1780         int old_flags = bond_dev->flags;
1781         netdev_features_t old_features = bond_dev->features;
1782
1783         /* slave is not a slave or master is not master of this slave */
1784         if (!(slave_dev->flags & IFF_SLAVE) ||
1785             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1786                 netdev_dbg(bond_dev, "cannot release %s\n",
1787                            slave_dev->name);
1788                 return -EINVAL;
1789         }
1790
1791         block_netpoll_tx();
1792
1793         slave = bond_get_slave_by_dev(bond, slave_dev);
1794         if (!slave) {
1795                 /* not a slave of this bond */
1796                 netdev_info(bond_dev, "%s not enslaved\n",
1797                             slave_dev->name);
1798                 unblock_netpoll_tx();
1799                 return -EINVAL;
1800         }
1801
1802         bond_sysfs_slave_del(slave);
1803
1804         /* recompute stats just before removing the slave */
1805         bond_get_stats(bond->dev, &bond->bond_stats);
1806
1807         bond_upper_dev_unlink(bond_dev, slave_dev);
1808         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1809          * for this slave anymore.
1810          */
1811         netdev_rx_handler_unregister(slave_dev);
1812
1813         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1814                 bond_3ad_unbind_slave(slave);
1815
1816         if (bond_mode_uses_xmit_hash(bond))
1817                 bond_update_slave_arr(bond, slave);
1818
1819         netdev_info(bond_dev, "Releasing %s interface %s\n",
1820                     bond_is_active_slave(slave) ? "active" : "backup",
1821                     slave_dev->name);
1822
1823         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1824
1825         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1826
1827         if (!all && (!bond->params.fail_over_mac ||
1828                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1829                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1830                     bond_has_slaves(bond))
1831                         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",
1832                                     slave_dev->name, slave->perm_hwaddr,
1833                                     bond_dev->name, slave_dev->name);
1834         }
1835
1836         if (rtnl_dereference(bond->primary_slave) == slave)
1837                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1838
1839         if (oldcurrent == slave)
1840                 bond_change_active_slave(bond, NULL);
1841
1842         if (bond_is_lb(bond)) {
1843                 /* Must be called only after the slave has been
1844                  * detached from the list and the curr_active_slave
1845                  * has been cleared (if our_slave == old_current),
1846                  * but before a new active slave is selected.
1847                  */
1848                 bond_alb_deinit_slave(bond, slave);
1849         }
1850
1851         if (all) {
1852                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1853         } else if (oldcurrent == slave) {
1854                 /* Note that we hold RTNL over this sequence, so there
1855                  * is no concern that another slave add/remove event
1856                  * will interfere.
1857                  */
1858                 bond_select_active_slave(bond);
1859         }
1860
1861         if (!bond_has_slaves(bond)) {
1862                 bond_set_carrier(bond);
1863                 eth_hw_addr_random(bond_dev);
1864         }
1865
1866         unblock_netpoll_tx();
1867         synchronize_rcu();
1868         bond->slave_cnt--;
1869
1870         if (!bond_has_slaves(bond)) {
1871                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1872                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1873         }
1874
1875         bond_compute_features(bond);
1876         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1877             (old_features & NETIF_F_VLAN_CHALLENGED))
1878                 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1879                             slave_dev->name, bond_dev->name);
1880
1881         vlan_vids_del_by_dev(slave_dev, bond_dev);
1882
1883         /* If the mode uses primary, then this case was handled above by
1884          * bond_change_active_slave(..., NULL)
1885          */
1886         if (!bond_uses_primary(bond)) {
1887                 /* unset promiscuity level from slave
1888                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1889                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1890                  * value of that flag before that change, as that was the value
1891                  * when this slave was attached, so we cache at the start of the
1892                  * function and use it here. Same goes for ALLMULTI below
1893                  */
1894                 if (old_flags & IFF_PROMISC)
1895                         dev_set_promiscuity(slave_dev, -1);
1896
1897                 /* unset allmulti level from slave */
1898                 if (old_flags & IFF_ALLMULTI)
1899                         dev_set_allmulti(slave_dev, -1);
1900
1901                 bond_hw_addr_flush(bond_dev, slave_dev);
1902         }
1903
1904         slave_disable_netpoll(slave);
1905
1906         /* close slave before restoring its mac address */
1907         dev_close(slave_dev);
1908
1909         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1910             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1911                 /* restore original ("permanent") mac address */
1912                 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1913                 addr.sa_family = slave_dev->type;
1914                 dev_set_mac_address(slave_dev, &addr);
1915         }
1916
1917         dev_set_mtu(slave_dev, slave->original_mtu);
1918
1919         slave_dev->priv_flags &= ~IFF_BONDING;
1920
1921         bond_free_slave(slave);
1922
1923         return 0;
1924 }
1925
1926 /* A wrapper used because of ndo_del_link */
1927 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1928 {
1929         return __bond_release_one(bond_dev, slave_dev, false);
1930 }
1931
1932 /* First release a slave and then destroy the bond if no more slaves are left.
1933  * Must be under rtnl_lock when this function is called.
1934  */
1935 static int  bond_release_and_destroy(struct net_device *bond_dev,
1936                                      struct net_device *slave_dev)
1937 {
1938         struct bonding *bond = netdev_priv(bond_dev);
1939         int ret;
1940
1941         ret = bond_release(bond_dev, slave_dev);
1942         if (ret == 0 && !bond_has_slaves(bond)) {
1943                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1944                 netdev_info(bond_dev, "Destroying bond %s\n",
1945                             bond_dev->name);
1946                 bond_remove_proc_entry(bond);
1947                 unregister_netdevice(bond_dev);
1948         }
1949         return ret;
1950 }
1951
1952 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1953 {
1954         struct bonding *bond = netdev_priv(bond_dev);
1955         bond_fill_ifbond(bond, info);
1956         return 0;
1957 }
1958
1959 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1960 {
1961         struct bonding *bond = netdev_priv(bond_dev);
1962         struct list_head *iter;
1963         int i = 0, res = -ENODEV;
1964         struct slave *slave;
1965
1966         bond_for_each_slave(bond, slave, iter) {
1967                 if (i++ == (int)info->slave_id) {
1968                         res = 0;
1969                         bond_fill_ifslave(slave, info);
1970                         break;
1971                 }
1972         }
1973
1974         return res;
1975 }
1976
1977 /*-------------------------------- Monitoring -------------------------------*/
1978
1979 /* called with rcu_read_lock() */
1980 static int bond_miimon_inspect(struct bonding *bond)
1981 {
1982         int link_state, commit = 0;
1983         struct list_head *iter;
1984         struct slave *slave;
1985         bool ignore_updelay;
1986
1987         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
1988
1989         bond_for_each_slave_rcu(bond, slave, iter) {
1990                 slave->new_link = BOND_LINK_NOCHANGE;
1991
1992                 link_state = bond_check_dev_link(bond, slave->dev, 0);
1993
1994                 switch (slave->link) {
1995                 case BOND_LINK_UP:
1996                         if (link_state)
1997                                 continue;
1998
1999                         bond_set_slave_link_state(slave, BOND_LINK_FAIL);
2000                         slave->delay = bond->params.downdelay;
2001                         if (slave->delay) {
2002                                 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2003                                             (BOND_MODE(bond) ==
2004                                              BOND_MODE_ACTIVEBACKUP) ?
2005                                              (bond_is_active_slave(slave) ?
2006                                               "active " : "backup ") : "",
2007                                             slave->dev->name,
2008                                             bond->params.downdelay * bond->params.miimon);
2009                         }
2010                         /*FALLTHRU*/
2011                 case BOND_LINK_FAIL:
2012                         if (link_state) {
2013                                 /* recovered before downdelay expired */
2014                                 bond_set_slave_link_state(slave, BOND_LINK_UP);
2015                                 slave->last_link_up = jiffies;
2016                                 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2017                                             (bond->params.downdelay - slave->delay) *
2018                                             bond->params.miimon,
2019                                             slave->dev->name);
2020                                 continue;
2021                         }
2022
2023                         if (slave->delay <= 0) {
2024                                 slave->new_link = BOND_LINK_DOWN;
2025                                 commit++;
2026                                 continue;
2027                         }
2028
2029                         slave->delay--;
2030                         break;
2031
2032                 case BOND_LINK_DOWN:
2033                         if (!link_state)
2034                                 continue;
2035
2036                         bond_set_slave_link_state(slave, BOND_LINK_BACK);
2037                         slave->delay = bond->params.updelay;
2038
2039                         if (slave->delay) {
2040                                 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2041                                             slave->dev->name,
2042                                             ignore_updelay ? 0 :
2043                                             bond->params.updelay *
2044                                             bond->params.miimon);
2045                         }
2046                         /*FALLTHRU*/
2047                 case BOND_LINK_BACK:
2048                         if (!link_state) {
2049                                 bond_set_slave_link_state(slave,
2050                                                           BOND_LINK_DOWN);
2051                                 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2052                                             (bond->params.updelay - slave->delay) *
2053                                             bond->params.miimon,
2054                                             slave->dev->name);
2055
2056                                 continue;
2057                         }
2058
2059                         if (ignore_updelay)
2060                                 slave->delay = 0;
2061
2062                         if (slave->delay <= 0) {
2063                                 slave->new_link = BOND_LINK_UP;
2064                                 commit++;
2065                                 ignore_updelay = false;
2066                                 continue;
2067                         }
2068
2069                         slave->delay--;
2070                         break;
2071                 }
2072         }
2073
2074         return commit;
2075 }
2076
2077 static void bond_miimon_commit(struct bonding *bond)
2078 {
2079         struct list_head *iter;
2080         struct slave *slave, *primary;
2081
2082         bond_for_each_slave(bond, slave, iter) {
2083                 switch (slave->new_link) {
2084                 case BOND_LINK_NOCHANGE:
2085                         continue;
2086
2087                 case BOND_LINK_UP:
2088                         bond_set_slave_link_state(slave, BOND_LINK_UP);
2089                         slave->last_link_up = jiffies;
2090
2091                         primary = rtnl_dereference(bond->primary_slave);
2092                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2093                                 /* prevent it from being the active one */
2094                                 bond_set_backup_slave(slave);
2095                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2096                                 /* make it immediately active */
2097                                 bond_set_active_slave(slave);
2098                         } else if (slave != primary) {
2099                                 /* prevent it from being the active one */
2100                                 bond_set_backup_slave(slave);
2101                         }
2102
2103                         netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2104                                     slave->dev->name,
2105                                     slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2106                                     slave->duplex ? "full" : "half");
2107
2108                         /* notify ad that the link status has changed */
2109                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2110                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2111
2112                         if (bond_is_lb(bond))
2113                                 bond_alb_handle_link_change(bond, slave,
2114                                                             BOND_LINK_UP);
2115
2116                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2117                                 bond_update_slave_arr(bond, NULL);
2118
2119                         if (!bond->curr_active_slave || slave == primary)
2120                                 goto do_failover;
2121
2122                         continue;
2123
2124                 case BOND_LINK_DOWN:
2125                         if (slave->link_failure_count < UINT_MAX)
2126                                 slave->link_failure_count++;
2127
2128                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2129
2130                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2131                             BOND_MODE(bond) == BOND_MODE_8023AD)
2132                                 bond_set_slave_inactive_flags(slave,
2133                                                               BOND_SLAVE_NOTIFY_NOW);
2134
2135                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2136                                     slave->dev->name);
2137
2138                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2139                                 bond_3ad_handle_link_change(slave,
2140                                                             BOND_LINK_DOWN);
2141
2142                         if (bond_is_lb(bond))
2143                                 bond_alb_handle_link_change(bond, slave,
2144                                                             BOND_LINK_DOWN);
2145
2146                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2147                                 bond_update_slave_arr(bond, NULL);
2148
2149                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2150                                 goto do_failover;
2151
2152                         continue;
2153
2154                 default:
2155                         netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2156                                    slave->new_link, slave->dev->name);
2157                         slave->new_link = BOND_LINK_NOCHANGE;
2158
2159                         continue;
2160                 }
2161
2162 do_failover:
2163                 block_netpoll_tx();
2164                 bond_select_active_slave(bond);
2165                 unblock_netpoll_tx();
2166         }
2167
2168         bond_set_carrier(bond);
2169 }
2170
2171 /* bond_mii_monitor
2172  *
2173  * Really a wrapper that splits the mii monitor into two phases: an
2174  * inspection, then (if inspection indicates something needs to be done)
2175  * an acquisition of appropriate locks followed by a commit phase to
2176  * implement whatever link state changes are indicated.
