drivers/net/bonding/bond_main.c

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 #include <linux/kernel.h>
  35 #include <linux/module.h>
  36 #include <linux/types.h>
  37 #include <linux/fcntl.h>
  38 #include <linux/interrupt.h>
  39 #include <linux/ptrace.h>
  40 #include <linux/ioport.h>
  41 #include <linux/in.h>
  42 #include <net/ip.h>
  43 #include <linux/ip.h>
  44 #include <linux/tcp.h>
  45 #include <linux/udp.h>
  46 #include <linux/slab.h>
  47 #include <linux/string.h>
  48 #include <linux/init.h>
  49 #include <linux/timer.h>
  50 #include <linux/socket.h>
  51 #include <linux/ctype.h>
  52 #include <linux/inet.h>
  53 #include <linux/bitops.h>
  54 #include <linux/io.h>
  55 #include <asm/dma.h>
  56 #include <linux/uaccess.h>
  57 #include <linux/errno.h>
  58 #include <linux/netdevice.h>
  59 #include <linux/inetdevice.h>
  60 #include <linux/igmp.h>
  61 #include <linux/etherdevice.h>
  62 #include <linux/skbuff.h>
  63 #include <net/sock.h>
  64 #include <linux/rtnetlink.h>
  65 #include <linux/smp.h>
  66 #include <linux/if_ether.h>
  67 #include <net/arp.h>
  68 #include <linux/mii.h>
  69 #include <linux/ethtool.h>
  70 #include <linux/if_vlan.h>
  71 #include <linux/if_bonding.h>
  72 #include <linux/jiffies.h>
  73 #include <linux/preempt.h>
  74 #include <net/route.h>
  75 #include <net/net_namespace.h>
  76 #include <net/netns/generic.h>
  77 #include <net/pkt_sched.h>
  78 #include <linux/rculist.h>
  79 #include <net/flow_dissector.h>
  80 #include <net/switchdev.h>
  81 #include <net/bonding.h>
  82 #include <net/bond_3ad.h>
  83 #include <net/bond_alb.h>
  84
  85 #include "bonding_priv.h"
  86
  87 /*---------------------------- Module parameters ----------------------------*/
  88
  89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  90
  91 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  92 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  93 static int num_peer_notif = 1;
  94 static int miimon;
  95 static int updelay;
  96 static int downdelay;
  97 static int use_carrier  = 1;
  98 static char *mode;
  99 static char *primary;
 100 static char *primary_reselect;
 101 static char *lacp_rate;
 102 static int min_links;
 103 static char *ad_select;
 104 static char *xmit_hash_policy;
 105 static int arp_interval;
 106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 107 static char *arp_validate;
 108 static char *arp_all_targets;
 109 static char *fail_over_mac;
 110 static int all_slaves_active;
 111 static struct bond_params bonding_defaults;
 112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 113 static int packets_per_slave = 1;
 114 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
 115
 116 module_param(max_bonds, int, 0);
 117 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 118 module_param(tx_queues, int, 0);
 119 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 120 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 121 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 122                                "failover event (alias of num_unsol_na)");
 123 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 124 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 125                                "failover event (alias of num_grat_arp)");
 126 module_param(miimon, int, 0);
 127 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 128 module_param(updelay, int, 0);
 129 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 130 module_param(downdelay, int, 0);
 131 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 132                             "in milliseconds");
 133 module_param(use_carrier, int, 0);
 134 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 135                               "0 for off, 1 for on (default)");
 136 module_param(mode, charp, 0);
 137 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 138                        "1 for active-backup, 2 for balance-xor, "
 139                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 140                        "6 for balance-alb");
 141 module_param(primary, charp, 0);
 142 MODULE_PARM_DESC(primary, "Primary network device to use");
 143 module_param(primary_reselect, charp, 0);
 144 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 145                                    "once it comes up; "
 146                                    "0 for always (default), "
 147                                    "1 for only if speed of primary is "
 148                                    "better, "
 149                                    "2 for only on active slave "
 150                                    "failure");
 151 module_param(lacp_rate, charp, 0);
 152 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 153                             "0 for slow, 1 for fast");
 154 module_param(ad_select, charp, 0);
 155 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 156                             "0 for stable (default), 1 for bandwidth, "
 157                             "2 for count");
 158 module_param(min_links, int, 0);
 159 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 160
 161 module_param(xmit_hash_policy, charp, 0);
 162 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 163                                    "0 for layer 2 (default), 1 for layer 3+4, "
 164                                    "2 for layer 2+3, 3 for encap layer 2+3, "
 165                                    "4 for encap layer 3+4");
 166 module_param(arp_interval, int, 0);
 167 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 168 module_param_array(arp_ip_target, charp, NULL, 0);
 169 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 170 module_param(arp_validate, charp, 0);
 171 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 172                                "0 for none (default), 1 for active, "
 173                                "2 for backup, 3 for all");
 174 module_param(arp_all_targets, charp, 0);
 175 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 176 module_param(fail_over_mac, charp, 0);
 177 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 178                                 "the same MAC; 0 for none (default), "
 179                                 "1 for active, 2 for follow");
 180 module_param(all_slaves_active, int, 0);
 181 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
 182                                      "by setting active flag for all slaves; "
 183                                      "0 for never (default), 1 for always.");
 184 module_param(resend_igmp, int, 0);
 185 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 186                               "link failure");
 187 module_param(packets_per_slave, int, 0);
 188 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
 189                                     "mode; 0 for a random slave, 1 packet per "
 190                                     "slave (default), >1 packets per slave.");
 191 module_param(lp_interval, uint, 0);
 192 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
 193                               "the bonding driver sends learning packets to "
 194                               "each slaves peer switch. The default is 1.");
 195
 196 /*----------------------------- Global variables ----------------------------*/
 197
 198 #ifdef CONFIG_NET_POLL_CONTROLLER
 199 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 200 #endif
 201
 202 int bond_net_id __read_mostly;
 203
 204 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 205 static int arp_ip_count;
 206 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 207 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 208 static int lacp_fast;
 209
 210 /*-------------------------- Forward declarations ---------------------------*/
 211
 212 static int bond_init(struct net_device *bond_dev);
 213 static void bond_uninit(struct net_device *bond_dev);
 214 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
 215                                                 struct rtnl_link_stats64 *stats);
 216 static void bond_slave_arr_handler(struct work_struct *work);
 217 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
 218                                   int mod);
 219
 220 /*---------------------------- General routines -----------------------------*/
 221
 222 const char *bond_mode_name(int mode)
 223 {
 224         static const char *names[] = {
 225                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 226                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 227                 [BOND_MODE_XOR] = "load balancing (xor)",
 228                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 229                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 230                 [BOND_MODE_TLB] = "transmit load balancing",
 231                 [BOND_MODE_ALB] = "adaptive load balancing",
 232         };
 233
 234         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 235                 return "unknown";
 236
 237         return names[mode];
 238 }
 239
 240 /*---------------------------------- VLAN -----------------------------------*/
 241
 242 /**
 243  * bond_dev_queue_xmit - Prepare skb for xmit.
 244  *
 245  * @bond: bond device that got this skb for tx.
 246  * @skb: hw accel VLAN tagged skb to transmit
 247  * @slave_dev: slave that is supposed to xmit this skbuff
 248  */
 249 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 250                         struct net_device *slave_dev)
 251 {
 252         skb->dev = slave_dev;
 253
 254         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 255                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 256         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 257
 258         if (unlikely(netpoll_tx_running(bond->dev)))
 259                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 260         else
 261                 dev_queue_xmit(skb);
 262 }
 263
 264 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 265  * We don't protect the slave list iteration with a lock because:
 266  * a. This operation is performed in IOCTL context,
 267  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 268  * c. Holding a lock with BH disabled while directly calling a base driver
 269  *    entry point is generally a BAD idea.
 270  *
 271  * The design of synchronization/protection for this operation in the 8021q
 272  * module is good for one or more VLAN devices over a single physical device
 273  * and cannot be extended for a teaming solution like bonding, so there is a
 274  * potential race condition here where a net device from the vlan group might
 275  * be referenced (either by a base driver or the 8021q code) while it is being
 276  * removed from the system. However, it turns out we're not making matters
 277  * worse, and if it works for regular VLAN usage it will work here too.
 278 */
 279
 280 /**
 281  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 282  * @bond_dev: bonding net device that got called
 283  * @vid: vlan id being added
 284  */
 285 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 286                                 __be16 proto, u16 vid)
 287 {
 288         struct bonding *bond = netdev_priv(bond_dev);
 289         struct slave *slave, *rollback_slave;
 290         struct list_head *iter;
 291         int res;
 292
 293         bond_for_each_slave(bond, slave, iter) {
 294                 res = vlan_vid_add(slave->dev, proto, vid);
 295                 if (res)
 296                         goto unwind;
 297         }
 298
 299         return 0;
 300
 301 unwind:
 302         /* unwind to the slave that failed */
 303         bond_for_each_slave(bond, rollback_slave, iter) {
 304                 if (rollback_slave == slave)
 305                         break;
 306
 307                 vlan_vid_del(rollback_slave->dev, proto, vid);
 308         }
 309
 310         return res;
 311 }
 312
 313 /**
 314  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 315  * @bond_dev: bonding net device that got called
 316  * @vid: vlan id being removed
 317  */
 318 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 319                                  __be16 proto, u16 vid)
 320 {
 321         struct bonding *bond = netdev_priv(bond_dev);
 322         struct list_head *iter;
 323         struct slave *slave;
 324
 325         bond_for_each_slave(bond, slave, iter)
 326                 vlan_vid_del(slave->dev, proto, vid);
 327
 328         if (bond_is_lb(bond))
 329                 bond_alb_clear_vlan(bond, vid);
 330
 331         return 0;
 332 }
 333
 334 /*------------------------------- Link status -------------------------------*/
 335
 336 /* Set the carrier state for the master according to the state of its
 337  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 338  * do special 802.3ad magic.
 339  *
 340  * Returns zero if carrier state does not change, nonzero if it does.
 341  */
 342 int bond_set_carrier(struct bonding *bond)
 343 {
 344         struct list_head *iter;
 345         struct slave *slave;
 346
 347         if (!bond_has_slaves(bond))
 348                 goto down;
 349
 350         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 351                 return bond_3ad_set_carrier(bond);
 352
 353         bond_for_each_slave(bond, slave, iter) {
 354                 if (slave->link == BOND_LINK_UP) {
 355                         if (!netif_carrier_ok(bond->dev)) {
 356                                 netif_carrier_on(bond->dev);
 357                                 return 1;
 358                         }
 359                         return 0;
 360                 }
 361         }
 362
 363 down:
 364         if (netif_carrier_ok(bond->dev)) {
 365                 netif_carrier_off(bond->dev);
 366                 return 1;
 367         }
 368         return 0;
 369 }
 370
 371 /* Get link speed and duplex from the slave's base driver
 372  * using ethtool. If for some reason the call fails or the
 373  * values are invalid, set speed and duplex to -1,
 374  * and return.
 375  */
 376 static void bond_update_speed_duplex(struct slave *slave)
 377 {
 378         struct net_device *slave_dev = slave->dev;
 379         struct ethtool_link_ksettings ecmd;
 380         int res;
 381
 382         slave->speed = SPEED_UNKNOWN;
 383         slave->duplex = DUPLEX_UNKNOWN;
 384
 385         res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
 386         if (res < 0)
 387                 return;
 388
 389         if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
 390                 return;
 391
 392         switch (ecmd.base.duplex) {
 393         case DUPLEX_FULL:
 394         case DUPLEX_HALF:
 395                 break;
 396         default:
 397                 return;
 398         }
 399
 400         slave->speed = ecmd.base.speed;
 401         slave->duplex = ecmd.base.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->name=%s slave_dev->addr_len=%d\n",
 622                    bond_dev, slave_dev, slave_dev->name, 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             !netif_carrier_ok(bond->dev) ||
 790             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 791                 return false;
 792
 793         return true;
 794 }
 795
 796 /**
 797  * change_active_interface - change the active slave into the specified one
 798  * @bond: our bonding struct
 799  * @new: the new slave to make the active one
 800  *
 801  * Set the new slave to the bond's settings and unset them on the old
 802  * curr_active_slave.
 803  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 804  *
 805  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 806  * because it is apparently the best available slave we have, even though its
 807  * updelay hasn't timed out yet.
 808  *
 809  * Caller must hold RTNL.
 810  */
 811 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 812 {
 813         struct slave *old_active;
 814
 815         ASSERT_RTNL();
 816
 817         old_active = rtnl_dereference(bond->curr_active_slave);
 818
 819         if (old_active == new_active)
 820                 return;
 821
 822         if (new_active) {
 823                 new_active->last_link_up = jiffies;
 824
 825                 if (new_active->link == BOND_LINK_BACK) {
 826                         if (bond_uses_primary(bond)) {
 827                                 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
 828                                             new_active->dev->name,
 829                                             (bond->params.updelay - new_active->delay) * bond->params.miimon);
 830                         }
 831
 832                         new_active->delay = 0;
 833                         bond_set_slave_link_state(new_active, BOND_LINK_UP,
 834                                                   BOND_SLAVE_NOTIFY_NOW);
 835
 836                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 837                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 838
 839                         if (bond_is_lb(bond))
 840                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 841                 } else {
 842                         if (bond_uses_primary(bond)) {
 843                                 netdev_info(bond->dev, "making interface %s the new active one\n",
 844                                             new_active->dev->name);
 845                         }
 846                 }
 847         }
 848
 849         if (bond_uses_primary(bond))
 850                 bond_hw_addr_swap(bond, new_active, old_active);
 851
 852         if (bond_is_lb(bond)) {
 853                 bond_alb_handle_active_change(bond, new_active);
 854                 if (old_active)
 855                         bond_set_slave_inactive_flags(old_active,
 856                                                       BOND_SLAVE_NOTIFY_NOW);
 857                 if (new_active)
 858                         bond_set_slave_active_flags(new_active,
 859                                                     BOND_SLAVE_NOTIFY_NOW);
 860         } else {
 861                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 862         }
 863
 864         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
 865                 if (old_active)
 866                         bond_set_slave_inactive_flags(old_active,
 867                                                       BOND_SLAVE_NOTIFY_NOW);
 868
 869                 if (new_active) {
 870                         bool should_notify_peers = false;
 871
 872                         bond_set_slave_active_flags(new_active,
 873                                                     BOND_SLAVE_NOTIFY_NOW);
 874
 875                         if (bond->params.fail_over_mac)
 876                                 bond_do_fail_over_mac(bond, new_active,
 877                                                       old_active);
 878
 879                         if (netif_running(bond->dev)) {
 880                                 bond->send_peer_notif =
 881                                         bond->params.num_peer_notif;
 882                                 should_notify_peers =
 883                                         bond_should_notify_peers(bond);
 884                         }
 885
 886                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 887                         if (should_notify_peers)
 888                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 889                                                          bond->dev);
 890                 }
 891         }
 892
 893         /* resend IGMP joins since active slave has changed or
 894          * all were sent on curr_active_slave.
 895          * resend only if bond is brought up with the affected
 896          * bonding modes and the retransmission is enabled
 897          */
 898         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 899             ((bond_uses_primary(bond) && new_active) ||
 900              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
 901                 bond->igmp_retrans = bond->params.resend_igmp;
 902                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 903         }
 904 }
 905
 906 /**
 907  * bond_select_active_slave - select a new active slave, if needed
 908  * @bond: our bonding struct
 909  *
 910  * This functions should be called when one of the following occurs:
 911  * - The old curr_active_slave has been released or lost its link.
 912  * - The primary_slave has got its link back.
 913  * - A slave has got its link back and there's no old curr_active_slave.
 914  *
 915  * Caller must hold RTNL.
 916  */
 917 void bond_select_active_slave(struct bonding *bond)
 918 {
 919         struct slave *best_slave;
 920         int rv;
 921
 922         ASSERT_RTNL();
 923
 924         best_slave = bond_find_best_slave(bond);
 925         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
 926                 bond_change_active_slave(bond, best_slave);
 927                 rv = bond_set_carrier(bond);
 928                 if (!rv)
 929                         return;
 930
 931                 if (netif_carrier_ok(bond->dev))
 932                         netdev_info(bond->dev, "first active interface up!\n");
 933                 else
 934                         netdev_info(bond->dev, "now running without any active interface!\n");
 935         }
 936 }
 937
 938 #ifdef CONFIG_NET_POLL_CONTROLLER
 939 static inline int slave_enable_netpoll(struct slave *slave)
 940 {
 941         struct netpoll *np;
 942         int err = 0;
 943
 944         np = kzalloc(sizeof(*np), GFP_KERNEL);
 945         err = -ENOMEM;
 946         if (!np)
 947                 goto out;
 948
 949         err = __netpoll_setup(np, slave->dev);
 950         if (err) {
 951                 kfree(np);
 952                 goto out;
 953         }
 954         slave->np = np;
 955 out:
 956         return err;
 957 }
 958 static inline void slave_disable_netpoll(struct slave *slave)
 959 {
 960         struct netpoll *np = slave->np;
 961
 962         if (!np)
 963                 return;
 964
 965         slave->np = NULL;
 966         __netpoll_free_async(np);
 967 }
 968
 969 static void bond_poll_controller(struct net_device *bond_dev)
 970 {
 971         struct bonding *bond = netdev_priv(bond_dev);
 972         struct slave *slave = NULL;
 973         struct list_head *iter;
 974         struct ad_info ad_info;
 975         struct netpoll_info *ni;
 976         const struct net_device_ops *ops;
 977
 978         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 979                 if (bond_3ad_get_active_agg_info(bond, &ad_info))
 980                         return;
 981
 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 }
1003
1004 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1005 {
1006         struct bonding *bond = netdev_priv(bond_dev);
1007         struct list_head *iter;
1008         struct slave *slave;
1009
1010         bond_for_each_slave(bond, slave, iter)
1011                 if (bond_slave_is_up(slave))
1012                         slave_disable_netpoll(slave);
1013 }
1014
1015 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1016 {
1017         struct bonding *bond = netdev_priv(dev);
1018         struct list_head *iter;
1019         struct slave *slave;
1020         int err = 0;
1021
1022         bond_for_each_slave(bond, slave, iter) {
1023                 err = slave_enable_netpoll(slave);
1024                 if (err) {
1025                         bond_netpoll_cleanup(dev);
1026                         break;
1027                 }
1028         }
1029         return err;
1030 }
1031 #else
1032 static inline int slave_enable_netpoll(struct slave *slave)
1033 {
1034         return 0;
1035 }
1036 static inline void slave_disable_netpoll(struct slave *slave)
1037 {
1038 }
1039 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1040 {
1041 }
1042 #endif
1043
1044 /*---------------------------------- IOCTL ----------------------------------*/
1045
1046 static netdev_features_t bond_fix_features(struct net_device *dev,
1047                                            netdev_features_t features)
1048 {
1049         struct bonding *bond = netdev_priv(dev);
1050         struct list_head *iter;
1051         netdev_features_t mask;
1052         struct slave *slave;
1053
1054         mask = features;
1055
1056         features &= ~NETIF_F_ONE_FOR_ALL;
1057         features |= NETIF_F_ALL_FOR_ALL;
1058
1059         bond_for_each_slave(bond, slave, iter) {
1060                 features = netdev_increment_features(features,
1061                                                      slave->dev->features,
1062                                                      mask);
1063         }
1064         features = netdev_add_tso_features(features, mask);
1065
1066         return features;
1067 }
1068
1069 #define BOND_VLAN_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1070                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1071                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1072
1073 #define BOND_ENC_FEATURES       (NETIF_F_HW_CSUM | NETIF_F_SG | \
1074                                  NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1075
1076 static void bond_compute_features(struct bonding *bond)
1077 {
1078         unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1079                                         IFF_XMIT_DST_RELEASE_PERM;
1080         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1081         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1082         struct net_device *bond_dev = bond->dev;
1083         struct list_head *iter;
1084         struct slave *slave;
1085         unsigned short max_hard_header_len = ETH_HLEN;
1086         unsigned int gso_max_size = GSO_MAX_SIZE;
1087         u16 gso_max_segs = GSO_MAX_SEGS;
1088
1089         if (!bond_has_slaves(bond))
1090                 goto done;
1091         vlan_features &= NETIF_F_ALL_FOR_ALL;
1092
1093         bond_for_each_slave(bond, slave, iter) {
1094                 vlan_features = netdev_increment_features(vlan_features,
1095                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1096
1097                 enc_features = netdev_increment_features(enc_features,
1098                                                          slave->dev->hw_enc_features,
1099                                                          BOND_ENC_FEATURES);
1100                 dst_release_flag &= slave->dev->priv_flags;
1101                 if (slave->dev->hard_header_len > max_hard_header_len)
1102                         max_hard_header_len = slave->dev->hard_header_len;
1103
1104                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1105                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1106         }
1107
1108 done:
1109         bond_dev->vlan_features = vlan_features;
1110         bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
1111         bond_dev->hard_header_len = max_hard_header_len;
1112         bond_dev->gso_max_segs = gso_max_segs;
1113         netif_set_gso_max_size(bond_dev, gso_max_size);
1114
1115         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1116         if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1117             dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1118                 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1119
1120         netdev_change_features(bond_dev);
1121 }
1122
1123 static void bond_setup_by_slave(struct net_device *bond_dev,
1124                                 struct net_device *slave_dev)
1125 {
1126         bond_dev->header_ops        = slave_dev->header_ops;
1127
1128         bond_dev->type              = slave_dev->type;
1129         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1130         bond_dev->addr_len          = slave_dev->addr_len;
1131
1132         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1133                 slave_dev->addr_len);
1134 }
1135
1136 /* On bonding slaves other than the currently active slave, suppress
1137  * duplicates except for alb non-mcast/bcast.
1138  */
1139 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1140                                             struct slave *slave,
1141                                             struct bonding *bond)
1142 {
1143         if (bond_is_slave_inactive(slave)) {
1144                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1145                     skb->pkt_type != PACKET_BROADCAST &&
1146                     skb->pkt_type != PACKET_MULTICAST)
1147                         return false;
1148                 return true;
1149         }
1150         return false;
1151 }
1152
1153 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1154 {
1155         struct sk_buff *skb = *pskb;
1156         struct slave *slave;
1157         struct bonding *bond;
1158         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1159                           struct slave *);
1160         int ret = RX_HANDLER_ANOTHER;
1161
1162         skb = skb_share_check(skb, GFP_ATOMIC);
1163         if (unlikely(!skb))
1164                 return RX_HANDLER_CONSUMED;
1165
1166         *pskb = skb;
1167
1168         slave = bond_slave_get_rcu(skb->dev);
1169         bond = slave->bond;
1170
1171         recv_probe = ACCESS_ONCE(bond->recv_probe);
1172         if (recv_probe) {
1173                 ret = recv_probe(skb, bond, slave);
1174                 if (ret == RX_HANDLER_CONSUMED) {
1175                         consume_skb(skb);
1176                         return ret;
1177                 }
1178         }
1179
1180         if (bond_should_deliver_exact_match(skb, slave, bond))
1181                 return RX_HANDLER_EXACT;
1182
1183         skb->dev = bond->dev;
1184
1185         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1186             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1187             skb->pkt_type == PACKET_HOST) {
1188
1189                 if (unlikely(skb_cow_head(skb,
1190                                           skb->data - skb_mac_header(skb)))) {
1191                         kfree_skb(skb);
1192                         return RX_HANDLER_CONSUMED;
1193                 }
1194                 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1195         }
1196
1197         return ret;
1198 }
1199
1200 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1201 {
1202         switch (BOND_MODE(bond)) {
1203         case BOND_MODE_ROUNDROBIN:
1204                 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1205         case BOND_MODE_ACTIVEBACKUP:
1206                 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1207         case BOND_MODE_BROADCAST:
1208                 return NETDEV_LAG_TX_TYPE_BROADCAST;
1209         case BOND_MODE_XOR:
1210         case BOND_MODE_8023AD:
1211                 return NETDEV_LAG_TX_TYPE_HASH;
1212         default:
1213                 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1214         }
1215 }
1216
1217 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave)
1218 {
1219         struct netdev_lag_upper_info lag_upper_info;
1220         int err;
1221
1222         lag_upper_info.tx_type = bond_lag_tx_type(bond);
1223         err = netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1224                                            &lag_upper_info);
1225         if (err)
1226                 return err;
1227         rtmsg_ifinfo(RTM_NEWLINK, slave->dev, IFF_SLAVE, GFP_KERNEL);
1228         return 0;
1229 }
1230
1231 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1232 {
1233         netdev_upper_dev_unlink(slave->dev, bond->dev);
1234         slave->dev->flags &= ~IFF_SLAVE;
1235         rtmsg_ifinfo(RTM_NEWLINK, slave->dev, IFF_SLAVE, GFP_KERNEL);
1236 }
1237
1238 static struct slave *bond_alloc_slave(struct bonding *bond)
1239 {
1240         struct slave *slave = NULL;
1241
1242         slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1243         if (!slave)
1244                 return NULL;
1245
1246         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1247                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1248                                                GFP_KERNEL);
1249                 if (!SLAVE_AD_INFO(slave)) {
1250                         kfree(slave);
1251                         return NULL;
1252                 }
1253         }
1254         return slave;
1255 }
1256
1257 static void bond_free_slave(struct slave *slave)
1258 {
1259         struct bonding *bond = bond_get_bond_by_slave(slave);
1260
1261         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1262                 kfree(SLAVE_AD_INFO(slave));
1263
1264         kfree(slave);
1265 }
1266
1267 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1268 {
1269         info->bond_mode = BOND_MODE(bond);
1270         info->miimon = bond->params.miimon;
1271         info->num_slaves = bond->slave_cnt;
1272 }
1273
1274 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1275 {
1276         strcpy(info->slave_name, slave->dev->name);
1277         info->link = slave->link;
1278         info->state = bond_slave_state(slave);
1279         info->link_failure_count = slave->link_failure_count;
1280 }
1281
1282 static void bond_netdev_notify(struct net_device *dev,
1283                                struct netdev_bonding_info *info)
1284 {
1285         rtnl_lock();
1286         netdev_bonding_info_change(dev, info);
1287         rtnl_unlock();
1288 }
1289
1290 static void bond_netdev_notify_work(struct work_struct *_work)
1291 {
1292         struct netdev_notify_work *w =
1293                 container_of(_work, struct netdev_notify_work, work.work);
1294
1295         bond_netdev_notify(w->dev, &w->bonding_info);
1296         dev_put(w->dev);
1297         kfree(w);
1298 }
1299
1300 void bond_queue_slave_event(struct slave *slave)
1301 {
1302         struct bonding *bond = slave->bond;
1303         struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1304
1305         if (!nnw)
1306                 return;
1307
1308         dev_hold(slave->dev);
1309         nnw->dev = slave->dev;
1310         bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1311         bond_fill_ifbond(bond, &nnw->bonding_info.master);
1312         INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1313
1314         queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1315 }
1316
1317 void bond_lower_state_changed(struct slave *slave)
1318 {
1319         struct netdev_lag_lower_state_info info;
1320
1321         info.link_up = slave->link == BOND_LINK_UP ||
1322                        slave->link == BOND_LINK_FAIL;
1323         info.tx_enabled = bond_is_active_slave(slave);
1324         netdev_lower_state_changed(slave->dev, &info);
1325 }
1326
1327 /* enslave device <slave> to bond device <master> */
1328 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1329 {
1330         struct bonding *bond = netdev_priv(bond_dev);
1331         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1332         struct slave *new_slave = NULL, *prev_slave;
1333         struct sockaddr addr;
1334         int link_reporting;
1335         int res = 0, i;
1336
1337         if (!bond->params.use_carrier &&
1338             slave_dev->ethtool_ops->get_link == NULL &&
1339             slave_ops->ndo_do_ioctl == NULL) {
1340                 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1341                             slave_dev->name);
1342         }
1343
1344         /* already enslaved */
1345         if (slave_dev->flags & IFF_SLAVE) {
1346                 netdev_dbg(bond_dev, "Error: Device was already enslaved\n");
1347                 return -EBUSY;
1348         }
1349
1350         if (bond_dev == slave_dev) {
1351                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1352                 return -EPERM;
1353         }
1354
1355         /* vlan challenged mutual exclusion */
1356         /* no need to lock since we're protected by rtnl_lock */
1357         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1358                 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1359                            slave_dev->name);
1360                 if (vlan_uses_dev(bond_dev)) {
1361                         netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1362                                    slave_dev->name, bond_dev->name);
1363                         return -EPERM;
1364                 } else {
1365                         netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1366                                     slave_dev->name, slave_dev->name,
1367                                     bond_dev->name);
1368                 }
1369         } else {
1370                 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1371                            slave_dev->name);
1372         }
1373
1374         /* Old ifenslave binaries are no longer supported.  These can
1375          * be identified with moderate accuracy by the state of the slave:
1376          * the current ifenslave will set the interface down prior to
1377          * enslaving it; the old ifenslave will not.