2177  */
2178 static void bond_mii_monitor(struct work_struct *work)
2179 {
2180         struct bonding *bond = container_of(work, struct bonding,
2181                                             mii_work.work);
2182         bool should_notify_peers = false;
2183         unsigned long delay;
2184
2185         delay = msecs_to_jiffies(bond->params.miimon);
2186
2187         if (!bond_has_slaves(bond))
2188                 goto re_arm;
2189
2190         rcu_read_lock();
2191
2192         should_notify_peers = bond_should_notify_peers(bond);
2193
2194         if (bond_miimon_inspect(bond)) {
2195                 rcu_read_unlock();
2196
2197                 /* Race avoidance with bond_close cancel of workqueue */
2198                 if (!rtnl_trylock()) {
2199                         delay = 1;
2200                         should_notify_peers = false;
2201                         goto re_arm;
2202                 }
2203
2204                 bond_miimon_commit(bond);
2205
2206                 rtnl_unlock();  /* might sleep, hold no other locks */
2207         } else
2208                 rcu_read_unlock();
2209
2210 re_arm:
2211         if (bond->params.miimon)
2212                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2213
2214         if (should_notify_peers) {
2215                 if (!rtnl_trylock())
2216                         return;
2217                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2218                 rtnl_unlock();
2219         }
2220 }
2221
2222 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2223 {
2224         struct net_device *upper;
2225         struct list_head *iter;
2226         bool ret = false;
2227
2228         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2229                 return true;
2230
2231         rcu_read_lock();
2232         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2233                 if (ip == bond_confirm_addr(upper, 0, ip)) {
2234                         ret = true;
2235                         break;
2236                 }
2237         }
2238         rcu_read_unlock();
2239
2240         return ret;
2241 }
2242
2243 /* We go to the (large) trouble of VLAN tagging ARP frames because
2244  * switches in VLAN mode (especially if ports are configured as
2245  * "native" to a VLAN) might not pass non-tagged frames.
2246  */
2247 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2248                           __be32 dest_ip, __be32 src_ip,
2249                           struct bond_vlan_tag *tags)
2250 {
2251         struct sk_buff *skb;
2252         struct bond_vlan_tag *outer_tag = tags;
2253
2254         netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2255                    arp_op, slave_dev->name, &dest_ip, &src_ip);
2256
2257         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2258                          NULL, slave_dev->dev_addr, NULL);
2259
2260         if (!skb) {
2261                 net_err_ratelimited("ARP packet allocation failed\n");
2262                 return;
2263         }
2264
2265         if (!tags || tags->vlan_proto == VLAN_N_VID)
2266                 goto xmit;
2267
2268         tags++;
2269
2270         /* Go through all the tags backwards and add them to the packet */
2271         while (tags->vlan_proto != VLAN_N_VID) {
2272                 if (!tags->vlan_id) {
2273                         tags++;
2274                         continue;
2275                 }
2276
2277                 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2278                            ntohs(outer_tag->vlan_proto), tags->vlan_id);
2279                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2280                                                 tags->vlan_id);
2281                 if (!skb) {
2282                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2283                         return;
2284                 }
2285
2286                 tags++;
2287         }
2288         /* Set the outer tag */
2289         if (outer_tag->vlan_id) {
2290                 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2291                            ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2292                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2293                                        outer_tag->vlan_id);
2294         }
2295
2296 xmit:
2297         arp_xmit(skb);
2298 }
2299
2300 /* Validate the device path between the @start_dev and the @end_dev.
2301  * The path is valid if the @end_dev is reachable through device
2302  * stacking.
2303  * When the path is validated, collect any vlan information in the
2304  * path.
2305  */
2306 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2307                                               struct net_device *end_dev,
2308                                               int level)
2309 {
2310         struct bond_vlan_tag *tags;
2311         struct net_device *upper;
2312         struct list_head  *iter;
2313
2314         if (start_dev == end_dev) {
2315                 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2316                 if (!tags)
2317                         return ERR_PTR(-ENOMEM);
2318                 tags[level].vlan_proto = VLAN_N_VID;
2319                 return tags;
2320         }
2321
2322         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2323                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2324                 if (IS_ERR_OR_NULL(tags)) {
2325                         if (IS_ERR(tags))
2326                                 return tags;
2327                         continue;
2328                 }
2329                 if (is_vlan_dev(upper)) {
2330                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2331                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2332                 }
2333
2334                 return tags;
2335         }
2336
2337         return NULL;
2338 }
2339
2340 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2341 {
2342         struct rtable *rt;
2343         struct bond_vlan_tag *tags;
2344         __be32 *targets = bond->params.arp_targets, addr;
2345         int i;
2346
2347         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2348                 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2349                 tags = NULL;
2350
2351                 /* Find out through which dev should the packet go */
2352                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2353                                      RTO_ONLINK, 0);
2354                 if (IS_ERR(rt)) {
2355                         /* there's no route to target - try to send arp
2356                          * probe to generate any traffic (arp_validate=0)
2357                          */
2358                         if (bond->params.arp_validate)
2359                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2360                                                      bond->dev->name,
2361                                                      &targets[i]);
2362                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2363                                       0, tags);
2364                         continue;
2365                 }
2366
2367                 /* bond device itself */
2368                 if (rt->dst.dev == bond->dev)
2369                         goto found;
2370
2371                 rcu_read_lock();
2372                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2373                 rcu_read_unlock();
2374
2375                 if (!IS_ERR_OR_NULL(tags))
2376                         goto found;
2377
2378                 /* Not our device - skip */
2379                 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2380                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2381
2382                 ip_rt_put(rt);
2383                 continue;
2384
2385 found:
2386                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2387                 ip_rt_put(rt);
2388                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2389                               addr, tags);
2390                 kfree(tags);
2391         }
2392 }
2393
2394 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2395 {
2396         int i;
2397
2398         if (!sip || !bond_has_this_ip(bond, tip)) {
2399                 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2400                            &sip, &tip);
2401                 return;
2402         }
2403
2404         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2405         if (i == -1) {
2406                 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2407                            &sip);
2408                 return;
2409         }
2410         slave->last_rx = jiffies;
2411         slave->target_last_arp_rx[i] = jiffies;
2412 }
2413
2414 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2415                  struct slave *slave)
2416 {
2417         struct arphdr *arp = (struct arphdr *)skb->data;
2418         struct slave *curr_active_slave;
2419         unsigned char *arp_ptr;
2420         __be32 sip, tip;
2421         int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2422
2423         if (!slave_do_arp_validate(bond, slave)) {
2424                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2425                     !slave_do_arp_validate_only(bond))
2426                         slave->last_rx = jiffies;
2427                 return RX_HANDLER_ANOTHER;
2428         } else if (!is_arp) {
2429                 return RX_HANDLER_ANOTHER;
2430         }
2431
2432         alen = arp_hdr_len(bond->dev);
2433
2434         netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2435                    skb->dev->name);
2436
2437         if (alen > skb_headlen(skb)) {
2438                 arp = kmalloc(alen, GFP_ATOMIC);
2439                 if (!arp)
2440                         goto out_unlock;
2441                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2442                         goto out_unlock;
2443         }
2444
2445         if (arp->ar_hln != bond->dev->addr_len ||
2446             skb->pkt_type == PACKET_OTHERHOST ||
2447             skb->pkt_type == PACKET_LOOPBACK ||
2448             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2449             arp->ar_pro != htons(ETH_P_IP) ||
2450             arp->ar_pln != 4)
2451                 goto out_unlock;
2452
2453         arp_ptr = (unsigned char *)(arp + 1);
2454         arp_ptr += bond->dev->addr_len;
2455         memcpy(&sip, arp_ptr, 4);
2456         arp_ptr += 4 + bond->dev->addr_len;
2457         memcpy(&tip, arp_ptr, 4);
2458
2459         netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2460                    slave->dev->name, bond_slave_state(slave),
2461                      bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2462                      &sip, &tip);
2463
2464         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2465
2466         /* Backup slaves won't see the ARP reply, but do come through
2467          * here for each ARP probe (so we swap the sip/tip to validate
2468          * the probe).  In a "redundant switch, common router" type of
2469          * configuration, the ARP probe will (hopefully) travel from
2470          * the active, through one switch, the router, then the other
2471          * switch before reaching the backup.
2472          *
2473          * We 'trust' the arp requests if there is an active slave and
2474          * it received valid arp reply(s) after it became active. This
2475          * is done to avoid endless looping when we can't reach the
2476          * arp_ip_target and fool ourselves with our own arp requests.
2477          */
2478
2479         if (bond_is_active_slave(slave))
2480                 bond_validate_arp(bond, slave, sip, tip);
2481         else if (curr_active_slave &&
2482                  time_after(slave_last_rx(bond, curr_active_slave),
2483                             curr_active_slave->last_link_up))
2484                 bond_validate_arp(bond, slave, tip, sip);
2485
2486 out_unlock:
2487         if (arp != (struct arphdr *)skb->data)
2488                 kfree(arp);
2489         return RX_HANDLER_ANOTHER;
2490 }
2491
2492 /* function to verify if we're in the arp_interval timeslice, returns true if
2493  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2494  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2495  */
2496 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2497                                   int mod)
2498 {
2499         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2500
2501         return time_in_range(jiffies,
2502                              last_act - delta_in_ticks,
2503                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2504 }
2505
2506 /* This function is called regularly to monitor each slave's link
2507  * ensuring that traffic is being sent and received when arp monitoring
2508  * is used in load-balancing mode. if the adapter has been dormant, then an
2509  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2510  * arp monitoring in active backup mode.
2511  */
2512 static void bond_loadbalance_arp_mon(struct work_struct *work)
2513 {
2514         struct bonding *bond = container_of(work, struct bonding,
2515                                             arp_work.work);
2516         struct slave *slave, *oldcurrent;
2517         struct list_head *iter;
2518         int do_failover = 0, slave_state_changed = 0;
2519
2520         if (!bond_has_slaves(bond))
2521                 goto re_arm;
2522
2523         rcu_read_lock();
2524
2525         oldcurrent = rcu_dereference(bond->curr_active_slave);
2526         /* see if any of the previous devices are up now (i.e. they have
2527          * xmt and rcv traffic). the curr_active_slave does not come into
2528          * the picture unless it is null. also, slave->last_link_up is not
2529          * needed here because we send an arp on each slave and give a slave
2530          * as long as it needs to get the tx/rx within the delta.
2531          * TODO: what about up/down delay in arp mode? it wasn't here before
2532          *       so it can wait
2533          */
2534         bond_for_each_slave_rcu(bond, slave, iter) {
2535                 unsigned long trans_start = dev_trans_start(slave->dev);
2536
2537                 if (slave->link != BOND_LINK_UP) {
2538                         if (bond_time_in_interval(bond, trans_start, 1) &&
2539                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2540
2541                                 slave->link  = BOND_LINK_UP;
2542                                 slave_state_changed = 1;
2543
2544                                 /* primary_slave has no meaning in round-robin
2545                                  * mode. the window of a slave being up and
2546                                  * curr_active_slave being null after enslaving
2547                                  * is closed.