1378          */
1379         if (slave_dev->flags & IFF_UP) {
1380                 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1381                            slave_dev->name);
1382                 return -EPERM;
1383         }
1384
1385         /* set bonding device ether type by slave - bonding netdevices are
1386          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1387          * there is a need to override some of the type dependent attribs/funcs.
1388          *
1389          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1390          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1391          */
1392         if (!bond_has_slaves(bond)) {
1393                 if (bond_dev->type != slave_dev->type) {
1394                         netdev_dbg(bond_dev, "change device type from %d to %d\n",
1395                                    bond_dev->type, slave_dev->type);
1396
1397                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1398                                                        bond_dev);
1399                         res = notifier_to_errno(res);
1400                         if (res) {
1401                                 netdev_err(bond_dev, "refused to change device type\n");
1402                                 return -EBUSY;
1403                         }
1404
1405                         /* Flush unicast and multicast addresses */
1406                         dev_uc_flush(bond_dev);
1407                         dev_mc_flush(bond_dev);
1408
1409                         if (slave_dev->type != ARPHRD_ETHER)
1410                                 bond_setup_by_slave(bond_dev, slave_dev);
1411                         else {
1412                                 ether_setup(bond_dev);
1413                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1414                         }
1415
1416                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1417                                                  bond_dev);
1418                 }
1419         } else if (bond_dev->type != slave_dev->type) {
1420                 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1421                            slave_dev->name, slave_dev->type, bond_dev->type);
1422                 return -EINVAL;
1423         }
1424
1425         if (slave_ops->ndo_set_mac_address == NULL) {
1426                 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1427                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1428                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1429                         if (!bond_has_slaves(bond)) {
1430                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1431                                 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1432                         } else {
1433                                 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");
1434                                 res = -EOPNOTSUPP;
1435                                 goto err_undo_flags;
1436                         }
1437                 }
1438         }
1439
1440         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1441
1442         /* If this is the first slave, then we need to set the master's hardware
1443          * address to be the same as the slave's.
1444          */
1445         if (!bond_has_slaves(bond) &&
1446             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1447                 bond_set_dev_addr(bond->dev, slave_dev);
1448
1449         new_slave = bond_alloc_slave(bond);
1450         if (!new_slave) {
1451                 res = -ENOMEM;
1452                 goto err_undo_flags;
1453         }
1454
1455         new_slave->bond = bond;
1456         new_slave->dev = slave_dev;
1457         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1458          * is set via sysfs or module option if desired.
1459          */
1460         new_slave->queue_id = 0;
1461
1462         /* Save slave's original mtu and then set it to match the bond */
1463         new_slave->original_mtu = slave_dev->mtu;
1464         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1465         if (res) {
1466                 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1467                 goto err_free;
1468         }
1469
1470         /* Save slave's original ("permanent") mac address for modes
1471          * that need it, and for restoring it upon release, and then
1472          * set it to the master's address
1473          */
1474         ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1475
1476         if (!bond->params.fail_over_mac ||
1477             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1478                 /* Set slave to master's mac address.  The application already
1479                  * set the master's mac address to that of the first slave
1480                  */
1481                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1482                 addr.sa_family = slave_dev->type;
1483                 res = dev_set_mac_address(slave_dev, &addr);
1484                 if (res) {
1485                         netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1486                         goto err_restore_mtu;
1487                 }
1488         }
1489
1490         /* set slave flag before open to prevent IPv6 addrconf */
1491         slave_dev->flags |= IFF_SLAVE;
1492
1493         /* open the slave since the application closed it */
1494         res = dev_open(slave_dev);
1495         if (res) {
1496                 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1497                 goto err_restore_mac;
1498         }
1499
1500         slave_dev->priv_flags |= IFF_BONDING;
1501         /* initialize slave stats */
1502         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1503
1504         if (bond_is_lb(bond)) {
1505                 /* bond_alb_init_slave() must be called before all other stages since
1506                  * it might fail and we do not want to have to undo everything
1507                  */
1508                 res = bond_alb_init_slave(bond, new_slave);
1509                 if (res)
1510                         goto err_close;
1511         }
1512
1513         /* If the mode uses primary, then the following is handled by
1514          * bond_change_active_slave().
1515          */
1516         if (!bond_uses_primary(bond)) {
1517                 /* set promiscuity level to new slave */
1518                 if (bond_dev->flags & IFF_PROMISC) {
1519                         res = dev_set_promiscuity(slave_dev, 1);
1520                         if (res)
1521                                 goto err_close;
1522                 }
1523
1524                 /* set allmulti level to new slave */
1525                 if (bond_dev->flags & IFF_ALLMULTI) {
1526                         res = dev_set_allmulti(slave_dev, 1);
1527                         if (res)
1528                                 goto err_close;
1529                 }
1530
1531                 netif_addr_lock_bh(bond_dev);
1532
1533                 dev_mc_sync_multiple(slave_dev, bond_dev);
1534                 dev_uc_sync_multiple(slave_dev, bond_dev);
1535
1536                 netif_addr_unlock_bh(bond_dev);
1537         }
1538
1539         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1540                 /* add lacpdu mc addr to mc list */
1541                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1542
1543                 dev_mc_add(slave_dev, lacpdu_multicast);
1544         }
1545
1546         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1547         if (res) {
1548                 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1549                            slave_dev->name);
1550                 goto err_close;
1551         }
1552
1553         prev_slave = bond_last_slave(bond);
1554
1555         new_slave->delay = 0;
1556         new_slave->link_failure_count = 0;
1557
1558         bond_update_speed_duplex(new_slave);
1559
1560         new_slave->last_rx = jiffies -
1561                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1562         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1563                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1564
1565         if (bond->params.miimon && !bond->params.use_carrier) {
1566                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1567
1568                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1569                         /* miimon is set but a bonded network driver
1570                          * does not support ETHTOOL/MII and
1571                          * arp_interval is not set.  Note: if
1572                          * use_carrier is enabled, we will never go
1573                          * here (because netif_carrier is always
1574                          * supported); thus, we don't need to change
1575                          * the messages for netif_carrier.
1576                          */
1577                         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",
1578                                     slave_dev->name);
1579                 } else if (link_reporting == -1) {
1580                         /* unable get link status using mii/ethtool */
1581                         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",
1582                                     slave_dev->name);
1583                 }
1584         }
1585
1586         /* check for initial state */
1587         new_slave->link = BOND_LINK_NOCHANGE;
1588         if (bond->params.miimon) {
1589                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1590                         if (bond->params.updelay) {
1591                                 bond_set_slave_link_state(new_slave,
1592                                                           BOND_LINK_BACK,
1593                                                           BOND_SLAVE_NOTIFY_NOW);
1594                                 new_slave->delay = bond->params.updelay;
1595                         } else {
1596                                 bond_set_slave_link_state(new_slave,
1597                                                           BOND_LINK_UP,
1598                                                           BOND_SLAVE_NOTIFY_NOW);
1599                         }
1600                 } else {
1601                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1602                                                   BOND_SLAVE_NOTIFY_NOW);
1603                 }
1604         } else if (bond->params.arp_interval) {
1605                 bond_set_slave_link_state(new_slave,
1606                                           (netif_carrier_ok(slave_dev) ?
1607                                           BOND_LINK_UP : BOND_LINK_DOWN),
1608                                           BOND_SLAVE_NOTIFY_NOW);
1609         } else {
1610                 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1611                                           BOND_SLAVE_NOTIFY_NOW);
1612         }
1613
1614         if (new_slave->link != BOND_LINK_DOWN)
1615                 new_slave->last_link_up = jiffies;
1616         netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1617                    new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1618                    (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1619
1620         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1621                 /* if there is a primary slave, remember it */
1622                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1623                         rcu_assign_pointer(bond->primary_slave, new_slave);
1624                         bond->force_primary = true;
1625                 }
1626         }
1627
1628         switch (BOND_MODE(bond)) {
1629         case BOND_MODE_ACTIVEBACKUP:
1630                 bond_set_slave_inactive_flags(new_slave,
1631                                               BOND_SLAVE_NOTIFY_NOW);
1632                 break;
1633         case BOND_MODE_8023AD:
1634                 /* in 802.3ad mode, the internal mechanism
1635                  * will activate the slaves in the selected
1636                  * aggregator
1637                  */
1638                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1639                 /* if this is the first slave */
1640                 if (!prev_slave) {
1641                         SLAVE_AD_INFO(new_slave)->id = 1;
1642                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1643                          * can be called only after the mac address of the bond is set
1644                          */
1645                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1646                 } else {
1647                         SLAVE_AD_INFO(new_slave)->id =
1648                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1649                 }
1650
1651                 bond_3ad_bind_slave(new_slave);
1652                 break;
1653         case BOND_MODE_TLB:
1654         case BOND_MODE_ALB:
1655                 bond_set_active_slave(new_slave);
1656                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1657                 break;
1658         default:
1659                 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1660
1661                 /* always active in trunk mode */
1662                 bond_set_active_slave(new_slave);
1663
1664                 /* In trunking mode there is little meaning to curr_active_slave
1665                  * anyway (it holds no special properties of the bond device),
1666                  * so we can change it without calling change_active_interface()
1667                  */
1668                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1669                     new_slave->link == BOND_LINK_UP)
1670                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1671
1672                 break;
1673         } /* switch(bond_mode) */
1674
1675 #ifdef CONFIG_NET_POLL_CONTROLLER
1676         slave_dev->npinfo = bond->dev->npinfo;
1677         if (slave_dev->npinfo) {
1678                 if (slave_enable_netpoll(new_slave)) {
1679                         netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1680                         res = -EBUSY;
1681                         goto err_detach;
1682                 }
1683         }
1684 #endif
1685
1686         if (!(bond_dev->features & NETIF_F_LRO))
1687                 dev_disable_lro(slave_dev);
1688
1689         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1690                                          new_slave);
1691         if (res) {
1692                 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1693                 goto err_detach;
1694         }
1695
1696         res = bond_master_upper_dev_link(bond, new_slave);
1697         if (res) {
1698                 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1699                 goto err_unregister;
1700         }
1701
1702         res = bond_sysfs_slave_add(new_slave);
1703         if (res) {
1704                 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1705                 goto err_upper_unlink;
1706         }
1707
1708         bond->slave_cnt++;
1709         bond_compute_features(bond);
1710         bond_set_carrier(bond);
1711
1712         if (bond_uses_primary(bond)) {
1713                 block_netpoll_tx();
1714                 bond_select_active_slave(bond);
1715                 unblock_netpoll_tx();
1716         }
1717
1718         if (bond_mode_uses_xmit_hash(bond))
1719                 bond_update_slave_arr(bond, NULL);
1720
1721         netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1722                     slave_dev->name,
1723                     bond_is_active_slave(new_slave) ? "an active" : "a backup",
1724                     new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1725
1726         /* enslave is successful */
1727         bond_queue_slave_event(new_slave);
1728         return 0;
1729
1730 /* Undo stages on error */
1731 err_upper_unlink:
1732         bond_upper_dev_unlink(bond, new_slave);
1733
1734 err_unregister:
1735         netdev_rx_handler_unregister(slave_dev);
1736
1737 err_detach:
1738         if (!bond_uses_primary(bond))
1739                 bond_hw_addr_flush(bond_dev, slave_dev);
1740
1741         vlan_vids_del_by_dev(slave_dev, bond_dev);
1742         if (rcu_access_pointer(bond->primary_slave) == new_slave)
1743                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1744         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1745                 block_netpoll_tx();
1746                 bond_change_active_slave(bond, NULL);
1747                 bond_select_active_slave(bond);
1748                 unblock_netpoll_tx();
1749         }
1750         /* either primary_slave or curr_active_slave might've changed */
1751         synchronize_rcu();
1752         slave_disable_netpoll(new_slave);
1753
1754 err_close:
1755         slave_dev->priv_flags &= ~IFF_BONDING;
1756         dev_close(slave_dev);
1757
1758 err_restore_mac:
1759         slave_dev->flags &= ~IFF_SLAVE;
1760         if (!bond->params.fail_over_mac ||
1761             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1762                 /* XXX TODO - fom follow mode needs to change master's
1763                  * MAC if this slave's MAC is in use by the bond, or at
1764                  * least print a warning.
1765                  */
1766                 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1767                 addr.sa_family = slave_dev->type;
1768                 dev_set_mac_address(slave_dev, &addr);
1769         }
1770
1771 err_restore_mtu:
1772         dev_set_mtu(slave_dev, new_slave->original_mtu);
1773
1774 err_free:
1775         bond_free_slave(new_slave);
1776
1777 err_undo_flags:
1778         /* Enslave of first slave has failed and we need to fix master's mac */
1779         if (!bond_has_slaves(bond)) {
1780                 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1781                                             slave_dev->dev_addr))
1782                         eth_hw_addr_random(bond_dev);
1783                 if (bond_dev->type != ARPHRD_ETHER) {
1784                         dev_close(bond_dev);
1785                         ether_setup(bond_dev);
1786                         bond_dev->flags |= IFF_MASTER;
1787                         bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1788                 }
1789         }
1790
1791         return res;
1792 }
1793
1794 /* Try to release the slave device <slave> from the bond device <master>
1795  * It is legal to access curr_active_slave without a lock because all the function
1796  * is RTNL-locked. If "all" is true it means that the function is being called
1797  * while destroying a bond interface and all slaves are being released.
1798  *
1799  * The rules for slave state should be:
1800  *   for Active/Backup:
1801  *     Active stays on all backups go down
1802  *   for Bonded connections:
1803  *     The first up interface should be left on and all others downed.
1804  */
1805 static int __bond_release_one(struct net_device *bond_dev,
1806                               struct net_device *slave_dev,
1807                               bool all)
1808 {
1809         struct bonding *bond = netdev_priv(bond_dev);
1810         struct slave *slave, *oldcurrent;
1811         struct sockaddr addr;
1812         int old_flags = bond_dev->flags;
1813         netdev_features_t old_features = bond_dev->features;
1814
1815         /* slave is not a slave or master is not master of this slave */
1816         if (!(slave_dev->flags & IFF_SLAVE) ||
1817             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1818                 netdev_dbg(bond_dev, "cannot release %s\n",
1819                            slave_dev->name);
1820                 return -EINVAL;
1821         }
1822
1823         block_netpoll_tx();
1824
1825         slave = bond_get_slave_by_dev(bond, slave_dev);
1826         if (!slave) {
1827                 /* not a slave of this bond */
1828                 netdev_info(bond_dev, "%s not enslaved\n",
1829                             slave_dev->name);
1830                 unblock_netpoll_tx();
1831                 return -EINVAL;
1832         }
1833
1834         bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1835
1836         bond_sysfs_slave_del(slave);
1837
1838         /* recompute stats just before removing the slave */
1839         bond_get_stats(bond->dev, &bond->bond_stats);
1840
1841         bond_upper_dev_unlink(bond, slave);
1842         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1843          * for this slave anymore.
1844          */
1845         netdev_rx_handler_unregister(slave_dev);
1846
1847         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1848                 bond_3ad_unbind_slave(slave);
1849
1850         if (bond_mode_uses_xmit_hash(bond))
1851                 bond_update_slave_arr(bond, slave);
1852
1853         netdev_info(bond_dev, "Releasing %s interface %s\n",
1854                     bond_is_active_slave(slave) ? "active" : "backup",
1855                     slave_dev->name);
1856
1857         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1858
1859         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1860
1861         if (!all && (!bond->params.fail_over_mac ||
1862                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1863                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1864                     bond_has_slaves(bond))
1865                         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",
1866                                     slave_dev->name, slave->perm_hwaddr,
1867                                     bond_dev->name, slave_dev->name);
1868         }
1869
1870         if (rtnl_dereference(bond->primary_slave) == slave)
1871                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1872
1873         if (oldcurrent == slave)
1874                 bond_change_active_slave(bond, NULL);
1875
1876         if (bond_is_lb(bond)) {
1877                 /* Must be called only after the slave has been
1878                  * detached from the list and the curr_active_slave
1879                  * has been cleared (if our_slave == old_current),
1880                  * but before a new active slave is selected.
1881                  */
1882                 bond_alb_deinit_slave(bond, slave);
1883         }
1884
1885         if (all) {
1886                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1887         } else if (oldcurrent == slave) {
1888                 /* Note that we hold RTNL over this sequence, so there
1889                  * is no concern that another slave add/remove event
1890                  * will interfere.
1891                  */
1892                 bond_select_active_slave(bond);
1893         }
1894
1895         if (!bond_has_slaves(bond)) {
1896                 bond_set_carrier(bond);
1897                 eth_hw_addr_random(bond_dev);
1898         }
1899
1900         unblock_netpoll_tx();
1901         synchronize_rcu();
1902         bond->slave_cnt--;
1903
1904         if (!bond_has_slaves(bond)) {
1905                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1906                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1907         }
1908
1909         bond_compute_features(bond);
1910         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1911             (old_features & NETIF_F_VLAN_CHALLENGED))
1912                 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1913                             slave_dev->name, bond_dev->name);
1914
1915         vlan_vids_del_by_dev(slave_dev, bond_dev);
1916
1917         /* If the mode uses primary, then this case was handled above by
1918          * bond_change_active_slave(..., NULL)
1919          */
1920         if (!bond_uses_primary(bond)) {
1921                 /* unset promiscuity level from slave
1922                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1923                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1924                  * value of that flag before that change, as that was the value
1925                  * when this slave was attached, so we cache at the start of the
1926                  * function and use it here. Same goes for ALLMULTI below
1927                  */
1928                 if (old_flags & IFF_PROMISC)
1929                         dev_set_promiscuity(slave_dev, -1);
1930
1931                 /* unset allmulti level from slave */
1932                 if (old_flags & IFF_ALLMULTI)
1933                         dev_set_allmulti(slave_dev, -1);
1934
1935                 bond_hw_addr_flush(bond_dev, slave_dev);
1936         }
1937
1938         slave_disable_netpoll(slave);
1939
1940         /* close slave before restoring its mac address */
1941         dev_close(slave_dev);
1942
1943         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1944             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1945                 /* restore original ("permanent") mac address */
1946                 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1947                 addr.sa_family = slave_dev->type;
1948                 dev_set_mac_address(slave_dev, &addr);
1949         }
1950
1951         dev_set_mtu(slave_dev, slave->original_mtu);
1952
1953         slave_dev->priv_flags &= ~IFF_BONDING;
1954
1955         bond_free_slave(slave);
1956
1957         return 0;
1958 }
1959
1960 /* A wrapper used because of ndo_del_link */
1961 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1962 {
1963         return __bond_release_one(bond_dev, slave_dev, false);
1964 }
1965
1966 /* First release a slave and then destroy the bond if no more slaves are left.
1967  * Must be under rtnl_lock when this function is called.
1968  */
1969 static int  bond_release_and_destroy(struct net_device *bond_dev,
1970                                      struct net_device *slave_dev)
1971 {
1972         struct bonding *bond = netdev_priv(bond_dev);
1973         int ret;
1974
1975         ret = bond_release(bond_dev, slave_dev);
1976         if (ret == 0 && !bond_has_slaves(bond)) {
1977                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1978                 netdev_info(bond_dev, "Destroying bond %s\n",
1979                             bond_dev->name);
1980                 bond_remove_proc_entry(bond);
1981                 unregister_netdevice(bond_dev);
1982         }
1983         return ret;
1984 }
1985
1986 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1987 {
1988         struct bonding *bond = netdev_priv(bond_dev);
1989         bond_fill_ifbond(bond, info);
1990         return 0;
1991 }
1992
1993 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1994 {
1995         struct bonding *bond = netdev_priv(bond_dev);
1996         struct list_head *iter;
1997         int i = 0, res = -ENODEV;
1998         struct slave *slave;
1999
2000         bond_for_each_slave(bond, slave, iter) {
2001                 if (i++ == (int)info->slave_id) {
2002                         res = 0;
2003                         bond_fill_ifslave(slave, info);
2004                         break;
2005                 }
2006         }
2007
2008         return res;
2009 }
2010
2011 /*-------------------------------- Monitoring -------------------------------*/
2012
2013 /* called with rcu_read_lock() */
2014 static int bond_miimon_inspect(struct bonding *bond)
2015 {
2016         int link_state, commit = 0;
2017         struct list_head *iter;
2018         struct slave *slave;
2019         bool ignore_updelay;
2020
2021         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2022
2023         bond_for_each_slave_rcu(bond, slave, iter) {
2024                 slave->new_link = BOND_LINK_NOCHANGE;
2025
2026                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2027
2028                 switch (slave->link) {
2029                 case BOND_LINK_UP:
2030                         if (link_state)
2031                                 continue;
2032
2033                         bond_set_slave_link_state(slave, BOND_LINK_FAIL,
2034                                                   BOND_SLAVE_NOTIFY_LATER);
2035                         slave->delay = bond->params.downdelay;
2036                         if (slave->delay) {
2037                                 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2038                                             (BOND_MODE(bond) ==
2039                                              BOND_MODE_ACTIVEBACKUP) ?
2040                                              (bond_is_active_slave(slave) ?
2041                                               "active " : "backup ") : "",
2042                                             slave->dev->name,
2043                                             bond->params.downdelay * bond->params.miimon);
2044                         }
2045                         /*FALLTHRU*/
2046                 case BOND_LINK_FAIL:
2047                         if (link_state) {
2048                                 /* recovered before downdelay expired */
2049                                 bond_set_slave_link_state(slave, BOND_LINK_UP,
2050                                                           BOND_SLAVE_NOTIFY_LATER);
2051                                 slave->last_link_up = jiffies;
2052                                 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2053                                             (bond->params.downdelay - slave->delay) *
2054                                             bond->params.miimon,
2055                                             slave->dev->name);
2056                                 continue;
2057                         }
2058
2059                         if (slave->delay <= 0) {
2060                                 slave->new_link = BOND_LINK_DOWN;
2061                                 commit++;
2062                                 continue;
2063                         }
2064
2065                         slave->delay--;
2066                         break;
2067
2068                 case BOND_LINK_DOWN:
2069                         if (!link_state)
2070                                 continue;
2071
2072                         bond_set_slave_link_state(slave, BOND_LINK_BACK,
2073                                                   BOND_SLAVE_NOTIFY_LATER);
2074                         slave->delay = bond->params.updelay;
2075
2076                         if (slave->delay) {
2077                                 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2078                                             slave->dev->name,
2079                                             ignore_updelay ? 0 :
2080                                             bond->params.updelay *
2081                                             bond->params.miimon);
2082                         }
2083                         /*FALLTHRU*/
2084                 case BOND_LINK_BACK:
2085                         if (!link_state) {
2086                                 bond_set_slave_link_state(slave,
2087                                                           BOND_LINK_DOWN,
2088                                                           BOND_SLAVE_NOTIFY_LATER);
2089                                 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2090                                             (bond->params.updelay - slave->delay) *
2091                                             bond->params.miimon,
2092                                             slave->dev->name);
2093
2094                                 continue;
2095                         }
2096
2097                         if (ignore_updelay)
2098                                 slave->delay = 0;
2099
2100                         if (slave->delay <= 0) {
2101                                 slave->new_link = BOND_LINK_UP;
2102                                 commit++;
2103                                 ignore_updelay = false;
2104                                 continue;
2105                         }
2106
2107                         slave->delay--;
2108                         break;
2109                 }
2110         }
2111
2112         return commit;
2113 }
2114
2115 static void bond_miimon_commit(struct bonding *bond)
2116 {
2117         struct list_head *iter;
2118         struct slave *slave, *primary;
2119
2120         bond_for_each_slave(bond, slave, iter) {
2121                 switch (slave->new_link) {
2122                 case BOND_LINK_NOCHANGE:
2123                         continue;
2124
2125                 case BOND_LINK_UP:
2126                         bond_update_speed_duplex(slave);
2127                         bond_set_slave_link_state(slave, BOND_LINK_UP,
2128                                                   BOND_SLAVE_NOTIFY_NOW);
2129                         slave->last_link_up = jiffies;
2130
2131                         primary = rtnl_dereference(bond->primary_slave);
2132                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2133                                 /* prevent it from being the active one */
2134                                 bond_set_backup_slave(slave);
2135                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2136                                 /* make it immediately active */
2137                                 bond_set_active_slave(slave);
2138                         } else if (slave != primary) {
2139                                 /* prevent it from being the active one */
2140                                 bond_set_backup_slave(slave);
2141                         }
2142
2143                         netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2144                                     slave->dev->name,
2145                                     slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2146                                     slave->duplex ? "full" : "half");
2147
2148                         /* notify ad that the link status has changed */
2149                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2150                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2151
2152                         if (bond_is_lb(bond))
2153                                 bond_alb_handle_link_change(bond, slave,
2154                                                             BOND_LINK_UP);
2155
2156                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2157                                 bond_update_slave_arr(bond, NULL);
2158
2159                         if (!bond->curr_active_slave || slave == primary)
2160                                 goto do_failover;
2161
2162                         continue;
2163
2164                 case BOND_LINK_DOWN:
2165                         if (slave->link_failure_count < UINT_MAX)
2166                                 slave->link_failure_count++;
2167
2168                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2169                                                   BOND_SLAVE_NOTIFY_NOW);
2170
2171                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2172                             BOND_MODE(bond) == BOND_MODE_8023AD)
2173                                 bond_set_slave_inactive_flags(slave,
2174                                                               BOND_SLAVE_NOTIFY_NOW);
2175
2176                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2177                                     slave->dev->name);
2178
2179                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2180                                 bond_3ad_handle_link_change(slave,
2181                                                             BOND_LINK_DOWN);
2182
2183                         if (bond_is_lb(bond))
2184                                 bond_alb_handle_link_change(bond, slave,
2185                                                             BOND_LINK_DOWN);
2186
2187                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2188                                 bond_update_slave_arr(bond, NULL);
2189
2190                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2191                                 goto do_failover;
2192
2193                         continue;
2194
2195                 default:
2196                         netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2197                                    slave->new_link, slave->dev->name);
2198                         slave->new_link = BOND_LINK_NOCHANGE;
2199
2200                         continue;
2201                 }
2202
2203 do_failover:
2204                 block_netpoll_tx();
2205                 bond_select_active_slave(bond);
2206                 unblock_netpoll_tx();
2207         }
2208
2209         bond_set_carrier(bond);
2210 }
2211
2212 /* bond_mii_monitor
2213  *
2214  * Really a wrapper that splits the mii monitor into two phases: an
2215  * inspection, then (if inspection indicates something needs to be done)
2216  * an acquisition of appropriate locks followed by a commit phase to
2217  * implement whatever link state changes are indicated.