2548                                  */
2549                                 if (!oldcurrent) {
2550                                         netdev_info(bond->dev, "link status definitely up for interface %s\n",
2551                                                     slave->dev->name);
2552                                         do_failover = 1;
2553                                 } else {
2554                                         netdev_info(bond->dev, "interface %s is now up\n",
2555                                                     slave->dev->name);
2556                                 }
2557                         }
2558                 } else {
2559                         /* slave->link == BOND_LINK_UP */
2560
2561                         /* not all switches will respond to an arp request
2562                          * when the source ip is 0, so don't take the link down
2563                          * if we don't know our ip yet
2564                          */
2565                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2566                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2567
2568                                 slave->link  = BOND_LINK_DOWN;
2569                                 slave_state_changed = 1;
2570
2571                                 if (slave->link_failure_count < UINT_MAX)
2572                                         slave->link_failure_count++;
2573
2574                                 netdev_info(bond->dev, "interface %s is now down\n",
2575                                             slave->dev->name);
2576
2577                                 if (slave == oldcurrent)
2578                                         do_failover = 1;
2579                         }
2580                 }
2581
2582                 /* note: if switch is in round-robin mode, all links
2583                  * must tx arp to ensure all links rx an arp - otherwise
2584                  * links may oscillate or not come up at all; if switch is
2585                  * in something like xor mode, there is nothing we can
2586                  * do - all replies will be rx'ed on same link causing slaves
2587                  * to be unstable during low/no traffic periods
2588                  */
2589                 if (bond_slave_is_up(slave))
2590                         bond_arp_send_all(bond, slave);
2591         }
2592
2593         rcu_read_unlock();
2594
2595         if (do_failover || slave_state_changed) {
2596                 if (!rtnl_trylock())
2597                         goto re_arm;
2598
2599                 if (slave_state_changed) {
2600                         bond_slave_state_change(bond);
2601                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2602                                 bond_update_slave_arr(bond, NULL);
2603                 }
2604                 if (do_failover) {
2605                         block_netpoll_tx();
2606                         bond_select_active_slave(bond);
2607                         unblock_netpoll_tx();
2608                 }
2609                 rtnl_unlock();
2610         }
2611
2612 re_arm:
2613         if (bond->params.arp_interval)
2614                 queue_delayed_work(bond->wq, &bond->arp_work,
2615                                    msecs_to_jiffies(bond->params.arp_interval));
2616 }
2617
2618 /* Called to inspect slaves for active-backup mode ARP monitor link state
2619  * changes.  Sets new_link in slaves to specify what action should take
2620  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2621  * to link states must be committed.
2622  *
2623  * Called with rcu_read_lock held.
2624  */
2625 static int bond_ab_arp_inspect(struct bonding *bond)
2626 {
2627         unsigned long trans_start, last_rx;
2628         struct list_head *iter;
2629         struct slave *slave;
2630         int commit = 0;
2631
2632         bond_for_each_slave_rcu(bond, slave, iter) {
2633                 slave->new_link = BOND_LINK_NOCHANGE;
2634                 last_rx = slave_last_rx(bond, slave);
2635
2636                 if (slave->link != BOND_LINK_UP) {
2637                         if (bond_time_in_interval(bond, last_rx, 1)) {
2638                                 slave->new_link = BOND_LINK_UP;
2639                                 commit++;
2640                         }
2641                         continue;
2642                 }
2643
2644                 /* Give slaves 2*delta after being enslaved or made
2645                  * active.  This avoids bouncing, as the last receive
2646                  * times need a full ARP monitor cycle to be updated.
2647                  */
2648                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2649                         continue;
2650
2651                 /* Backup slave is down if:
2652                  * - No current_arp_slave AND
2653                  * - more than 3*delta since last receive AND
2654                  * - the bond has an IP address
2655                  *
2656                  * Note: a non-null current_arp_slave indicates
2657                  * the curr_active_slave went down and we are
2658                  * searching for a new one; under this condition
2659                  * we only take the curr_active_slave down - this
2660                  * gives each slave a chance to tx/rx traffic
2661                  * before being taken out
2662                  */
2663                 if (!bond_is_active_slave(slave) &&
2664                     !rcu_access_pointer(bond->current_arp_slave) &&
2665                     !bond_time_in_interval(bond, last_rx, 3)) {
2666                         slave->new_link = BOND_LINK_DOWN;
2667                         commit++;
2668                 }
2669
2670                 /* Active slave is down if:
2671                  * - more than 2*delta since transmitting OR
2672                  * - (more than 2*delta since receive AND
2673                  *    the bond has an IP address)
2674                  */
2675                 trans_start = dev_trans_start(slave->dev);
2676                 if (bond_is_active_slave(slave) &&
2677                     (!bond_time_in_interval(bond, trans_start, 2) ||
2678                      !bond_time_in_interval(bond, last_rx, 2))) {
2679                         slave->new_link = BOND_LINK_DOWN;
2680                         commit++;
2681                 }
2682         }
2683
2684         return commit;
2685 }
2686
2687 /* Called to commit link state changes noted by inspection step of
2688  * active-backup mode ARP monitor.
2689  *
2690  * Called with RTNL hold.
2691  */
2692 static void bond_ab_arp_commit(struct bonding *bond)
2693 {
2694         unsigned long trans_start;
2695         struct list_head *iter;
2696         struct slave *slave;
2697
2698         bond_for_each_slave(bond, slave, iter) {
2699                 switch (slave->new_link) {
2700                 case BOND_LINK_NOCHANGE:
2701                         continue;
2702
2703                 case BOND_LINK_UP:
2704                         trans_start = dev_trans_start(slave->dev);
2705                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
2706                             (!rtnl_dereference(bond->curr_active_slave) &&
2707                              bond_time_in_interval(bond, trans_start, 1))) {
2708                                 struct slave *current_arp_slave;
2709
2710                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2711                                 bond_set_slave_link_state(slave, BOND_LINK_UP);
2712                                 if (current_arp_slave) {
2713                                         bond_set_slave_inactive_flags(
2714                                                 current_arp_slave,
2715                                                 BOND_SLAVE_NOTIFY_NOW);
2716                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2717                                 }
2718
2719                                 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2720                                             slave->dev->name);
2721
2722                                 if (!rtnl_dereference(bond->curr_active_slave) ||
2723                                     slave == rtnl_dereference(bond->primary_slave))
2724                                         goto do_failover;
2725
2726                         }
2727
2728                         continue;
2729
2730                 case BOND_LINK_DOWN:
2731                         if (slave->link_failure_count < UINT_MAX)
2732                                 slave->link_failure_count++;
2733
2734                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2735                         bond_set_slave_inactive_flags(slave,
2736                                                       BOND_SLAVE_NOTIFY_NOW);
2737
2738                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2739                                     slave->dev->name);
2740
2741                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
2742                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2743                                 goto do_failover;
2744                         }
2745
2746                         continue;
2747
2748                 default:
2749                         netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2750                                    slave->new_link, slave->dev->name);
2751                         continue;
2752                 }
2753
2754 do_failover:
2755                 block_netpoll_tx();
2756                 bond_select_active_slave(bond);
2757                 unblock_netpoll_tx();
2758         }
2759
2760         bond_set_carrier(bond);
2761 }
2762
2763 /* Send ARP probes for active-backup mode ARP monitor.
2764  *
2765  * Called with rcu_read_lock held.
2766  */
2767 static bool bond_ab_arp_probe(struct bonding *bond)
2768 {
2769         struct slave *slave, *before = NULL, *new_slave = NULL,
2770                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2771                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2772         struct list_head *iter;
2773         bool found = false;
2774         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2775
2776         if (curr_arp_slave && curr_active_slave)
2777                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2778                             curr_arp_slave->dev->name,
2779                             curr_active_slave->dev->name);
2780
2781         if (curr_active_slave) {
2782                 bond_arp_send_all(bond, curr_active_slave);
2783                 return should_notify_rtnl;
2784         }
2785
2786         /* if we don't have a curr_active_slave, search for the next available
2787          * backup slave from the current_arp_slave and make it the candidate
2788          * for becoming the curr_active_slave
2789          */
2790
2791         if (!curr_arp_slave) {
2792                 curr_arp_slave = bond_first_slave_rcu(bond);
2793                 if (!curr_arp_slave)
2794                         return should_notify_rtnl;
2795         }
2796
2797         bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2798
2799         bond_for_each_slave_rcu(bond, slave, iter) {
2800                 if (!found && !before && bond_slave_is_up(slave))
2801                         before = slave;
2802
2803                 if (found && !new_slave && bond_slave_is_up(slave))
2804                         new_slave = slave;
2805                 /* if the link state is up at this point, we
2806                  * mark it down - this can happen if we have
2807                  * simultaneous link failures and
2808                  * reselect_active_interface doesn't make this
2809                  * one the current slave so it is still marked
2810                  * up when it is actually down
2811                  */
2812                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2813                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2814                         if (slave->link_failure_count < UINT_MAX)
2815                                 slave->link_failure_count++;
2816
2817                         bond_set_slave_inactive_flags(slave,
2818                                                       BOND_SLAVE_NOTIFY_LATER);
2819
2820                         netdev_info(bond->dev, "backup interface %s is now down\n",
2821                                     slave->dev->name);
2822                 }
2823                 if (slave == curr_arp_slave)
2824                         found = true;
2825         }
2826
2827         if (!new_slave && before)
2828                 new_slave = before;
2829
2830         if (!new_slave)
2831                 goto check_state;
2832
2833         bond_set_slave_link_state(new_slave, BOND_LINK_BACK);
2834         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2835         bond_arp_send_all(bond, new_slave);
2836         new_slave->last_link_up = jiffies;
2837         rcu_assign_pointer(bond->current_arp_slave, new_slave);
2838
2839 check_state:
2840         bond_for_each_slave_rcu(bond, slave, iter) {
2841                 if (slave->should_notify) {
2842                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2843                         break;
2844                 }
2845         }
2846         return should_notify_rtnl;
2847 }
2848
2849 static void bond_activebackup_arp_mon(struct work_struct *work)
2850 {
2851         struct bonding *bond = container_of(work, struct bonding,
2852                                             arp_work.work);
2853         bool should_notify_peers = false;
2854         bool should_notify_rtnl = false;
2855         int delta_in_ticks;
2856
2857         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2858
2859         if (!bond_has_slaves(bond))
2860                 goto re_arm;
2861
2862         rcu_read_lock();
2863
2864         should_notify_peers = bond_should_notify_peers(bond);
2865
2866         if (bond_ab_arp_inspect(bond)) {
2867                 rcu_read_unlock();
2868
2869                 /* Race avoidance with bond_close flush of workqueue */
2870                 if (!rtnl_trylock()) {
2871                         delta_in_ticks = 1;
2872                         should_notify_peers = false;
2873                         goto re_arm;
2874                 }
2875
2876                 bond_ab_arp_commit(bond);
2877
2878                 rtnl_unlock();
2879                 rcu_read_lock();
2880         }
2881
2882         should_notify_rtnl = bond_ab_arp_probe(bond);
2883         rcu_read_unlock();
2884
2885 re_arm:
2886         if (bond->params.arp_interval)
2887                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2888
2889         if (should_notify_peers || should_notify_rtnl) {
2890                 if (!rtnl_trylock())
2891                         return;
2892
2893                 if (should_notify_peers)
2894                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2895                                                  bond->dev);
2896                 if (should_notify_rtnl)
2897                         bond_slave_state_notify(bond);
2898
2899                 rtnl_unlock();
2900         }
2901 }
2902
2903 /*-------------------------- netdev event handling --------------------------*/
2904
2905 /* Change device name */
2906 static int bond_event_changename(struct bonding *bond)
2907 {
2908         bond_remove_proc_entry(bond);
2909         bond_create_proc_entry(bond);
2910
2911         bond_debug_reregister(bond);
2912
2913         return NOTIFY_DONE;
2914 }
2915
2916 static int bond_master_netdev_event(unsigned long event,
2917                                     struct net_device *bond_dev)
2918 {
2919         struct bonding *event_bond = netdev_priv(bond_dev);
2920
2921         switch (event) {
2922         case NETDEV_CHANGENAME:
2923                 return bond_event_changename(event_bond);
2924         case NETDEV_UNREGISTER:
2925                 bond_remove_proc_entry(event_bond);
2926                 break;
2927         case NETDEV_REGISTER:
2928                 bond_create_proc_entry(event_bond);
2929                 break;
2930         case NETDEV_NOTIFY_PEERS:
2931                 if (event_bond->send_peer_notif)
2932                         event_bond->send_peer_notif--;
2933                 break;
2934         default:
2935                 break;
2936         }
2937
2938         return NOTIFY_DONE;
2939 }
2940
2941 static int bond_slave_netdev_event(unsigned long event,
2942                                    struct net_device *slave_dev)
2943 {
2944         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
2945         struct bonding *bond;
2946         struct net_device *bond_dev;
2947         u32 old_speed;
2948         u8 old_duplex;
2949
2950         /* A netdev event can be generated while enslaving a device
2951          * before netdev_rx_handler_register is called in which case
2952          * slave will be NULL
2953          */
2954         if (!slave)
2955                 return NOTIFY_DONE;
2956         bond_dev = slave->bond->dev;
2957         bond = slave->bond;
2958         primary = rtnl_dereference(bond->primary_slave);
2959
2960         switch (event) {
2961         case NETDEV_UNREGISTER:
2962                 if (bond_dev->type != ARPHRD_ETHER)
2963                         bond_release_and_destroy(bond_dev, slave_dev);
2964                 else
2965                         bond_release(bond_dev, slave_dev);
2966                 break;
2967         case NETDEV_UP:
2968         case NETDEV_CHANGE:
2969                 old_speed = slave->speed;
2970                 old_duplex = slave->duplex;
2971
2972                 bond_update_speed_duplex(slave);
2973
2974                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2975                         if (old_speed != slave->speed)
2976                                 bond_3ad_adapter_speed_changed(slave);
2977                         if (old_duplex != slave->duplex)
2978                                 bond_3ad_adapter_duplex_changed(slave);
2979                 }
2980                 /* Fallthrough */
2981         case NETDEV_DOWN:
2982                 /* Refresh slave-array if applicable!