2218  */
2219 static void bond_mii_monitor(struct work_struct *work)
2220 {
2221         struct bonding *bond = container_of(work, struct bonding,
2222                                             mii_work.work);
2223         bool should_notify_peers = false;
2224         unsigned long delay;
2225
2226         delay = msecs_to_jiffies(bond->params.miimon);
2227
2228         if (!bond_has_slaves(bond))
2229                 goto re_arm;
2230
2231         rcu_read_lock();
2232
2233         should_notify_peers = bond_should_notify_peers(bond);
2234
2235         if (bond_miimon_inspect(bond)) {
2236                 rcu_read_unlock();
2237
2238                 /* Race avoidance with bond_close cancel of workqueue */
2239                 if (!rtnl_trylock()) {
2240                         delay = 1;
2241                         should_notify_peers = false;
2242                         goto re_arm;
2243                 }
2244
2245                 bond_miimon_commit(bond);
2246
2247                 rtnl_unlock();  /* might sleep, hold no other locks */
2248         } else
2249                 rcu_read_unlock();
2250
2251 re_arm:
2252         if (bond->params.miimon)
2253                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2254
2255         if (should_notify_peers) {
2256                 if (!rtnl_trylock())
2257                         return;
2258                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2259                 rtnl_unlock();
2260         }
2261 }
2262
2263 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2264 {
2265         struct net_device *upper;
2266         struct list_head *iter;
2267         bool ret = false;
2268
2269         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2270                 return true;
2271
2272         rcu_read_lock();
2273         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2274                 if (ip == bond_confirm_addr(upper, 0, ip)) {
2275                         ret = true;
2276                         break;
2277                 }
2278         }
2279         rcu_read_unlock();
2280
2281         return ret;
2282 }
2283
2284 /* We go to the (large) trouble of VLAN tagging ARP frames because
2285  * switches in VLAN mode (especially if ports are configured as
2286  * "native" to a VLAN) might not pass non-tagged frames.
2287  */
2288 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2289                           __be32 dest_ip, __be32 src_ip,
2290                           struct bond_vlan_tag *tags)
2291 {
2292         struct sk_buff *skb;
2293         struct bond_vlan_tag *outer_tag = tags;
2294
2295         netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2296                    arp_op, slave_dev->name, &dest_ip, &src_ip);
2297
2298         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2299                          NULL, slave_dev->dev_addr, NULL);
2300
2301         if (!skb) {
2302                 net_err_ratelimited("ARP packet allocation failed\n");
2303                 return;
2304         }
2305
2306         if (!tags || tags->vlan_proto == VLAN_N_VID)
2307                 goto xmit;
2308
2309         tags++;
2310
2311         /* Go through all the tags backwards and add them to the packet */
2312         while (tags->vlan_proto != VLAN_N_VID) {
2313                 if (!tags->vlan_id) {
2314                         tags++;
2315                         continue;
2316                 }
2317
2318                 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2319                            ntohs(outer_tag->vlan_proto), tags->vlan_id);
2320                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2321                                                 tags->vlan_id);
2322                 if (!skb) {
2323                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2324                         return;
2325                 }
2326
2327                 tags++;
2328         }
2329         /* Set the outer tag */
2330         if (outer_tag->vlan_id) {
2331                 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2332                            ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2333                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2334                                        outer_tag->vlan_id);
2335         }
2336
2337 xmit:
2338         arp_xmit(skb);
2339 }
2340
2341 /* Validate the device path between the @start_dev and the @end_dev.
2342  * The path is valid if the @end_dev is reachable through device
2343  * stacking.
2344  * When the path is validated, collect any vlan information in the
2345  * path.
2346  */
2347 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2348                                               struct net_device *end_dev,
2349                                               int level)
2350 {
2351         struct bond_vlan_tag *tags;
2352         struct net_device *upper;
2353         struct list_head  *iter;
2354
2355         if (start_dev == end_dev) {
2356                 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2357                 if (!tags)
2358                         return ERR_PTR(-ENOMEM);
2359                 tags[level].vlan_proto = VLAN_N_VID;
2360                 return tags;
2361         }
2362
2363         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2364                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2365                 if (IS_ERR_OR_NULL(tags)) {
2366                         if (IS_ERR(tags))
2367                                 return tags;
2368                         continue;
2369                 }
2370                 if (is_vlan_dev(upper)) {
2371                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2372                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2373                 }
2374
2375                 return tags;
2376         }
2377
2378         return NULL;
2379 }
2380
2381 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2382 {
2383         struct rtable *rt;
2384         struct bond_vlan_tag *tags;
2385         __be32 *targets = bond->params.arp_targets, addr;
2386         int i;
2387
2388         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2389                 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2390                 tags = NULL;
2391
2392                 /* Find out through which dev should the packet go */
2393                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2394                                      RTO_ONLINK, 0);
2395                 if (IS_ERR(rt)) {
2396                         /* there's no route to target - try to send arp
2397                          * probe to generate any traffic (arp_validate=0)
2398                          */
2399                         if (bond->params.arp_validate)
2400                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2401                                                      bond->dev->name,
2402                                                      &targets[i]);
2403                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2404                                       0, tags);
2405                         continue;
2406                 }
2407
2408                 /* bond device itself */
2409                 if (rt->dst.dev == bond->dev)
2410                         goto found;
2411
2412                 rcu_read_lock();
2413                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2414                 rcu_read_unlock();
2415
2416                 if (!IS_ERR_OR_NULL(tags))
2417                         goto found;
2418
2419                 /* Not our device - skip */
2420                 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2421                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2422
2423                 ip_rt_put(rt);
2424                 continue;
2425
2426 found:
2427                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2428                 ip_rt_put(rt);
2429                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2430                               addr, tags);
2431                 kfree(tags);
2432         }
2433 }
2434
2435 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2436 {
2437         int i;
2438
2439         if (!sip || !bond_has_this_ip(bond, tip)) {
2440                 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2441                            &sip, &tip);
2442                 return;
2443         }
2444
2445         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2446         if (i == -1) {
2447                 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2448                            &sip);
2449                 return;
2450         }
2451         slave->last_rx = jiffies;
2452         slave->target_last_arp_rx[i] = jiffies;
2453 }
2454
2455 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2456                  struct slave *slave)
2457 {
2458         struct arphdr *arp = (struct arphdr *)skb->data;
2459         struct slave *curr_active_slave, *curr_arp_slave;
2460         unsigned char *arp_ptr;
2461         __be32 sip, tip;
2462         int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2463
2464         if (!slave_do_arp_validate(bond, slave)) {
2465                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2466                     !slave_do_arp_validate_only(bond))
2467                         slave->last_rx = jiffies;
2468                 return RX_HANDLER_ANOTHER;
2469         } else if (!is_arp) {
2470                 return RX_HANDLER_ANOTHER;
2471         }
2472
2473         alen = arp_hdr_len(bond->dev);
2474
2475         netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2476                    skb->dev->name);
2477
2478         if (alen > skb_headlen(skb)) {
2479                 arp = kmalloc(alen, GFP_ATOMIC);
2480                 if (!arp)
2481                         goto out_unlock;
2482                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2483                         goto out_unlock;
2484         }
2485
2486         if (arp->ar_hln != bond->dev->addr_len ||
2487             skb->pkt_type == PACKET_OTHERHOST ||
2488             skb->pkt_type == PACKET_LOOPBACK ||
2489             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2490             arp->ar_pro != htons(ETH_P_IP) ||
2491             arp->ar_pln != 4)
2492                 goto out_unlock;
2493
2494         arp_ptr = (unsigned char *)(arp + 1);
2495         arp_ptr += bond->dev->addr_len;
2496         memcpy(&sip, arp_ptr, 4);
2497         arp_ptr += 4 + bond->dev->addr_len;
2498         memcpy(&tip, arp_ptr, 4);
2499
2500         netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2501                    slave->dev->name, bond_slave_state(slave),
2502                      bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2503                      &sip, &tip);
2504
2505         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2506         curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2507
2508         /* We 'trust' the received ARP enough to validate it if:
2509          *
2510          * (a) the slave receiving the ARP is active (which includes the
2511          * current ARP slave, if any), or
2512          *
2513          * (b) the receiving slave isn't active, but there is a currently
2514          * active slave and it received valid arp reply(s) after it became
2515          * the currently active slave, or
2516          *
2517          * (c) there is an ARP slave that sent an ARP during the prior ARP
2518          * interval, and we receive an ARP reply on any slave.  We accept
2519          * these because switch FDB update delays may deliver the ARP
2520          * reply to a slave other than the sender of the ARP request.
2521          *
2522          * Note: for (b), backup slaves are receiving the broadcast ARP
2523          * request, not a reply.  This request passes from the sending
2524          * slave through the L2 switch(es) to the receiving slave.  Since
2525          * this is checking the request, sip/tip are swapped for
2526          * validation.
2527          *
2528          * This is done to avoid endless looping when we can't reach the
2529          * arp_ip_target and fool ourselves with our own arp requests.
2530          */
2531         if (bond_is_active_slave(slave))
2532                 bond_validate_arp(bond, slave, sip, tip);
2533         else if (curr_active_slave &&
2534                  time_after(slave_last_rx(bond, curr_active_slave),
2535                             curr_active_slave->last_link_up))
2536                 bond_validate_arp(bond, slave, tip, sip);
2537         else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2538                  bond_time_in_interval(bond,
2539                                        dev_trans_start(curr_arp_slave->dev), 1))
2540                 bond_validate_arp(bond, slave, sip, tip);
2541
2542 out_unlock:
2543         if (arp != (struct arphdr *)skb->data)
2544                 kfree(arp);
2545         return RX_HANDLER_ANOTHER;
2546 }
2547
2548 /* function to verify if we're in the arp_interval timeslice, returns true if
2549  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2550  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2551  */
2552 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2553                                   int mod)
2554 {
2555         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2556
2557         return time_in_range(jiffies,
2558                              last_act - delta_in_ticks,
2559                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2560 }
2561
2562 /* This function is called regularly to monitor each slave's link
2563  * ensuring that traffic is being sent and received when arp monitoring
2564  * is used in load-balancing mode. if the adapter has been dormant, then an
2565  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2566  * arp monitoring in active backup mode.
2567  */
2568 static void bond_loadbalance_arp_mon(struct work_struct *work)
2569 {
2570         struct bonding *bond = container_of(work, struct bonding,
2571                                             arp_work.work);
2572         struct slave *slave, *oldcurrent;
2573         struct list_head *iter;
2574         int do_failover = 0, slave_state_changed = 0;
2575
2576         if (!bond_has_slaves(bond))
2577                 goto re_arm;
2578
2579         rcu_read_lock();
2580
2581         oldcurrent = rcu_dereference(bond->curr_active_slave);
2582         /* see if any of the previous devices are up now (i.e. they have
2583          * xmt and rcv traffic). the curr_active_slave does not come into
2584          * the picture unless it is null. also, slave->last_link_up is not
2585          * needed here because we send an arp on each slave and give a slave
2586          * as long as it needs to get the tx/rx within the delta.