2983                  * If the setup does not use miimon or arpmon (mode-specific!),
2984                  * then these events will not cause the slave-array to be
2985                  * refreshed. This will cause xmit to use a slave that is not
2986                  * usable. Avoid such situation by refeshing the array at these
2987                  * events. If these (miimon/arpmon) parameters are configured
2988                  * then array gets refreshed twice and that should be fine!
2989                  */
2990                 if (bond_mode_uses_xmit_hash(bond))
2991                         bond_update_slave_arr(bond, NULL);
2992                 break;
2993         case NETDEV_CHANGEMTU:
2994                 /* TODO: Should slaves be allowed to
2995                  * independently alter their MTU?  For
2996                  * an active-backup bond, slaves need
2997                  * not be the same type of device, so
2998                  * MTUs may vary.  For other modes,
2999                  * slaves arguably should have the
3000                  * same MTUs. To do this, we'd need to
3001                  * take over the slave's change_mtu
3002                  * function for the duration of their
3003                  * servitude.
3004                  */
3005                 break;
3006         case NETDEV_CHANGENAME:
3007                 /* we don't care if we don't have primary set */
3008                 if (!bond_uses_primary(bond) ||
3009                     !bond->params.primary[0])
3010                         break;
3011
3012                 if (slave == primary) {
3013                         /* slave's name changed - he's no longer primary */
3014                         RCU_INIT_POINTER(bond->primary_slave, NULL);
3015                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3016                         /* we have a new primary slave */
3017                         rcu_assign_pointer(bond->primary_slave, slave);
3018                 } else { /* we didn't change primary - exit */
3019                         break;
3020                 }
3021
3022                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3023                             primary ? slave_dev->name : "none");
3024
3025                 block_netpoll_tx();
3026                 bond_select_active_slave(bond);
3027                 unblock_netpoll_tx();
3028                 break;
3029         case NETDEV_FEAT_CHANGE:
3030                 bond_compute_features(bond);
3031                 break;
3032         case NETDEV_RESEND_IGMP:
3033                 /* Propagate to master device */
3034                 call_netdevice_notifiers(event, slave->bond->dev);
3035                 break;
3036         default:
3037                 break;
3038         }
3039
3040         return NOTIFY_DONE;
3041 }
3042
3043 /* bond_netdev_event: handle netdev notifier chain events.
3044  *
3045  * This function receives events for the netdev chain.  The caller (an
3046  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3047  * locks for us to safely manipulate the slave devices (RTNL lock,
3048  * dev_probe_lock).
3049  */
3050 static int bond_netdev_event(struct notifier_block *this,
3051                              unsigned long event, void *ptr)
3052 {
3053         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3054
3055         netdev_dbg(event_dev, "event: %lx\n", event);
3056
3057         if (!(event_dev->priv_flags & IFF_BONDING))
3058                 return NOTIFY_DONE;
3059
3060         if (event_dev->flags & IFF_MASTER) {
3061                 netdev_dbg(event_dev, "IFF_MASTER\n");
3062                 return bond_master_netdev_event(event, event_dev);
3063         }
3064
3065         if (event_dev->flags & IFF_SLAVE) {
3066                 netdev_dbg(event_dev, "IFF_SLAVE\n");
3067                 return bond_slave_netdev_event(event, event_dev);
3068         }
3069
3070         return NOTIFY_DONE;
3071 }
3072
3073 static struct notifier_block bond_netdev_notifier = {
3074         .notifier_call = bond_netdev_event,
3075 };
3076
3077 /*---------------------------- Hashing Policies -----------------------------*/
3078
3079 /* L2 hash helper */
3080 static inline u32 bond_eth_hash(struct sk_buff *skb)
3081 {
3082         struct ethhdr *ep, hdr_tmp;
3083
3084         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3085         if (ep)
3086                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3087         return 0;
3088 }
3089
3090 /* Extract the appropriate headers based on bond's xmit policy */
3091 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3092                               struct flow_keys *fk)
3093 {
3094         const struct ipv6hdr *iph6;
3095         const struct iphdr *iph;
3096         int noff, proto = -1;
3097
3098         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3099                 return skb_flow_dissect_flow_keys(skb, fk);
3100
3101         fk->ports.ports = 0;
3102         noff = skb_network_offset(skb);
3103         if (skb->protocol == htons(ETH_P_IP)) {
3104                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3105                         return false;
3106                 iph = ip_hdr(skb);
3107                 iph_to_flow_copy_v4addrs(fk, iph);
3108                 noff += iph->ihl << 2;
3109                 if (!ip_is_fragment(iph))
3110                         proto = iph->protocol;
3111         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3112                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3113                         return false;
3114                 iph6 = ipv6_hdr(skb);
3115                 iph_to_flow_copy_v6addrs(fk, iph6);
3116                 noff += sizeof(*iph6);
3117                 proto = iph6->nexthdr;
3118         } else {
3119                 return false;
3120         }
3121         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3122                 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3123
3124         return true;
3125 }
3126
3127 /**
3128  * bond_xmit_hash - generate a hash value based on the xmit policy
3129  * @bond: bonding device
3130  * @skb: buffer to use for headers
3131  *
3132  * This function will extract the necessary headers from the skb buffer and use
3133  * them to generate a hash based on the xmit_policy set in the bonding device
3134  */
3135 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3136 {
3137         struct flow_keys flow;
3138         u32 hash;
3139
3140         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3141             !bond_flow_dissect(bond, skb, &flow))
3142                 return bond_eth_hash(skb);
3143
3144         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3145             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3146                 hash = bond_eth_hash(skb);
3147         else
3148                 hash = (__force u32)flow.ports.ports;
3149         hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3150                 (__force u32)flow_get_u32_src(&flow);
3151         hash ^= (hash >> 16);
3152         hash ^= (hash >> 8);
3153
3154         return hash;
3155 }
3156
3157 /*-------------------------- Device entry points ----------------------------*/
3158
3159 static void bond_work_init_all(struct bonding *bond)
3160 {
3161         INIT_DELAYED_WORK(&bond->mcast_work,
3162                           bond_resend_igmp_join_requests_delayed);
3163         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3164         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3165         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3166                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3167         else
3168                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3169         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3170         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3171 }
3172
3173 static void bond_work_cancel_all(struct bonding *bond)
3174 {
3175         cancel_delayed_work_sync(&bond->mii_work);
3176         cancel_delayed_work_sync(&bond->arp_work);
3177         cancel_delayed_work_sync(&bond->alb_work);
3178         cancel_delayed_work_sync(&bond->ad_work);
3179         cancel_delayed_work_sync(&bond->mcast_work);
3180         cancel_delayed_work_sync(&bond->slave_arr_work);
3181 }
3182
3183 static int bond_open(struct net_device *bond_dev)
3184 {
3185         struct bonding *bond = netdev_priv(bond_dev);
3186         struct list_head *iter;
3187         struct slave *slave;
3188
3189         /* reset slave->backup and slave->inactive */
3190         if (bond_has_slaves(bond)) {
3191                 bond_for_each_slave(bond, slave, iter) {
3192                         if (bond_uses_primary(bond) &&
3193                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3194                                 bond_set_slave_inactive_flags(slave,
3195                                                               BOND_SLAVE_NOTIFY_NOW);
3196                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3197                                 bond_set_slave_active_flags(slave,
3198                                                             BOND_SLAVE_NOTIFY_NOW);
3199                         }
3200                 }
3201         }
3202
3203         bond_work_init_all(bond);
3204
3205         if (bond_is_lb(bond)) {
3206                 /* bond_alb_initialize must be called before the timer
3207                  * is started.
3208                  */
3209                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3210                         return -ENOMEM;
3211                 if (bond->params.tlb_dynamic_lb)
3212                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3213         }
3214
3215         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3216                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3217
3218         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3219                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3220                 bond->recv_probe = bond_arp_rcv;
3221         }
3222
3223         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3224                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3225                 /* register to receive LACPDUs */
3226                 bond->recv_probe = bond_3ad_lacpdu_recv;
3227                 bond_3ad_initiate_agg_selection(bond, 1);
3228         }
3229
3230         if (bond_mode_uses_xmit_hash(bond))
3231                 bond_update_slave_arr(bond, NULL);
3232
3233         return 0;
3234 }
3235
3236 static int bond_close(struct net_device *bond_dev)
3237 {
3238         struct bonding *bond = netdev_priv(bond_dev);
3239
3240         bond_work_cancel_all(bond);
3241         bond->send_peer_notif = 0;
3242         if (bond_is_lb(bond))
3243                 bond_alb_deinitialize(bond);
3244         bond->recv_probe = NULL;
3245
3246         return 0;
3247 }
3248
3249 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3250                                                 struct rtnl_link_stats64 *stats)
3251 {
3252         struct bonding *bond = netdev_priv(bond_dev);
3253         struct rtnl_link_stats64 temp;
3254         struct list_head *iter;
3255         struct slave *slave;
3256
3257         memcpy(stats, &bond->bond_stats, sizeof(*stats));
3258
3259         bond_for_each_slave(bond, slave, iter) {
3260                 const struct rtnl_link_stats64 *sstats =
3261                         dev_get_stats(slave->dev, &temp);
3262                 struct rtnl_link_stats64 *pstats = &slave->slave_stats;
3263
3264                 stats->rx_packets +=  sstats->rx_packets - pstats->rx_packets;
3265                 stats->rx_bytes += sstats->rx_bytes - pstats->rx_bytes;
3266                 stats->rx_errors += sstats->rx_errors - pstats->rx_errors;
3267                 stats->rx_dropped += sstats->rx_dropped - pstats->rx_dropped;
3268
3269                 stats->tx_packets += sstats->tx_packets - pstats->tx_packets;;
3270                 stats->tx_bytes += sstats->tx_bytes - pstats->tx_bytes;
3271                 stats->tx_errors += sstats->tx_errors - pstats->tx_errors;
3272                 stats->tx_dropped += sstats->tx_dropped - pstats->tx_dropped;
3273
3274                 stats->multicast += sstats->multicast - pstats->multicast;
3275                 stats->collisions += sstats->collisions - pstats->collisions;
3276
3277                 stats->rx_length_errors += sstats->rx_length_errors - pstats->rx_length_errors;
3278                 stats->rx_over_errors += sstats->rx_over_errors - pstats->rx_over_errors;
3279                 stats->rx_crc_errors += sstats->rx_crc_errors - pstats->rx_crc_errors;
3280                 stats->rx_frame_errors += sstats->rx_frame_errors - pstats->rx_frame_errors;
3281                 stats->rx_fifo_errors += sstats->rx_fifo_errors - pstats->rx_fifo_errors;
3282                 stats->rx_missed_errors += sstats->rx_missed_errors - pstats->rx_missed_errors;
3283
3284                 stats->tx_aborted_errors += sstats->tx_aborted_errors - pstats->tx_aborted_errors;
3285                 stats->tx_carrier_errors += sstats->tx_carrier_errors - pstats->tx_carrier_errors;
3286                 stats->tx_fifo_errors += sstats->tx_fifo_errors - pstats->tx_fifo_errors;
3287                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors - pstats->tx_heartbeat_errors;
3288                 stats->tx_window_errors += sstats->tx_window_errors - pstats->tx_window_errors;
3289
3290                 /* save off the slave stats for the next run */
3291                 memcpy(pstats, sstats, sizeof(*sstats));
3292         }
3293         memcpy(&bond->bond_stats, stats, sizeof(*stats));
3294
3295         return stats;
3296 }
3297
3298 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3299 {
3300         struct bonding *bond = netdev_priv(bond_dev);
3301         struct net_device *slave_dev = NULL;
3302         struct ifbond k_binfo;
3303         struct ifbond __user *u_binfo = NULL;
3304         struct ifslave k_sinfo;
3305         struct ifslave __user *u_sinfo = NULL;
3306         struct mii_ioctl_data *mii = NULL;
3307         struct bond_opt_value newval;
3308         struct net *net;
3309         int res = 0;
3310
3311         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3312
3313         switch (cmd) {
3314         case SIOCGMIIPHY:
3315                 mii = if_mii(ifr);
3316                 if (!mii)
3317                         return -EINVAL;
3318
3319                 mii->phy_id = 0;
3320                 /* Fall Through */
3321         case SIOCGMIIREG:
3322                 /* We do this again just in case we were called by SIOCGMIIREG
3323                  * instead of SIOCGMIIPHY.
3324                  */
3325                 mii = if_mii(ifr);
3326                 if (!mii)
3327                         return -EINVAL;
3328
3329                 if (mii->reg_num == 1) {
3330                         mii->val_out = 0;
3331                         if (netif_carrier_ok(bond->dev))
3332                                 mii->val_out = BMSR_LSTATUS;
3333                 }
3334
3335                 return 0;
3336         case BOND_INFO_QUERY_OLD:
3337         case SIOCBONDINFOQUERY:
3338                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3339
3340                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3341                         return -EFAULT;
3342
3343                 res = bond_info_query(bond_dev, &k_binfo);
3344                 if (res == 0 &&
3345                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3346                         return -EFAULT;
3347
3348                 return res;
3349         case BOND_SLAVE_INFO_QUERY_OLD:
3350         case SIOCBONDSLAVEINFOQUERY:
3351                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3352
3353                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3354                         return -EFAULT;
3355
3356                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3357                 if (res == 0 &&
3358                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3359                         return -EFAULT;
3360
3361                 return res;
3362         default:
3363                 break;
3364         }
3365
3366         net = dev_net(bond_dev);
3367
3368         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3369                 return -EPERM;
3370
3371         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3372
3373         netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3374
3375         if (!slave_dev)
3376                 return -ENODEV;
3377
3378         netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3379         switch (cmd) {
3380         case BOND_ENSLAVE_OLD:
3381         case SIOCBONDENSLAVE:
3382                 res = bond_enslave(bond_dev, slave_dev);
3383                 break;
3384         case BOND_RELEASE_OLD:
3385         case SIOCBONDRELEASE:
3386                 res = bond_release(bond_dev, slave_dev);
3387                 break;
3388         case BOND_SETHWADDR_OLD:
3389         case SIOCBONDSETHWADDR:
3390                 bond_set_dev_addr(bond_dev, slave_dev);
3391                 res = 0;
3392                 break;
3393         case BOND_CHANGE_ACTIVE_OLD:
3394         case SIOCBONDCHANGEACTIVE:
3395                 bond_opt_initstr(&newval, slave_dev->name);
3396                 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3397                 break;
3398         default:
3399                 res = -EOPNOTSUPP;
3400         }
3401
3402         return res;
3403 }
3404
3405 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3406 {
3407         struct bonding *bond = netdev_priv(bond_dev);
3408
3409         if (change & IFF_PROMISC)
3410                 bond_set_promiscuity(bond,
3411                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3412
3413         if (change & IFF_ALLMULTI)
3414                 bond_set_allmulti(bond,
3415                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3416 }
3417
3418 static void bond_set_rx_mode(struct net_device *bond_dev)
3419 {
3420         struct bonding *bond = netdev_priv(bond_dev);
3421         struct list_head *iter;
3422         struct slave *slave;
3423
3424         rcu_read_lock();
3425         if (bond_uses_primary(bond)) {
3426                 slave = rcu_dereference(bond->curr_active_slave);
3427                 if (slave) {
3428                         dev_uc_sync(slave->dev, bond_dev);
3429                         dev_mc_sync(slave->dev, bond_dev);
3430                 }
3431         } else {
3432                 bond_for_each_slave_rcu(bond, slave, iter) {
3433                         dev_uc_sync_multiple(slave->dev, bond_dev);
3434                         dev_mc_sync_multiple(slave->dev, bond_dev);
3435                 }
3436         }
3437         rcu_read_unlock();
3438 }
3439
3440 static int bond_neigh_init(struct neighbour *n)
3441 {
3442         struct bonding *bond = netdev_priv(n->dev);
3443         const struct net_device_ops *slave_ops;
3444         struct neigh_parms parms;
3445         struct slave *slave;
3446         int ret;
3447
3448         slave = bond_first_slave(bond);
3449         if (!slave)
3450                 return 0;
3451         slave_ops = slave->dev->netdev_ops;
3452         if (!slave_ops->ndo_neigh_setup)
3453                 return 0;
3454
3455         parms.neigh_setup = NULL;
3456         parms.neigh_cleanup = NULL;
3457         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3458         if (ret)
3459                 return ret;
3460
3461         /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3462          * after the last slave has been detached.  Assumes that all slaves
3463          * utilize the same neigh_cleanup (true at this writing as only user
3464          * is ipoib).
3465          */
3466         n->parms->neigh_cleanup = parms.neigh_cleanup;
3467
3468         if (!parms.neigh_setup)
3469                 return 0;
3470
3471         return parms.neigh_setup(n);
3472 }
3473
3474 /* The bonding ndo_neigh_setup is called at init time beofre any
3475  * slave exists. So we must declare proxy setup function which will
3476  * be used at run time to resolve the actual slave neigh param setup.
3477  *
3478  * It's also called by master devices (such as vlans) to setup their
3479  * underlying devices. In that case - do nothing, we're already set up from
3480  * our init.
3481  */
3482 static int bond_neigh_setup(struct net_device *dev,
3483                             struct neigh_parms *parms)
3484 {
3485         /* modify only our neigh_parms */
3486         if (parms->dev == dev)
3487                 parms->neigh_setup = bond_neigh_init;
3488
3489         return 0;
3490 }
3491
3492 /* Change the MTU of all of a master's slaves to match the master */
3493 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3494 {
3495         struct bonding *bond = netdev_priv(bond_dev);
3496         struct slave *slave, *rollback_slave;
3497         struct list_head *iter;
3498         int res = 0;
3499
3500         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3501
3502         bond_for_each_slave(bond, slave, iter) {
3503                 netdev_dbg(bond_dev, "s %p c_m %p\n",
3504                            slave, slave->dev->netdev_ops->ndo_change_mtu);
3505
3506                 res = dev_set_mtu(slave->dev, new_mtu);
3507
3508                 if (res) {
3509                         /* If we failed to set the slave's mtu to the new value
3510                          * we must abort the operation even in ACTIVE_BACKUP
3511                          * mode, because if we allow the backup slaves to have
3512                          * different mtu values than the active slave we'll
3513                          * need to change their mtu when doing a failover. That
3514                          * means changing their mtu from timer context, which
3515                          * is probably not a good idea.
3516                          */
3517                         netdev_dbg(bond_dev, "err %d %s\n", res,
3518                                    slave->dev->name);
3519                         goto unwind;
3520                 }
3521         }
3522
3523         bond_dev->mtu = new_mtu;
3524
3525         return 0;
3526
3527 unwind:
3528         /* unwind from head to the slave that failed */
3529         bond_for_each_slave(bond, rollback_slave, iter) {
3530                 int tmp_res;
3531
3532                 if (rollback_slave == slave)
3533                         break;
3534
3535                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3536                 if (tmp_res) {
3537                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3538                                    tmp_res, rollback_slave->dev->name);
3539                 }
3540         }
3541
3542         return res;
3543 }
3544
3545 /* Change HW address
3546  *
3547  * Note that many devices must be down to change the HW address, and
3548  * downing the master releases all slaves.  We can make bonds full of
3549  * bonding devices to test this, however.
3550  */
3551 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3552 {
3553         struct bonding *bond = netdev_priv(bond_dev);
3554         struct slave *slave, *rollback_slave;
3555         struct sockaddr *sa = addr, tmp_sa;
3556         struct list_head *iter;
3557         int res = 0;
3558
3559         if (BOND_MODE(bond) == BOND_MODE_ALB)
3560                 return bond_alb_set_mac_address(bond_dev, addr);
3561
3562
3563         netdev_dbg(bond_dev, "bond=%p\n", bond);
3564
3565         /* If fail_over_mac is enabled, do nothing and return success.
3566          * Returning an error causes ifenslave to fail.
3567          */
3568         if (bond->params.fail_over_mac &&
3569             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3570                 return 0;
3571
3572         if (!is_valid_ether_addr(sa->sa_data))
3573                 return -EADDRNOTAVAIL;
3574
3575         bond_for_each_slave(bond, slave, iter) {
3576                 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3577                 res = dev_set_mac_address(slave->dev, addr);
3578                 if (res) {
3579                         /* TODO: consider downing the slave
3580                          * and retry ?
3581                          * User should expect communications
3582                          * breakage anyway until ARP finish
3583                          * updating, so...
3584                          */
3585                         netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3586                         goto unwind;
3587                 }
3588         }
3589
3590         /* success */
3591         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3592         return 0;
3593
3594 unwind:
3595         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3596         tmp_sa.sa_family = bond_dev->type;
3597
3598         /* unwind from head to the slave that failed */
3599         bond_for_each_slave(bond, rollback_slave, iter) {
3600                 int tmp_res;
3601
3602                 if (rollback_slave == slave)
3603                         break;
3604
3605                 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3606                 if (tmp_res) {
3607                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3608                                    tmp_res, rollback_slave->dev->name);
3609                 }
3610         }
3611
3612         return res;
3613 }
3614
3615 /**
3616  * bond_xmit_slave_id - transmit skb through slave with slave_id
3617  * @bond: bonding device that is transmitting
3618  * @skb: buffer to transmit
3619  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3620  *
3621  * This function tries to transmit through slave with slave_id but in case
3622  * it fails, it tries to find the first available slave for transmission.
3623  * The skb is consumed in all cases, thus the function is void.