2587          * TODO: what about up/down delay in arp mode? it wasn't here before
2588          *       so it can wait
2589          */
2590         bond_for_each_slave_rcu(bond, slave, iter) {
2591                 unsigned long trans_start = dev_trans_start(slave->dev);
2592
2593                 if (slave->link != BOND_LINK_UP) {
2594                         if (bond_time_in_interval(bond, trans_start, 1) &&
2595                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2596
2597                                 slave->link  = BOND_LINK_UP;
2598                                 slave_state_changed = 1;
2599
2600                                 /* primary_slave has no meaning in round-robin
2601                                  * mode. the window of a slave being up and
2602                                  * curr_active_slave being null after enslaving
2603                                  * is closed.
2604                                  */
2605                                 if (!oldcurrent) {
2606                                         netdev_info(bond->dev, "link status definitely up for interface %s\n",
2607                                                     slave->dev->name);
2608                                         do_failover = 1;
2609                                 } else {
2610                                         netdev_info(bond->dev, "interface %s is now up\n",
2611                                                     slave->dev->name);
2612                                 }
2613                         }
2614                 } else {
2615                         /* slave->link == BOND_LINK_UP */
2616
2617                         /* not all switches will respond to an arp request
2618                          * when the source ip is 0, so don't take the link down
2619                          * if we don't know our ip yet
2620                          */
2621                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2622                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2623
2624                                 slave->link  = BOND_LINK_DOWN;
2625                                 slave_state_changed = 1;
2626
2627                                 if (slave->link_failure_count < UINT_MAX)
2628                                         slave->link_failure_count++;
2629
2630                                 netdev_info(bond->dev, "interface %s is now down\n",
2631                                             slave->dev->name);
2632
2633                                 if (slave == oldcurrent)
2634                                         do_failover = 1;
2635                         }
2636                 }
2637
2638                 /* note: if switch is in round-robin mode, all links
2639                  * must tx arp to ensure all links rx an arp - otherwise
2640                  * links may oscillate or not come up at all; if switch is
2641                  * in something like xor mode, there is nothing we can
2642                  * do - all replies will be rx'ed on same link causing slaves
2643                  * to be unstable during low/no traffic periods
2644                  */
2645                 if (bond_slave_is_up(slave))
2646                         bond_arp_send_all(bond, slave);
2647         }
2648
2649         rcu_read_unlock();
2650
2651         if (do_failover || slave_state_changed) {
2652                 if (!rtnl_trylock())
2653                         goto re_arm;
2654
2655                 if (slave_state_changed) {
2656                         bond_slave_state_change(bond);
2657                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2658                                 bond_update_slave_arr(bond, NULL);
2659                 }
2660                 if (do_failover) {
2661                         block_netpoll_tx();
2662                         bond_select_active_slave(bond);
2663                         unblock_netpoll_tx();
2664                 }
2665                 rtnl_unlock();
2666         }
2667
2668 re_arm:
2669         if (bond->params.arp_interval)
2670                 queue_delayed_work(bond->wq, &bond->arp_work,
2671                                    msecs_to_jiffies(bond->params.arp_interval));
2672 }
2673
2674 /* Called to inspect slaves for active-backup mode ARP monitor link state
2675  * changes.  Sets new_link in slaves to specify what action should take
2676  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2677  * to link states must be committed.
2678  *
2679  * Called with rcu_read_lock held.
2680  */
2681 static int bond_ab_arp_inspect(struct bonding *bond)
2682 {
2683         unsigned long trans_start, last_rx;
2684         struct list_head *iter;
2685         struct slave *slave;
2686         int commit = 0;
2687
2688         bond_for_each_slave_rcu(bond, slave, iter) {
2689                 slave->new_link = BOND_LINK_NOCHANGE;
2690                 last_rx = slave_last_rx(bond, slave);
2691
2692                 if (slave->link != BOND_LINK_UP) {
2693                         if (bond_time_in_interval(bond, last_rx, 1)) {
2694                                 slave->new_link = BOND_LINK_UP;
2695                                 commit++;
2696                         }
2697                         continue;
2698                 }
2699
2700                 /* Give slaves 2*delta after being enslaved or made
2701                  * active.  This avoids bouncing, as the last receive
2702                  * times need a full ARP monitor cycle to be updated.
2703                  */
2704                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2705                         continue;
2706
2707                 /* Backup slave is down if:
2708                  * - No current_arp_slave AND
2709                  * - more than 3*delta since last receive AND
2710                  * - the bond has an IP address
2711                  *
2712                  * Note: a non-null current_arp_slave indicates
2713                  * the curr_active_slave went down and we are
2714                  * searching for a new one; under this condition
2715                  * we only take the curr_active_slave down - this
2716                  * gives each slave a chance to tx/rx traffic
2717                  * before being taken out
2718                  */
2719                 if (!bond_is_active_slave(slave) &&
2720                     !rcu_access_pointer(bond->current_arp_slave) &&
2721                     !bond_time_in_interval(bond, last_rx, 3)) {
2722                         slave->new_link = BOND_LINK_DOWN;
2723                         commit++;
2724                 }
2725
2726                 /* Active slave is down if:
2727                  * - more than 2*delta since transmitting OR
2728                  * - (more than 2*delta since receive AND
2729                  *    the bond has an IP address)
2730                  */
2731                 trans_start = dev_trans_start(slave->dev);
2732                 if (bond_is_active_slave(slave) &&
2733                     (!bond_time_in_interval(bond, trans_start, 2) ||
2734                      !bond_time_in_interval(bond, last_rx, 2))) {
2735                         slave->new_link = BOND_LINK_DOWN;
2736                         commit++;
2737                 }
2738         }
2739
2740         return commit;
2741 }
2742
2743 /* Called to commit link state changes noted by inspection step of
2744  * active-backup mode ARP monitor.
2745  *
2746  * Called with RTNL hold.
2747  */
2748 static void bond_ab_arp_commit(struct bonding *bond)
2749 {
2750         unsigned long trans_start;
2751         struct list_head *iter;
2752         struct slave *slave;
2753
2754         bond_for_each_slave(bond, slave, iter) {
2755                 switch (slave->new_link) {
2756                 case BOND_LINK_NOCHANGE:
2757                         continue;
2758
2759                 case BOND_LINK_UP:
2760                         trans_start = dev_trans_start(slave->dev);
2761                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
2762                             (!rtnl_dereference(bond->curr_active_slave) &&
2763                              bond_time_in_interval(bond, trans_start, 1))) {
2764                                 struct slave *current_arp_slave;
2765
2766                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2767                                 bond_set_slave_link_state(slave, BOND_LINK_UP,
2768                                                           BOND_SLAVE_NOTIFY_NOW);
2769                                 if (current_arp_slave) {
2770                                         bond_set_slave_inactive_flags(
2771                                                 current_arp_slave,
2772                                                 BOND_SLAVE_NOTIFY_NOW);
2773                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2774                                 }
2775
2776                                 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2777                                             slave->dev->name);
2778
2779                                 if (!rtnl_dereference(bond->curr_active_slave) ||
2780                                     slave == rtnl_dereference(bond->primary_slave))
2781                                         goto do_failover;
2782
2783                         }
2784
2785                         continue;
2786
2787                 case BOND_LINK_DOWN:
2788                         if (slave->link_failure_count < UINT_MAX)
2789                                 slave->link_failure_count++;
2790
2791                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2792                                                   BOND_SLAVE_NOTIFY_NOW);
2793                         bond_set_slave_inactive_flags(slave,
2794                                                       BOND_SLAVE_NOTIFY_NOW);
2795
2796                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2797                                     slave->dev->name);
2798
2799                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
2800                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2801                                 goto do_failover;
2802                         }
2803
2804                         continue;
2805
2806                 default:
2807                         netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2808                                    slave->new_link, slave->dev->name);
2809                         continue;
2810                 }
2811
2812 do_failover:
2813                 block_netpoll_tx();
2814                 bond_select_active_slave(bond);
2815                 unblock_netpoll_tx();
2816         }
2817
2818         bond_set_carrier(bond);
2819 }
2820
2821 /* Send ARP probes for active-backup mode ARP monitor.
2822  *
2823  * Called with rcu_read_lock held.
2824  */
2825 static bool bond_ab_arp_probe(struct bonding *bond)
2826 {
2827         struct slave *slave, *before = NULL, *new_slave = NULL,
2828                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2829                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2830         struct list_head *iter;
2831         bool found = false;
2832         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2833
2834         if (curr_arp_slave && curr_active_slave)
2835                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2836                             curr_arp_slave->dev->name,
2837                             curr_active_slave->dev->name);
2838
2839         if (curr_active_slave) {
2840                 bond_arp_send_all(bond, curr_active_slave);
2841                 return should_notify_rtnl;
2842         }
2843
2844         /* if we don't have a curr_active_slave, search for the next available
2845          * backup slave from the current_arp_slave and make it the candidate
2846          * for becoming the curr_active_slave
2847          */
2848
2849         if (!curr_arp_slave) {
2850                 curr_arp_slave = bond_first_slave_rcu(bond);
2851                 if (!curr_arp_slave)
2852                         return should_notify_rtnl;
2853         }
2854
2855         bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2856
2857         bond_for_each_slave_rcu(bond, slave, iter) {
2858                 if (!found && !before && bond_slave_is_up(slave))
2859                         before = slave;
2860
2861                 if (found && !new_slave && bond_slave_is_up(slave))
2862                         new_slave = slave;
2863                 /* if the link state is up at this point, we
2864                  * mark it down - this can happen if we have
2865                  * simultaneous link failures and
2866                  * reselect_active_interface doesn't make this
2867                  * one the current slave so it is still marked
2868                  * up when it is actually down
2869                  */
2870                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2871                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2872                                                   BOND_SLAVE_NOTIFY_LATER);
2873                         if (slave->link_failure_count < UINT_MAX)
2874                                 slave->link_failure_count++;
2875
2876                         bond_set_slave_inactive_flags(slave,
2877                                                       BOND_SLAVE_NOTIFY_LATER);
2878
2879                         netdev_info(bond->dev, "backup interface %s is now down\n",
2880                                     slave->dev->name);
2881                 }
2882                 if (slave == curr_arp_slave)
2883                         found = true;
2884         }
2885
2886         if (!new_slave && before)
2887                 new_slave = before;
2888
2889         if (!new_slave)
2890                 goto check_state;
2891
2892         bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2893                                   BOND_SLAVE_NOTIFY_LATER);
2894         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2895         bond_arp_send_all(bond, new_slave);
2896         new_slave->last_link_up = jiffies;
2897         rcu_assign_pointer(bond->current_arp_slave, new_slave);
2898
2899 check_state:
2900         bond_for_each_slave_rcu(bond, slave, iter) {
2901                 if (slave->should_notify || slave->should_notify_link) {
2902                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2903                         break;
2904                 }
2905         }
2906         return should_notify_rtnl;
2907 }
2908
2909 static void bond_activebackup_arp_mon(struct work_struct *work)
2910 {
2911         struct bonding *bond = container_of(work, struct bonding,
2912                                             arp_work.work);
2913         bool should_notify_peers = false;
2914         bool should_notify_rtnl = false;
2915         int delta_in_ticks;
2916
2917         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2918
2919         if (!bond_has_slaves(bond))
2920                 goto re_arm;
2921
2922         rcu_read_lock();
2923
2924         should_notify_peers = bond_should_notify_peers(bond);
2925
2926         if (bond_ab_arp_inspect(bond)) {
2927                 rcu_read_unlock();
2928
2929                 /* Race avoidance with bond_close flush of workqueue */
2930                 if (!rtnl_trylock()) {
2931                         delta_in_ticks = 1;
2932                         should_notify_peers = false;
2933                         goto re_arm;
2934                 }
2935
2936                 bond_ab_arp_commit(bond);
2937
2938                 rtnl_unlock();
2939                 rcu_read_lock();
2940         }
2941
2942         should_notify_rtnl = bond_ab_arp_probe(bond);
2943         rcu_read_unlock();
2944
2945 re_arm:
2946         if (bond->params.arp_interval)
2947                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2948
2949         if (should_notify_peers || should_notify_rtnl) {
2950                 if (!rtnl_trylock())
2951                         return;
2952
2953                 if (should_notify_peers)
2954                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2955                                                  bond->dev);
2956                 if (should_notify_rtnl) {
2957                         bond_slave_state_notify(bond);
2958                         bond_slave_link_notify(bond);
2959                 }
2960
2961                 rtnl_unlock();
2962         }
2963 }
2964
2965 /*-------------------------- netdev event handling --------------------------*/
2966
2967 /* Change device name */
2968 static int bond_event_changename(struct bonding *bond)
2969 {
2970         bond_remove_proc_entry(bond);
2971         bond_create_proc_entry(bond);
2972
2973         bond_debug_reregister(bond);
2974
2975         return NOTIFY_DONE;
2976 }
2977
2978 static int bond_master_netdev_event(unsigned long event,
2979                                     struct net_device *bond_dev)
2980 {
2981         struct bonding *event_bond = netdev_priv(bond_dev);
2982
2983         switch (event) {
2984         case NETDEV_CHANGENAME:
2985                 return bond_event_changename(event_bond);
2986         case NETDEV_UNREGISTER:
2987                 bond_remove_proc_entry(event_bond);
2988                 break;
2989         case NETDEV_REGISTER:
2990                 bond_create_proc_entry(event_bond);
2991                 break;
2992         case NETDEV_NOTIFY_PEERS:
2993                 if (event_bond->send_peer_notif)
2994                         event_bond->send_peer_notif--;
2995                 break;
2996         default:
2997                 break;
2998         }
2999
3000         return NOTIFY_DONE;
3001 }
3002
3003 static int bond_slave_netdev_event(unsigned long event,
3004                                    struct net_device *slave_dev)
3005 {
3006         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3007         struct bonding *bond;
3008         struct net_device *bond_dev;
3009
3010         /* A netdev event can be generated while enslaving a device
3011          * before netdev_rx_handler_register is called in which case
3012          * slave will be NULL
3013          */
3014         if (!slave)
3015                 return NOTIFY_DONE;
3016         bond_dev = slave->bond->dev;
3017         bond = slave->bond;
3018         primary = rtnl_dereference(bond->primary_slave);
3019
3020         switch (event) {
3021         case NETDEV_UNREGISTER:
3022                 if (bond_dev->type != ARPHRD_ETHER)
3023                         bond_release_and_destroy(bond_dev, slave_dev);
3024                 else
3025                         bond_release(bond_dev, slave_dev);
3026                 break;
3027         case NETDEV_UP:
3028         case NETDEV_CHANGE:
3029                 bond_update_speed_duplex(slave);
3030                 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3031                         bond_3ad_adapter_speed_duplex_changed(slave);
3032                 /* Fallthrough */
3033         case NETDEV_DOWN:
3034                 /* Refresh slave-array if applicable!