3624  */
3625 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3626 {
3627         struct list_head *iter;
3628         struct slave *slave;
3629         int i = slave_id;
3630
3631         /* Here we start from the slave with slave_id */
3632         bond_for_each_slave_rcu(bond, slave, iter) {
3633                 if (--i < 0) {
3634                         if (bond_slave_can_tx(slave)) {
3635                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3636                                 return;
3637                         }
3638                 }
3639         }
3640
3641         /* Here we start from the first slave up to slave_id */
3642         i = slave_id;
3643         bond_for_each_slave_rcu(bond, slave, iter) {
3644                 if (--i < 0)
3645                         break;
3646                 if (bond_slave_can_tx(slave)) {
3647                         bond_dev_queue_xmit(bond, skb, slave->dev);
3648                         return;
3649                 }
3650         }
3651         /* no slave that can tx has been found */
3652         bond_tx_drop(bond->dev, skb);
3653 }
3654
3655 /**
3656  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3657  * @bond: bonding device to use
3658  *
3659  * Based on the value of the bonding device's packets_per_slave parameter
3660  * this function generates a slave id, which is usually used as the next
3661  * slave to transmit through.
3662  */
3663 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3664 {
3665         u32 slave_id;
3666         struct reciprocal_value reciprocal_packets_per_slave;
3667         int packets_per_slave = bond->params.packets_per_slave;
3668
3669         switch (packets_per_slave) {
3670         case 0:
3671                 slave_id = prandom_u32();
3672                 break;
3673         case 1:
3674                 slave_id = bond->rr_tx_counter;
3675                 break;
3676         default:
3677                 reciprocal_packets_per_slave =
3678                         bond->params.reciprocal_packets_per_slave;
3679                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3680                                              reciprocal_packets_per_slave);
3681                 break;
3682         }
3683         bond->rr_tx_counter++;
3684
3685         return slave_id;
3686 }
3687
3688 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3689 {
3690         struct bonding *bond = netdev_priv(bond_dev);
3691         struct iphdr *iph = ip_hdr(skb);
3692         struct slave *slave;
3693         u32 slave_id;
3694
3695         /* Start with the curr_active_slave that joined the bond as the
3696          * default for sending IGMP traffic.  For failover purposes one
3697          * needs to maintain some consistency for the interface that will
3698          * send the join/membership reports.  The curr_active_slave found
3699          * will send all of this type of traffic.
3700          */
3701         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3702                 slave = rcu_dereference(bond->curr_active_slave);
3703                 if (slave)
3704                         bond_dev_queue_xmit(bond, skb, slave->dev);
3705                 else
3706                         bond_xmit_slave_id(bond, skb, 0);
3707         } else {
3708                 int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3709
3710                 if (likely(slave_cnt)) {
3711                         slave_id = bond_rr_gen_slave_id(bond);
3712                         bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3713                 } else {
3714                         bond_tx_drop(bond_dev, skb);
3715                 }
3716         }
3717
3718         return NETDEV_TX_OK;
3719 }
3720
3721 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3722  * the bond has a usable interface.
3723  */
3724 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3725 {
3726         struct bonding *bond = netdev_priv(bond_dev);
3727         struct slave *slave;
3728
3729         slave = rcu_dereference(bond->curr_active_slave);
3730         if (slave)
3731                 bond_dev_queue_xmit(bond, skb, slave->dev);
3732         else
3733                 bond_tx_drop(bond_dev, skb);
3734
3735         return NETDEV_TX_OK;
3736 }
3737
3738 /* Use this to update slave_array when (a) it's not appropriate to update
3739  * slave_array right away (note that update_slave_array() may sleep)
3740  * and / or (b) RTNL is not held.
3741  */
3742 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3743 {
3744         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3745 }
3746
3747 /* Slave array work handler. Holds only RTNL */
3748 static void bond_slave_arr_handler(struct work_struct *work)
3749 {
3750         struct bonding *bond = container_of(work, struct bonding,
3751                                             slave_arr_work.work);
3752         int ret;
3753
3754         if (!rtnl_trylock())
3755                 goto err;
3756
3757         ret = bond_update_slave_arr(bond, NULL);
3758         rtnl_unlock();
3759         if (ret) {
3760                 pr_warn_ratelimited("Failed to update slave array from WT\n");
3761                 goto err;
3762         }
3763         return;
3764
3765 err:
3766         bond_slave_arr_work_rearm(bond, 1);
3767 }
3768
3769 /* Build the usable slaves array in control path for modes that use xmit-hash
3770  * to determine the slave interface -
3771  * (a) BOND_MODE_8023AD
3772  * (b) BOND_MODE_XOR
3773  * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3774  *
3775  * The caller is expected to hold RTNL only and NO other lock!
3776  */
3777 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3778 {
3779         struct slave *slave;
3780         struct list_head *iter;
3781         struct bond_up_slave *new_arr, *old_arr;
3782         int slaves_in_agg;
3783         int agg_id = 0;
3784         int ret = 0;
3785
3786 #ifdef CONFIG_LOCKDEP
3787         WARN_ON(lockdep_is_held(&bond->mode_lock));
3788 #endif
3789
3790         new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3791                           GFP_KERNEL);
3792         if (!new_arr) {
3793                 ret = -ENOMEM;
3794                 pr_err("Failed to build slave-array.\n");
3795                 goto out;
3796         }
3797         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3798                 struct ad_info ad_info;
3799
3800                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3801                         pr_debug("bond_3ad_get_active_agg_info failed\n");
3802                         kfree_rcu(new_arr, rcu);
3803                         /* No active aggragator means it's not safe to use
3804                          * the previous array.
3805                          */
3806                         old_arr = rtnl_dereference(bond->slave_arr);
3807                         if (old_arr) {
3808                                 RCU_INIT_POINTER(bond->slave_arr, NULL);
3809                                 kfree_rcu(old_arr, rcu);
3810                         }
3811                         goto out;
3812                 }
3813                 slaves_in_agg = ad_info.ports;
3814                 agg_id = ad_info.aggregator_id;
3815         }
3816         bond_for_each_slave(bond, slave, iter) {
3817                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3818                         struct aggregator *agg;
3819
3820                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
3821                         if (!agg || agg->aggregator_identifier != agg_id)
3822                                 continue;
3823                 }
3824                 if (!bond_slave_can_tx(slave))
3825                         continue;
3826                 if (skipslave == slave)
3827                         continue;
3828                 new_arr->arr[new_arr->count++] = slave;
3829         }
3830
3831         old_arr = rtnl_dereference(bond->slave_arr);
3832         rcu_assign_pointer(bond->slave_arr, new_arr);
3833         if (old_arr)
3834                 kfree_rcu(old_arr, rcu);
3835 out:
3836         if (ret != 0 && skipslave) {
3837                 int idx;
3838
3839                 /* Rare situation where caller has asked to skip a specific
3840                  * slave but allocation failed (most likely!). BTW this is
3841                  * only possible when the call is initiated from
3842                  * __bond_release_one(). In this situation; overwrite the
3843                  * skipslave entry in the array with the last entry from the
3844                  * array to avoid a situation where the xmit path may choose
3845                  * this to-be-skipped slave to send a packet out.
3846                  */
3847                 old_arr = rtnl_dereference(bond->slave_arr);
3848                 for (idx = 0; idx < old_arr->count; idx++) {
3849                         if (skipslave == old_arr->arr[idx]) {
3850                                 old_arr->arr[idx] =
3851                                     old_arr->arr[old_arr->count-1];
3852                                 old_arr->count--;
3853                                 break;
3854                         }
3855                 }
3856         }
3857         return ret;
3858 }
3859
3860 /* Use this Xmit function for 3AD as well as XOR modes. The current
3861  * usable slave array is formed in the control path. The xmit function
3862  * just calculates hash and sends the packet out.
3863  */
3864 static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3865 {
3866         struct bonding *bond = netdev_priv(dev);
3867         struct slave *slave;
3868         struct bond_up_slave *slaves;
3869         unsigned int count;
3870
3871         slaves = rcu_dereference(bond->slave_arr);
3872         count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3873         if (likely(count)) {
3874                 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3875                 bond_dev_queue_xmit(bond, skb, slave->dev);
3876         } else {
3877                 bond_tx_drop(dev, skb);
3878         }
3879
3880         return NETDEV_TX_OK;
3881 }
3882
3883 /* in broadcast mode, we send everything to all usable interfaces. */
3884 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3885 {
3886         struct bonding *bond = netdev_priv(bond_dev);
3887         struct slave *slave = NULL;
3888         struct list_head *iter;
3889
3890         bond_for_each_slave_rcu(bond, slave, iter) {
3891                 if (bond_is_last_slave(bond, slave))
3892                         break;
3893                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3894                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3895
3896                         if (!skb2) {
3897                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3898                                                     bond_dev->name, __func__);
3899                                 continue;
3900                         }
3901                         bond_dev_queue_xmit(bond, skb2, slave->dev);
3902                 }
3903         }
3904         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3905                 bond_dev_queue_xmit(bond, skb, slave->dev);
3906         else
3907                 bond_tx_drop(bond_dev, skb);
3908
3909         return NETDEV_TX_OK;
3910 }
3911
3912 /*------------------------- Device initialization ---------------------------*/
3913
3914 /* Lookup the slave that corresponds to a qid */
3915 static inline int bond_slave_override(struct bonding *bond,
3916                                       struct sk_buff *skb)
3917 {
3918         struct slave *slave = NULL;
3919         struct list_head *iter;
3920
3921         if (!skb->queue_mapping)
3922                 return 1;
3923
3924         /* Find out if any slaves have the same mapping as this skb. */
3925         bond_for_each_slave_rcu(bond, slave, iter) {
3926                 if (slave->queue_id == skb->queue_mapping) {
3927                         if (bond_slave_is_up(slave) &&
3928                             slave->link == BOND_LINK_UP) {
3929                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3930                                 return 0;
3931                         }
3932                         /* If the slave isn't UP, use default transmit policy. */
3933                         break;
3934                 }
3935         }
3936
3937         return 1;
3938 }
3939
3940
3941 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3942                              void *accel_priv, select_queue_fallback_t fallback)
3943 {
3944         /* This helper function exists to help dev_pick_tx get the correct
3945          * destination queue.  Using a helper function skips a call to
3946          * skb_tx_hash and will put the skbs in the queue we expect on their
3947          * way down to the bonding driver.
3948          */
3949         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3950
3951         /* Save the original txq to restore before passing to the driver */
3952         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3953
3954         if (unlikely(txq >= dev->real_num_tx_queues)) {
3955                 do {
3956                         txq -= dev->real_num_tx_queues;
3957                 } while (txq >= dev->real_num_tx_queues);
3958         }
3959         return txq;
3960 }
3961
3962 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3963 {
3964         struct bonding *bond = netdev_priv(dev);
3965
3966         if (bond_should_override_tx_queue(bond) &&
3967             !bond_slave_override(bond, skb))
3968                 return NETDEV_TX_OK;
3969
3970         switch (BOND_MODE(bond)) {
3971         case BOND_MODE_ROUNDROBIN:
3972                 return bond_xmit_roundrobin(skb, dev);
3973         case BOND_MODE_ACTIVEBACKUP:
3974                 return bond_xmit_activebackup(skb, dev);
3975         case BOND_MODE_8023AD:
3976         case BOND_MODE_XOR:
3977                 return bond_3ad_xor_xmit(skb, dev);
3978         case BOND_MODE_BROADCAST:
3979                 return bond_xmit_broadcast(skb, dev);
3980         case BOND_MODE_ALB:
3981                 return bond_alb_xmit(skb, dev);
3982         case BOND_MODE_TLB:
3983                 return bond_tlb_xmit(skb, dev);
3984         default:
3985                 /* Should never happen, mode already checked */
3986                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
3987                 WARN_ON_ONCE(1);
3988                 bond_tx_drop(dev, skb);
3989                 return NETDEV_TX_OK;
3990         }
3991 }
3992
3993 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3994 {
3995         struct bonding *bond = netdev_priv(dev);
3996         netdev_tx_t ret = NETDEV_TX_OK;
3997
3998         /* If we risk deadlock from transmitting this in the
3999          * netpoll path, tell netpoll to queue the frame for later tx
4000          */
4001         if (unlikely(is_netpoll_tx_blocked(dev)))
4002                 return NETDEV_TX_BUSY;
4003
4004         rcu_read_lock();
4005         if (bond_has_slaves(bond))
4006                 ret = __bond_start_xmit(skb, dev);
4007         else
4008                 bond_tx_drop(dev, skb);
4009         rcu_read_unlock();
4010
4011         return ret;
4012 }
4013
4014 static int bond_ethtool_get_settings(struct net_device *bond_dev,
4015                                      struct ethtool_cmd *ecmd)
4016 {
4017         struct bonding *bond = netdev_priv(bond_dev);
4018         unsigned long speed = 0;
4019         struct list_head *iter;
4020         struct slave *slave;
4021
4022         ecmd->duplex = DUPLEX_UNKNOWN;
4023         ecmd->port = PORT_OTHER;
4024
4025         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4026          * do not need to check mode.  Though link speed might not represent
4027          * the true receive or transmit bandwidth (not all modes are symmetric)
4028          * this is an accurate maximum.
4029          */
4030         bond_for_each_slave(bond, slave, iter) {
4031                 if (bond_slave_can_tx(slave)) {
4032                         if (slave->speed != SPEED_UNKNOWN)
4033                                 speed += slave->speed;
4034                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
4035                             slave->duplex != DUPLEX_UNKNOWN)
4036                                 ecmd->duplex = slave->duplex;
4037                 }
4038         }
4039         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
4040
4041         return 0;
4042 }
4043
4044 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4045                                      struct ethtool_drvinfo *drvinfo)
4046 {
4047         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4048         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4049         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4050                  BOND_ABI_VERSION);
4051 }
4052
4053 static const struct ethtool_ops bond_ethtool_ops = {
4054         .get_drvinfo            = bond_ethtool_get_drvinfo,
4055         .get_settings           = bond_ethtool_get_settings,
4056         .get_link               = ethtool_op_get_link,
4057 };
4058
4059 static const struct net_device_ops bond_netdev_ops = {
4060         .ndo_init               = bond_init,
4061         .ndo_uninit             = bond_uninit,
4062         .ndo_open               = bond_open,
4063         .ndo_stop               = bond_close,
4064         .ndo_start_xmit         = bond_start_xmit,
4065         .ndo_select_queue       = bond_select_queue,
4066         .ndo_get_stats64        = bond_get_stats,
4067         .ndo_do_ioctl           = bond_do_ioctl,
4068         .ndo_change_rx_flags    = bond_change_rx_flags,
4069         .ndo_set_rx_mode        = bond_set_rx_mode,
4070         .ndo_change_mtu         = bond_change_mtu,
4071         .ndo_set_mac_address    = bond_set_mac_address,
4072         .ndo_neigh_setup        = bond_neigh_setup,
4073         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4074         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4075 #ifdef CONFIG_NET_POLL_CONTROLLER
4076         .ndo_netpoll_setup      = bond_netpoll_setup,
4077         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4078         .ndo_poll_controller    = bond_poll_controller,
4079 #endif
4080         .ndo_add_slave          = bond_enslave,
4081         .ndo_del_slave          = bond_release,
4082         .ndo_fix_features       = bond_fix_features,
4083         .ndo_bridge_setlink     = switchdev_port_bridge_setlink,
4084         .ndo_bridge_getlink     = switchdev_port_bridge_getlink,
4085         .ndo_bridge_dellink     = switchdev_port_bridge_dellink,
4086         .ndo_fdb_add            = switchdev_port_fdb_add,
4087         .ndo_fdb_del            = switchdev_port_fdb_del,
4088         .ndo_fdb_dump           = switchdev_port_fdb_dump,
4089         .ndo_features_check     = passthru_features_check,
4090 };
4091
4092 static const struct device_type bond_type = {
4093         .name = "bond",
4094 };
4095
4096 static void bond_destructor(struct net_device *bond_dev)
4097 {
4098         struct bonding *bond = netdev_priv(bond_dev);
4099         if (bond->wq)
4100                 destroy_workqueue(bond->wq);
4101         free_netdev(bond_dev);
4102 }
4103
4104 void bond_setup(struct net_device *bond_dev)
4105 {
4106         struct bonding *bond = netdev_priv(bond_dev);
4107
4108         spin_lock_init(&bond->mode_lock);
4109         bond->params = bonding_defaults;
4110
4111         /* Initialize pointers */
4112         bond->dev = bond_dev;
4113
4114         /* Initialize the device entry points */
4115         ether_setup(bond_dev);
4116         bond_dev->netdev_ops = &bond_netdev_ops;
4117         bond_dev->ethtool_ops = &bond_ethtool_ops;
4118
4119         bond_dev->destructor = bond_destructor;
4120
4121         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4122
4123         /* Initialize the device options */
4124         bond_dev->tx_queue_len = 0;
4125         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4126         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT;
4127         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4128
4129         /* don't acquire bond device's netif_tx_lock when transmitting */
4130         bond_dev->features |= NETIF_F_LLTX;
4131
4132         /* By default, we declare the bond to be fully
4133          * VLAN hardware accelerated capable. Special
4134          * care is taken in the various xmit functions
4135          * when there are slaves that are not hw accel
4136          * capable
4137          */
4138
4139         /* Don't allow bond devices to change network namespaces. */
4140         bond_dev->features |= NETIF_F_NETNS_LOCAL;
4141
4142         bond_dev->hw_features = BOND_VLAN_FEATURES |
4143                                 NETIF_F_HW_VLAN_CTAG_TX |
4144                                 NETIF_F_HW_VLAN_CTAG_RX |
4145                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4146
4147         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4148         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4149         bond_dev->features |= bond_dev->hw_features;
4150 }
4151
4152 /* Destroy a bonding device.
4153  * Must be under rtnl_lock when this function is called.
4154  */
4155 static void bond_uninit(struct net_device *bond_dev)
4156 {
4157         struct bonding *bond = netdev_priv(bond_dev);
4158         struct list_head *iter;
4159         struct slave *slave;
4160         struct bond_up_slave *arr;
4161
4162         bond_netpoll_cleanup(bond_dev);
4163
4164         /* Release the bonded slaves */
4165         bond_for_each_slave(bond, slave, iter)
4166                 __bond_release_one(bond_dev, slave->dev, true);
4167         netdev_info(bond_dev, "Released all slaves\n");
4168
4169         arr = rtnl_dereference(bond->slave_arr);
4170         if (arr) {
4171                 RCU_INIT_POINTER(bond->slave_arr, NULL);
4172                 kfree_rcu(arr, rcu);
4173         }
4174
4175         list_del(&bond->bond_list);
4176
4177         bond_debug_unregister(bond);
4178 }
4179
4180 /*------------------------- Module initialization ---------------------------*/
4181
4182 static int bond_check_params(struct bond_params *params)
4183 {
4184         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4185         struct bond_opt_value newval;
4186         const struct bond_opt_value *valptr;
4187         int arp_all_targets_value;
4188         u16 ad_actor_sys_prio = 0;
4189         u16 ad_user_port_key = 0;
4190
4191         /* Convert string parameters. */
4192         if (mode) {
4193                 bond_opt_initstr(&newval, mode);
4194                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4195                 if (!valptr) {
4196                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4197                         return -EINVAL;
4198                 }
4199                 bond_mode = valptr->value;
4200         }
4201
4202         if (xmit_hash_policy) {
4203                 if ((bond_mode != BOND_MODE_XOR) &&
4204                     (bond_mode != BOND_MODE_8023AD) &&
4205                     (bond_mode != BOND_MODE_TLB)) {
4206                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4207                                 bond_mode_name(bond_mode));
4208                 } else {
4209                         bond_opt_initstr(&newval, xmit_hash_policy);
4210                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4211                                                 &newval);
4212                         if (!valptr) {
4213                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4214                                        xmit_hash_policy);
4215                                 return -EINVAL;
4216                         }
4217                         xmit_hashtype = valptr->value;
4218                 }
4219         }
4220
4221         if (lacp_rate) {
4222                 if (bond_mode != BOND_MODE_8023AD) {
4223                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4224                                 bond_mode_name(bond_mode));
4225                 } else {
4226                         bond_opt_initstr(&newval, lacp_rate);
4227                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4228                                                 &newval);
4229                         if (!valptr) {
4230                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4231                                        lacp_rate);
4232                                 return -EINVAL;
4233                         }
4234                         lacp_fast = valptr->value;
4235                 }
4236         }
4237
4238         if (ad_select) {
4239                 bond_opt_initstr(&newval, ad_select);
4240                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4241                                         &newval);
4242                 if (!valptr) {
4243                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4244                         return -EINVAL;
4245                 }
4246                 params->ad_select = valptr->value;
4247                 if (bond_mode != BOND_MODE_8023AD)
4248                         pr_warn("ad_select param only affects 802.3ad mode\n");
4249         } else {
4250                 params->ad_select = BOND_AD_STABLE;
4251         }
4252
4253         if (max_bonds < 0) {
4254                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4255                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4256                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4257         }
4258
4259         if (miimon < 0) {
4260                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4261                         miimon, INT_MAX);
4262                 miimon = 0;
4263         }
4264
4265         if (updelay < 0) {
4266                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4267                         updelay, INT_MAX);
4268                 updelay = 0;
4269         }
4270
4271         if (downdelay < 0) {
4272                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4273                         downdelay, INT_MAX);
4274                 downdelay = 0;
4275         }
4276
4277         if ((use_carrier != 0) && (use_carrier != 1)) {
4278                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4279                         use_carrier);
4280                 use_carrier = 1;
4281         }
4282
4283         if (num_peer_notif < 0 || num_peer_notif > 255) {
4284                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4285                         num_peer_notif);
4286                 num_peer_notif = 1;
4287         }
4288
4289         /* reset values for 802.3ad/TLB/ALB */
4290         if (!bond_mode_uses_arp(bond_mode)) {
4291                 if (!miimon) {
4292                         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");
4293                         pr_warn("Forcing miimon to 100msec\n");
4294                         miimon = BOND_DEFAULT_MIIMON;
4295                 }
4296         }
4297
4298         if (tx_queues < 1 || tx_queues > 255) {
4299                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4300                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4301                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4302         }
4303
4304         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4305                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4306                         all_slaves_active);
4307                 all_slaves_active = 0;
4308         }
4309
4310         if (resend_igmp < 0 || resend_igmp > 255) {
4311                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4312                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4313                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4314         }
4315
4316         bond_opt_initval(&newval, packets_per_slave);
4317         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4318                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4319                         packets_per_slave, USHRT_MAX);
4320                 packets_per_slave = 1;
4321         }
4322
4323         if (bond_mode == BOND_MODE_ALB) {
4324                 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",
4325                           updelay);
4326         }
4327
4328         if (!miimon) {
4329                 if (updelay || downdelay) {
4330                         /* just warn the user the up/down delay will have
4331                          * no effect since miimon is zero...