3035                  * If the setup does not use miimon or arpmon (mode-specific!),
3036                  * then these events will not cause the slave-array to be
3037                  * refreshed. This will cause xmit to use a slave that is not
3038                  * usable. Avoid such situation by refeshing the array at these
3039                  * events. If these (miimon/arpmon) parameters are configured
3040                  * then array gets refreshed twice and that should be fine!
3041                  */
3042                 if (bond_mode_uses_xmit_hash(bond))
3043                         bond_update_slave_arr(bond, NULL);
3044                 break;
3045         case NETDEV_CHANGEMTU:
3046                 /* TODO: Should slaves be allowed to
3047                  * independently alter their MTU?  For
3048                  * an active-backup bond, slaves need
3049                  * not be the same type of device, so
3050                  * MTUs may vary.  For other modes,
3051                  * slaves arguably should have the
3052                  * same MTUs. To do this, we'd need to
3053                  * take over the slave's change_mtu
3054                  * function for the duration of their
3055                  * servitude.
3056                  */
3057                 break;
3058         case NETDEV_CHANGENAME:
3059                 /* we don't care if we don't have primary set */
3060                 if (!bond_uses_primary(bond) ||
3061                     !bond->params.primary[0])
3062                         break;
3063
3064                 if (slave == primary) {
3065                         /* slave's name changed - he's no longer primary */
3066                         RCU_INIT_POINTER(bond->primary_slave, NULL);
3067                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3068                         /* we have a new primary slave */
3069                         rcu_assign_pointer(bond->primary_slave, slave);
3070                 } else { /* we didn't change primary - exit */
3071                         break;
3072                 }
3073
3074                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3075                             primary ? slave_dev->name : "none");
3076
3077                 block_netpoll_tx();
3078                 bond_select_active_slave(bond);
3079                 unblock_netpoll_tx();
3080                 break;
3081         case NETDEV_FEAT_CHANGE:
3082                 bond_compute_features(bond);
3083                 break;
3084         case NETDEV_RESEND_IGMP:
3085                 /* Propagate to master device */
3086                 call_netdevice_notifiers(event, slave->bond->dev);
3087                 break;
3088         default:
3089                 break;
3090         }
3091
3092         return NOTIFY_DONE;
3093 }
3094
3095 /* bond_netdev_event: handle netdev notifier chain events.
3096  *
3097  * This function receives events for the netdev chain.  The caller (an
3098  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3099  * locks for us to safely manipulate the slave devices (RTNL lock,
3100  * dev_probe_lock).
3101  */
3102 static int bond_netdev_event(struct notifier_block *this,
3103                              unsigned long event, void *ptr)
3104 {
3105         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3106
3107         netdev_dbg(event_dev, "event: %lx\n", event);
3108
3109         if (!(event_dev->priv_flags & IFF_BONDING))
3110                 return NOTIFY_DONE;
3111
3112         if (event_dev->flags & IFF_MASTER) {
3113                 netdev_dbg(event_dev, "IFF_MASTER\n");
3114                 return bond_master_netdev_event(event, event_dev);
3115         }
3116
3117         if (event_dev->flags & IFF_SLAVE) {
3118                 netdev_dbg(event_dev, "IFF_SLAVE\n");
3119                 return bond_slave_netdev_event(event, event_dev);
3120         }
3121
3122         return NOTIFY_DONE;
3123 }
3124
3125 static struct notifier_block bond_netdev_notifier = {
3126         .notifier_call = bond_netdev_event,
3127 };
3128
3129 /*---------------------------- Hashing Policies -----------------------------*/
3130
3131 /* L2 hash helper */
3132 static inline u32 bond_eth_hash(struct sk_buff *skb)
3133 {
3134         struct ethhdr *ep, hdr_tmp;
3135
3136         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3137         if (ep)
3138                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3139         return 0;
3140 }
3141
3142 /* Extract the appropriate headers based on bond's xmit policy */
3143 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3144                               struct flow_keys *fk)
3145 {
3146         const struct ipv6hdr *iph6;
3147         const struct iphdr *iph;
3148         int noff, proto = -1;
3149
3150         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3151                 return skb_flow_dissect_flow_keys(skb, fk, 0);
3152
3153         fk->ports.ports = 0;
3154         noff = skb_network_offset(skb);
3155         if (skb->protocol == htons(ETH_P_IP)) {
3156                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3157                         return false;
3158                 iph = ip_hdr(skb);
3159                 iph_to_flow_copy_v4addrs(fk, iph);
3160                 noff += iph->ihl << 2;
3161                 if (!ip_is_fragment(iph))
3162                         proto = iph->protocol;
3163         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3164                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3165                         return false;
3166                 iph6 = ipv6_hdr(skb);
3167                 iph_to_flow_copy_v6addrs(fk, iph6);
3168                 noff += sizeof(*iph6);
3169                 proto = iph6->nexthdr;
3170         } else {
3171                 return false;
3172         }
3173         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3174                 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3175
3176         return true;
3177 }
3178
3179 /**
3180  * bond_xmit_hash - generate a hash value based on the xmit policy
3181  * @bond: bonding device
3182  * @skb: buffer to use for headers
3183  *
3184  * This function will extract the necessary headers from the skb buffer and use
3185  * them to generate a hash based on the xmit_policy set in the bonding device
3186  */
3187 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3188 {
3189         struct flow_keys flow;
3190         u32 hash;
3191
3192         if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3193             skb->l4_hash)
3194                 return skb->hash;
3195
3196         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3197             !bond_flow_dissect(bond, skb, &flow))
3198                 return bond_eth_hash(skb);
3199
3200         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3201             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3202                 hash = bond_eth_hash(skb);
3203         else
3204                 hash = (__force u32)flow.ports.ports;
3205         hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3206                 (__force u32)flow_get_u32_src(&flow);
3207         hash ^= (hash >> 16);
3208         hash ^= (hash >> 8);
3209
3210         return hash;
3211 }
3212
3213 /*-------------------------- Device entry points ----------------------------*/
3214
3215 static void bond_work_init_all(struct bonding *bond)
3216 {
3217         INIT_DELAYED_WORK(&bond->mcast_work,
3218                           bond_resend_igmp_join_requests_delayed);
3219         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3220         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3221         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3222                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3223         else
3224                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3225         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3226         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3227 }
3228
3229 static void bond_work_cancel_all(struct bonding *bond)
3230 {
3231         cancel_delayed_work_sync(&bond->mii_work);
3232         cancel_delayed_work_sync(&bond->arp_work);
3233         cancel_delayed_work_sync(&bond->alb_work);
3234         cancel_delayed_work_sync(&bond->ad_work);
3235         cancel_delayed_work_sync(&bond->mcast_work);
3236         cancel_delayed_work_sync(&bond->slave_arr_work);
3237 }
3238
3239 static int bond_open(struct net_device *bond_dev)
3240 {
3241         struct bonding *bond = netdev_priv(bond_dev);
3242         struct list_head *iter;
3243         struct slave *slave;
3244
3245         /* reset slave->backup and slave->inactive */
3246         if (bond_has_slaves(bond)) {
3247                 bond_for_each_slave(bond, slave, iter) {
3248                         if (bond_uses_primary(bond) &&
3249                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3250                                 bond_set_slave_inactive_flags(slave,
3251                                                               BOND_SLAVE_NOTIFY_NOW);
3252                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3253                                 bond_set_slave_active_flags(slave,
3254                                                             BOND_SLAVE_NOTIFY_NOW);
3255                         }
3256                 }
3257         }
3258
3259         bond_work_init_all(bond);
3260
3261         if (bond_is_lb(bond)) {
3262                 /* bond_alb_initialize must be called before the timer
3263                  * is started.
3264                  */
3265                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3266                         return -ENOMEM;
3267                 if (bond->params.tlb_dynamic_lb)
3268                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3269         }
3270
3271         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3272                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3273
3274         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3275                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3276                 bond->recv_probe = bond_arp_rcv;
3277         }
3278
3279         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3280                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3281                 /* register to receive LACPDUs */
3282                 bond->recv_probe = bond_3ad_lacpdu_recv;
3283                 bond_3ad_initiate_agg_selection(bond, 1);
3284         }
3285
3286         if (bond_mode_uses_xmit_hash(bond))
3287                 bond_update_slave_arr(bond, NULL);
3288
3289         return 0;
3290 }
3291
3292 static int bond_close(struct net_device *bond_dev)
3293 {
3294         struct bonding *bond = netdev_priv(bond_dev);
3295
3296         bond_work_cancel_all(bond);
3297         bond->send_peer_notif = 0;
3298         if (bond_is_lb(bond))
3299                 bond_alb_deinitialize(bond);
3300         bond->recv_probe = NULL;
3301
3302         return 0;
3303 }
3304
3305 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3306  * that some drivers can provide 32bit values only.
3307  */
3308 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3309                             const struct rtnl_link_stats64 *_new,
3310                             const struct rtnl_link_stats64 *_old)
3311 {
3312         const u64 *new = (const u64 *)_new;
3313         const u64 *old = (const u64 *)_old;
3314         u64 *res = (u64 *)_res;
3315         int i;
3316
3317         for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3318                 u64 nv = new[i];
3319                 u64 ov = old[i];
3320
3321                 /* detects if this particular field is 32bit only */
3322                 if (((nv | ov) >> 32) == 0)
3323                         res[i] += (u32)nv - (u32)ov;
3324                 else
3325                         res[i] += nv - ov;
3326         }
3327 }
3328
3329 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3330                                                 struct rtnl_link_stats64 *stats)
3331 {
3332         struct bonding *bond = netdev_priv(bond_dev);
3333         struct rtnl_link_stats64 temp;
3334         struct list_head *iter;
3335         struct slave *slave;
3336
3337         spin_lock(&bond->stats_lock);
3338         memcpy(stats, &bond->bond_stats, sizeof(*stats));
3339
3340         rcu_read_lock();
3341         bond_for_each_slave_rcu(bond, slave, iter) {
3342                 const struct rtnl_link_stats64 *new =
3343                         dev_get_stats(slave->dev, &temp);
3344
3345                 bond_fold_stats(stats, new, &slave->slave_stats);
3346
3347                 /* save off the slave stats for the next run */
3348                 memcpy(&slave->slave_stats, new, sizeof(*new));
3349         }
3350         rcu_read_unlock();
3351
3352         memcpy(&bond->bond_stats, stats, sizeof(*stats));
3353         spin_unlock(&bond->stats_lock);
3354
3355         return stats;
3356 }
3357
3358 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3359 {
3360         struct bonding *bond = netdev_priv(bond_dev);
3361         struct net_device *slave_dev = NULL;
3362         struct ifbond k_binfo;
3363         struct ifbond __user *u_binfo = NULL;
3364         struct ifslave k_sinfo;
3365         struct ifslave __user *u_sinfo = NULL;
3366         struct mii_ioctl_data *mii = NULL;
3367         struct bond_opt_value newval;
3368         struct net *net;
3369         int res = 0;
3370
3371         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3372
3373         switch (cmd) {
3374         case SIOCGMIIPHY:
3375                 mii = if_mii(ifr);
3376                 if (!mii)
3377                         return -EINVAL;
3378
3379                 mii->phy_id = 0;
3380                 /* Fall Through */
3381         case SIOCGMIIREG:
3382                 /* We do this again just in case we were called by SIOCGMIIREG
3383                  * instead of SIOCGMIIPHY.