4332                          */
4333                         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",
4334                                 updelay, downdelay);
4335                 }
4336         } else {
4337                 /* don't allow arp monitoring */
4338                 if (arp_interval) {
4339                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4340                                 miimon, arp_interval);
4341                         arp_interval = 0;
4342                 }
4343
4344                 if ((updelay % miimon) != 0) {
4345                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4346                                 updelay, miimon, (updelay / miimon) * miimon);
4347                 }
4348
4349                 updelay /= miimon;
4350
4351                 if ((downdelay % miimon) != 0) {
4352                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4353                                 downdelay, miimon,
4354                                 (downdelay / miimon) * miimon);
4355                 }
4356
4357                 downdelay /= miimon;
4358         }
4359
4360         if (arp_interval < 0) {
4361                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4362                         arp_interval, INT_MAX);
4363                 arp_interval = 0;
4364         }
4365
4366         for (arp_ip_count = 0, i = 0;
4367              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4368                 __be32 ip;
4369
4370                 /* not a complete check, but good enough to catch mistakes */
4371                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4372                     !bond_is_ip_target_ok(ip)) {
4373                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4374                                 arp_ip_target[i]);
4375                         arp_interval = 0;
4376                 } else {
4377                         if (bond_get_targets_ip(arp_target, ip) == -1)
4378                                 arp_target[arp_ip_count++] = ip;
4379                         else
4380                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4381                                         &ip);
4382                 }
4383         }
4384
4385         if (arp_interval && !arp_ip_count) {
4386                 /* don't allow arping if no arp_ip_target given... */
4387                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4388                         arp_interval);
4389                 arp_interval = 0;
4390         }
4391
4392         if (arp_validate) {
4393                 if (!arp_interval) {
4394                         pr_err("arp_validate requires arp_interval\n");
4395                         return -EINVAL;
4396                 }
4397
4398                 bond_opt_initstr(&newval, arp_validate);
4399                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4400                                         &newval);
4401                 if (!valptr) {
4402                         pr_err("Error: invalid arp_validate \"%s\"\n",
4403                                arp_validate);
4404                         return -EINVAL;
4405                 }
4406                 arp_validate_value = valptr->value;
4407         } else {
4408                 arp_validate_value = 0;
4409         }
4410
4411         arp_all_targets_value = 0;
4412         if (arp_all_targets) {
4413                 bond_opt_initstr(&newval, arp_all_targets);
4414                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4415                                         &newval);
4416                 if (!valptr) {
4417                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4418                                arp_all_targets);
4419                         arp_all_targets_value = 0;
4420                 } else {
4421                         arp_all_targets_value = valptr->value;
4422                 }
4423         }
4424
4425         if (miimon) {
4426                 pr_info("MII link monitoring set to %d ms\n", miimon);
4427         } else if (arp_interval) {
4428                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4429                                           arp_validate_value);
4430                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4431                         arp_interval, valptr->string, arp_ip_count);
4432
4433                 for (i = 0; i < arp_ip_count; i++)
4434                         pr_cont(" %s", arp_ip_target[i]);
4435
4436                 pr_cont("\n");
4437
4438         } else if (max_bonds) {
4439                 /* miimon and arp_interval not set, we need one so things
4440                  * work as expected, see bonding.txt for details
4441                  */
4442                 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");
4443         }
4444
4445         if (primary && !bond_mode_uses_primary(bond_mode)) {
4446                 /* currently, using a primary only makes sense
4447                  * in active backup, TLB or ALB modes
4448                  */
4449                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4450                         primary, bond_mode_name(bond_mode));
4451                 primary = NULL;
4452         }
4453
4454         if (primary && primary_reselect) {
4455                 bond_opt_initstr(&newval, primary_reselect);
4456                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4457                                         &newval);
4458                 if (!valptr) {
4459                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4460                                primary_reselect);
4461                         return -EINVAL;
4462                 }
4463                 primary_reselect_value = valptr->value;
4464         } else {
4465                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4466         }
4467
4468         if (fail_over_mac) {
4469                 bond_opt_initstr(&newval, fail_over_mac);
4470                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4471                                         &newval);
4472                 if (!valptr) {
4473                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4474                                fail_over_mac);
4475                         return -EINVAL;
4476                 }
4477                 fail_over_mac_value = valptr->value;
4478                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4479                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4480         } else {
4481                 fail_over_mac_value = BOND_FOM_NONE;
4482         }
4483
4484         bond_opt_initstr(&newval, "default");
4485         valptr = bond_opt_parse(
4486                         bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4487                                      &newval);
4488         if (!valptr) {
4489                 pr_err("Error: No ad_actor_sys_prio default value");
4490                 return -EINVAL;
4491         }
4492         ad_actor_sys_prio = valptr->value;
4493
4494         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4495                                 &newval);
4496         if (!valptr) {
4497                 pr_err("Error: No ad_user_port_key default value");
4498                 return -EINVAL;
4499         }
4500         ad_user_port_key = valptr->value;
4501
4502         if (lp_interval == 0) {
4503                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4504                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4505                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4506         }
4507
4508         /* fill params struct with the proper values */
4509         params->mode = bond_mode;
4510         params->xmit_policy = xmit_hashtype;
4511         params->miimon = miimon;
4512         params->num_peer_notif = num_peer_notif;
4513         params->arp_interval = arp_interval;
4514         params->arp_validate = arp_validate_value;
4515         params->arp_all_targets = arp_all_targets_value;
4516         params->updelay = updelay;
4517         params->downdelay = downdelay;
4518         params->use_carrier = use_carrier;
4519         params->lacp_fast = lacp_fast;
4520         params->primary[0] = 0;
4521         params->primary_reselect = primary_reselect_value;
4522         params->fail_over_mac = fail_over_mac_value;
4523         params->tx_queues = tx_queues;
4524         params->all_slaves_active = all_slaves_active;
4525         params->resend_igmp = resend_igmp;
4526         params->min_links = min_links;
4527         params->lp_interval = lp_interval;
4528         params->packets_per_slave = packets_per_slave;
4529         params->tlb_dynamic_lb = 1; /* Default value */
4530         params->ad_actor_sys_prio = ad_actor_sys_prio;
4531         eth_zero_addr(params->ad_actor_system);
4532         params->ad_user_port_key = ad_user_port_key;
4533         if (packets_per_slave > 0) {
4534                 params->reciprocal_packets_per_slave =
4535                         reciprocal_value(packets_per_slave);
4536         } else {
4537                 /* reciprocal_packets_per_slave is unused if
4538                  * packets_per_slave is 0 or 1, just initialize it
4539                  */
4540                 params->reciprocal_packets_per_slave =
4541                         (struct reciprocal_value) { 0 };
4542         }
4543
4544         if (primary) {
4545                 strncpy(params->primary, primary, IFNAMSIZ);
4546                 params->primary[IFNAMSIZ - 1] = 0;
4547         }
4548
4549         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4550
4551         return 0;
4552 }
4553
4554 static struct lock_class_key bonding_netdev_xmit_lock_key;
4555 static struct lock_class_key bonding_netdev_addr_lock_key;
4556 static struct lock_class_key bonding_tx_busylock_key;
4557
4558 static void bond_set_lockdep_class_one(struct net_device *dev,
4559                                        struct netdev_queue *txq,
4560                                        void *_unused)
4561 {
4562         lockdep_set_class(&txq->_xmit_lock,
4563                           &bonding_netdev_xmit_lock_key);
4564 }
4565
4566 static void bond_set_lockdep_class(struct net_device *dev)
4567 {
4568         lockdep_set_class(&dev->addr_list_lock,
4569                           &bonding_netdev_addr_lock_key);
4570         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4571         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4572 }
4573
4574 /* Called from registration process */
4575 static int bond_init(struct net_device *bond_dev)
4576 {
4577         struct bonding *bond = netdev_priv(bond_dev);
4578         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4579
4580         netdev_dbg(bond_dev, "Begin bond_init\n");
4581
4582         bond->wq = create_singlethread_workqueue(bond_dev->name);
4583         if (!bond->wq)
4584                 return -ENOMEM;
4585
4586         bond_set_lockdep_class(bond_dev);
4587
4588         list_add_tail(&bond->bond_list, &bn->dev_list);
4589
4590         bond_prepare_sysfs_group(bond);
4591
4592         bond_debug_register(bond);
4593
4594         /* Ensure valid dev_addr */
4595         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4596             bond_dev->addr_assign_type == NET_ADDR_PERM)
4597                 eth_hw_addr_random(bond_dev);
4598
4599         return 0;
4600 }
4601
4602 unsigned int bond_get_num_tx_queues(void)
4603 {
4604         return tx_queues;
4605 }
4606
4607 /* Create a new bond based on the specified name and bonding parameters.
4608  * If name is NULL, obtain a suitable "bond%d" name for us.
4609  * Caller must NOT hold rtnl_lock; we need to release it here before we
4610  * set up our sysfs entries.
4611  */
4612 int bond_create(struct net *net, const char *name)
4613 {
4614         struct net_device *bond_dev;
4615         struct bonding *bond;
4616         struct alb_bond_info *bond_info;
4617         int res;
4618
4619         rtnl_lock();
4620
4621         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4622                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
4623                                    bond_setup, tx_queues);
4624         if (!bond_dev) {
4625                 pr_err("%s: eek! can't alloc netdev!\n", name);
4626                 rtnl_unlock();
4627                 return -ENOMEM;
4628         }
4629
4630         /*
4631          * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4632          * It is set to 0 by default which is wrong.
4633          */
4634         bond = netdev_priv(bond_dev);
4635         bond_info = &(BOND_ALB_INFO(bond));
4636         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4637
4638         dev_net_set(bond_dev, net);
4639         bond_dev->rtnl_link_ops = &bond_link_ops;
4640
4641         res = register_netdevice(bond_dev);
4642
4643         netif_carrier_off(bond_dev);
4644
4645         rtnl_unlock();
4646         if (res < 0)
4647                 bond_destructor(bond_dev);
4648         return res;
4649 }
4650
4651 static int __net_init bond_net_init(struct net *net)
4652 {
4653         struct bond_net *bn = net_generic(net, bond_net_id);
4654
4655         bn->net = net;
4656         INIT_LIST_HEAD(&bn->dev_list);
4657
4658         bond_create_proc_dir(bn);
4659         bond_create_sysfs(bn);
4660
4661         return 0;
4662 }
4663
4664 static void __net_exit bond_net_exit(struct net *net)
4665 {
4666         struct bond_net *bn = net_generic(net, bond_net_id);
4667         struct bonding *bond, *tmp_bond;
4668         LIST_HEAD(list);
4669
4670         bond_destroy_sysfs(bn);
4671
4672         /* Kill off any bonds created after unregistering bond rtnl ops */
4673         rtnl_lock();
4674         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4675                 unregister_netdevice_queue(bond->dev, &list);
4676         unregister_netdevice_many(&list);
4677         rtnl_unlock();
4678
4679         bond_destroy_proc_dir(bn);
4680 }
4681
4682 static struct pernet_operations bond_net_ops = {
4683         .init = bond_net_init,
4684         .exit = bond_net_exit,
4685         .id   = &bond_net_id,
4686         .size = sizeof(struct bond_net),
4687 };
4688
4689 static int __init bonding_init(void)
4690 {
4691         int i;
4692         int res;
4693
4694         pr_info("%s", bond_version);
4695
4696         res = bond_check_params(&bonding_defaults);
4697         if (res)
4698                 goto out;
4699
4700         res = register_pernet_subsys(&bond_net_ops);
4701         if (res)
4702                 goto out;
4703
4704         res = bond_netlink_init();
4705         if (res)
4706                 goto err_link;
4707
4708         bond_create_debugfs();
4709
4710         for (i = 0; i < max_bonds; i++) {
4711                 res = bond_create(&init_net, NULL);
4712                 if (res)
4713                         goto err;
4714         }
4715
4716         register_netdevice_notifier(&bond_netdev_notifier);
4717 out:
4718         return res;
4719 err:
4720         bond_destroy_debugfs();
4721         bond_netlink_fini();
4722 err_link:
4723         unregister_pernet_subsys(&bond_net_ops);
4724         goto out;
4725
4726 }
4727
4728 static void __exit bonding_exit(void)
4729 {
4730         unregister_netdevice_notifier(&bond_netdev_notifier);
4731
4732         bond_destroy_debugfs();
4733
4734         bond_netlink_fini();
4735         unregister_pernet_subsys(&bond_net_ops);
4736
4737 #ifdef CONFIG_NET_POLL_CONTROLLER
4738         /* Make sure we don't have an imbalance on our netpoll blocking */
4739         WARN_ON(atomic_read(&netpoll_block_tx));
4740 #endif
4741 }
4742
4743 module_init(bonding_init);
4744 module_exit(bonding_exit);
4745 MODULE_LICENSE("GPL");
4746 MODULE_VERSION(DRV_VERSION);
4747 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4748 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");