3384                  */
3385                 mii = if_mii(ifr);
3386                 if (!mii)
3387                         return -EINVAL;
3388
3389                 if (mii->reg_num == 1) {
3390                         mii->val_out = 0;
3391                         if (netif_carrier_ok(bond->dev))
3392                                 mii->val_out = BMSR_LSTATUS;
3393                 }
3394
3395                 return 0;
3396         case BOND_INFO_QUERY_OLD:
3397         case SIOCBONDINFOQUERY:
3398                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3399
3400                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3401                         return -EFAULT;
3402
3403                 res = bond_info_query(bond_dev, &k_binfo);
3404                 if (res == 0 &&
3405                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3406                         return -EFAULT;
3407
3408                 return res;
3409         case BOND_SLAVE_INFO_QUERY_OLD:
3410         case SIOCBONDSLAVEINFOQUERY:
3411                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3412
3413                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3414                         return -EFAULT;
3415
3416                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3417                 if (res == 0 &&
3418                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3419                         return -EFAULT;
3420
3421                 return res;
3422         default:
3423                 break;
3424         }
3425
3426         net = dev_net(bond_dev);
3427
3428         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3429                 return -EPERM;
3430
3431         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3432
3433         netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3434
3435         if (!slave_dev)
3436                 return -ENODEV;
3437
3438         netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3439         switch (cmd) {
3440         case BOND_ENSLAVE_OLD:
3441         case SIOCBONDENSLAVE:
3442                 res = bond_enslave(bond_dev, slave_dev);
3443                 break;
3444         case BOND_RELEASE_OLD:
3445         case SIOCBONDRELEASE:
3446                 res = bond_release(bond_dev, slave_dev);
3447                 break;
3448         case BOND_SETHWADDR_OLD:
3449         case SIOCBONDSETHWADDR:
3450                 bond_set_dev_addr(bond_dev, slave_dev);
3451                 res = 0;
3452                 break;
3453         case BOND_CHANGE_ACTIVE_OLD:
3454         case SIOCBONDCHANGEACTIVE:
3455                 bond_opt_initstr(&newval, slave_dev->name);
3456                 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3457                 break;
3458         default:
3459                 res = -EOPNOTSUPP;
3460         }
3461
3462         return res;
3463 }
3464
3465 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3466 {
3467         struct bonding *bond = netdev_priv(bond_dev);
3468
3469         if (change & IFF_PROMISC)
3470                 bond_set_promiscuity(bond,
3471                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3472
3473         if (change & IFF_ALLMULTI)
3474                 bond_set_allmulti(bond,
3475                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3476 }
3477
3478 static void bond_set_rx_mode(struct net_device *bond_dev)
3479 {
3480         struct bonding *bond = netdev_priv(bond_dev);
3481         struct list_head *iter;
3482         struct slave *slave;
3483
3484         rcu_read_lock();
3485         if (bond_uses_primary(bond)) {
3486                 slave = rcu_dereference(bond->curr_active_slave);
3487                 if (slave) {
3488                         dev_uc_sync(slave->dev, bond_dev);
3489                         dev_mc_sync(slave->dev, bond_dev);
3490                 }
3491         } else {
3492                 bond_for_each_slave_rcu(bond, slave, iter) {
3493                         dev_uc_sync_multiple(slave->dev, bond_dev);
3494                         dev_mc_sync_multiple(slave->dev, bond_dev);
3495                 }
3496         }
3497         rcu_read_unlock();
3498 }
3499
3500 static int bond_neigh_init(struct neighbour *n)
3501 {
3502         struct bonding *bond = netdev_priv(n->dev);
3503         const struct net_device_ops *slave_ops;
3504         struct neigh_parms parms;
3505         struct slave *slave;
3506         int ret;
3507
3508         slave = bond_first_slave(bond);
3509         if (!slave)
3510                 return 0;
3511         slave_ops = slave->dev->netdev_ops;
3512         if (!slave_ops->ndo_neigh_setup)
3513                 return 0;
3514
3515         parms.neigh_setup = NULL;
3516         parms.neigh_cleanup = NULL;
3517         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3518         if (ret)
3519                 return ret;
3520
3521         /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3522          * after the last slave has been detached.  Assumes that all slaves
3523          * utilize the same neigh_cleanup (true at this writing as only user
3524          * is ipoib).
3525          */
3526         n->parms->neigh_cleanup = parms.neigh_cleanup;
3527
3528         if (!parms.neigh_setup)
3529                 return 0;
3530
3531         return parms.neigh_setup(n);
3532 }
3533
3534 /* The bonding ndo_neigh_setup is called at init time beofre any
3535  * slave exists. So we must declare proxy setup function which will
3536  * be used at run time to resolve the actual slave neigh param setup.
3537  *
3538  * It's also called by master devices (such as vlans) to setup their
3539  * underlying devices. In that case - do nothing, we're already set up from
3540  * our init.
3541  */
3542 static int bond_neigh_setup(struct net_device *dev,
3543                             struct neigh_parms *parms)
3544 {
3545         /* modify only our neigh_parms */
3546         if (parms->dev == dev)
3547                 parms->neigh_setup = bond_neigh_init;
3548
3549         return 0;
3550 }
3551
3552 /* Change the MTU of all of a master's slaves to match the master */
3553 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3554 {
3555         struct bonding *bond = netdev_priv(bond_dev);
3556         struct slave *slave, *rollback_slave;
3557         struct list_head *iter;
3558         int res = 0;
3559
3560         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3561
3562         bond_for_each_slave(bond, slave, iter) {
3563                 netdev_dbg(bond_dev, "s %p c_m %p\n",
3564                            slave, slave->dev->netdev_ops->ndo_change_mtu);
3565
3566                 res = dev_set_mtu(slave->dev, new_mtu);
3567
3568                 if (res) {
3569                         /* If we failed to set the slave's mtu to the new value
3570                          * we must abort the operation even in ACTIVE_BACKUP
3571                          * mode, because if we allow the backup slaves to have
3572                          * different mtu values than the active slave we'll
3573                          * need to change their mtu when doing a failover. That
3574                          * means changing their mtu from timer context, which
3575                          * is probably not a good idea.
3576                          */
3577                         netdev_dbg(bond_dev, "err %d %s\n", res,
3578                                    slave->dev->name);
3579                         goto unwind;
3580                 }
3581         }
3582
3583         bond_dev->mtu = new_mtu;
3584
3585         return 0;
3586
3587 unwind:
3588         /* unwind from head to the slave that failed */
3589         bond_for_each_slave(bond, rollback_slave, iter) {
3590                 int tmp_res;
3591
3592                 if (rollback_slave == slave)
3593                         break;
3594
3595                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3596                 if (tmp_res) {
3597                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3598                                    tmp_res, rollback_slave->dev->name);
3599                 }
3600         }
3601
3602         return res;
3603 }
3604
3605 /* Change HW address
3606  *
3607  * Note that many devices must be down to change the HW address, and
3608  * downing the master releases all slaves.  We can make bonds full of
3609  * bonding devices to test this, however.
3610  */
3611 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3612 {
3613         struct bonding *bond = netdev_priv(bond_dev);
3614         struct slave *slave, *rollback_slave;
3615         struct sockaddr *sa = addr, tmp_sa;
3616         struct list_head *iter;
3617         int res = 0;
3618
3619         if (BOND_MODE(bond) == BOND_MODE_ALB)
3620                 return bond_alb_set_mac_address(bond_dev, addr);
3621
3622
3623         netdev_dbg(bond_dev, "bond=%p\n", bond);
3624
3625         /* If fail_over_mac is enabled, do nothing and return success.
3626          * Returning an error causes ifenslave to fail.
3627          */
3628         if (bond->params.fail_over_mac &&
3629             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3630                 return 0;
3631
3632         if (!is_valid_ether_addr(sa->sa_data))
3633                 return -EADDRNOTAVAIL;
3634
3635         bond_for_each_slave(bond, slave, iter) {
3636                 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3637                 res = dev_set_mac_address(slave->dev, addr);
3638                 if (res) {
3639                         /* TODO: consider downing the slave
3640                          * and retry ?
3641                          * User should expect communications
3642                          * breakage anyway until ARP finish
3643                          * updating, so...
3644                          */
3645                         netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3646                         goto unwind;
3647                 }
3648         }
3649
3650         /* success */
3651         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3652         return 0;
3653
3654 unwind:
3655         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3656         tmp_sa.sa_family = bond_dev->type;
3657
3658         /* unwind from head to the slave that failed */
3659         bond_for_each_slave(bond, rollback_slave, iter) {
3660                 int tmp_res;
3661
3662                 if (rollback_slave == slave)
3663                         break;
3664
3665                 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3666                 if (tmp_res) {
3667                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3668                                    tmp_res, rollback_slave->dev->name);
3669                 }
3670         }
3671
3672         return res;
3673 }
3674
3675 /**
3676  * bond_xmit_slave_id - transmit skb through slave with slave_id
3677  * @bond: bonding device that is transmitting
3678  * @skb: buffer to transmit
3679  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3680  *
3681  * This function tries to transmit through slave with slave_id but in case
3682  * it fails, it tries to find the first available slave for transmission.
3683  * The skb is consumed in all cases, thus the function is void.
3684  */
3685 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3686 {
3687         struct list_head *iter;
3688         struct slave *slave;
3689         int i = slave_id;
3690
3691         /* Here we start from the slave with slave_id */
3692         bond_for_each_slave_rcu(bond, slave, iter) {
3693                 if (--i < 0) {
3694                         if (bond_slave_can_tx(slave)) {
3695                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3696                                 return;
3697                         }
3698                 }
3699         }
3700
3701         /* Here we start from the first slave up to slave_id */
3702         i = slave_id;
3703         bond_for_each_slave_rcu(bond, slave, iter) {
3704                 if (--i < 0)
3705                         break;
3706                 if (bond_slave_can_tx(slave)) {
3707                         bond_dev_queue_xmit(bond, skb, slave->dev);
3708                         return;
3709                 }
3710         }
3711         /* no slave that can tx has been found */
3712         bond_tx_drop(bond->dev, skb);
3713 }
3714
3715 /**
3716  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3717  * @bond: bonding device to use
3718  *
3719  * Based on the value of the bonding device's packets_per_slave parameter
3720  * this function generates a slave id, which is usually used as the next
3721  * slave to transmit through.
3722  */
3723 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3724 {
3725         u32 slave_id;
3726         struct reciprocal_value reciprocal_packets_per_slave;
3727         int packets_per_slave = bond->params.packets_per_slave;
3728
3729         switch (packets_per_slave) {
3730         case 0:
3731                 slave_id = prandom_u32();
3732                 break;
3733         case 1:
3734                 slave_id = bond->rr_tx_counter;
3735                 break;
3736         default:
3737                 reciprocal_packets_per_slave =
3738                         bond->params.reciprocal_packets_per_slave;
3739                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3740                                              reciprocal_packets_per_slave);
3741                 break;
3742         }
3743         bond->rr_tx_counter++;
3744
3745         return slave_id;
3746 }
3747
3748 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3749 {
3750         struct bonding *bond = netdev_priv(bond_dev);
3751         struct iphdr *iph = ip_hdr(skb);
3752         struct slave *slave;
3753         u32 slave_id;
3754
3755         /* Start with the curr_active_slave that joined the bond as the
3756          * default for sending IGMP traffic.  For failover purposes one
3757          * needs to maintain some consistency for the interface that will
3758          * send the join/membership reports.  The curr_active_slave found
3759          * will send all of this type of traffic.
3760          */
3761         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3762                 slave = rcu_dereference(bond->curr_active_slave);
3763                 if (slave)
3764                         bond_dev_queue_xmit(bond, skb, slave->dev);
3765                 else
3766                         bond_xmit_slave_id(bond, skb, 0);
3767         } else {
3768                 int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3769
3770                 if (likely(slave_cnt)) {
3771                         slave_id = bond_rr_gen_slave_id(bond);
3772                         bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3773                 } else {
3774                         bond_tx_drop(bond_dev, skb);
3775                 }
3776         }
3777
3778         return NETDEV_TX_OK;
3779 }
3780
3781 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3782  * the bond has a usable interface.
3783  */
3784 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3785 {
3786         struct bonding *bond = netdev_priv(bond_dev);
3787         struct slave *slave;
3788
3789         slave = rcu_dereference(bond->curr_active_slave);
3790         if (slave)
3791                 bond_dev_queue_xmit(bond, skb, slave->dev);
3792         else
3793                 bond_tx_drop(bond_dev, skb);
3794
3795         return NETDEV_TX_OK;
3796 }
3797
3798 /* Use this to update slave_array when (a) it's not appropriate to update
3799  * slave_array right away (note that update_slave_array() may sleep)
3800  * and / or (b) RTNL is not held.
3801  */
3802 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3803 {
3804         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3805 }
3806
3807 /* Slave array work handler. Holds only RTNL */
3808 static void bond_slave_arr_handler(struct work_struct *work)
3809 {
3810         struct bonding *bond = container_of(work, struct bonding,
3811                                             slave_arr_work.work);
3812         int ret;
3813
3814         if (!rtnl_trylock())
3815                 goto err;
3816
3817         ret = bond_update_slave_arr(bond, NULL);
3818         rtnl_unlock();
3819         if (ret) {
3820                 pr_warn_ratelimited("Failed to update slave array from WT\n");
3821                 goto err;
3822         }
3823         return;
3824
3825 err:
3826         bond_slave_arr_work_rearm(bond, 1);
3827 }
3828
3829 /* Build the usable slaves array in control path for modes that use xmit-hash
3830  * to determine the slave interface -
3831  * (a) BOND_MODE_8023AD
3832  * (b) BOND_MODE_XOR
3833  * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3834  *
3835  * The caller is expected to hold RTNL only and NO other lock!
3836  */
3837 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3838 {
3839         struct slave *slave;
3840         struct list_head *iter;
3841         struct bond_up_slave *new_arr, *old_arr;
3842         int agg_id = 0;
3843         int ret = 0;
3844
3845 #ifdef CONFIG_LOCKDEP
3846         WARN_ON(lockdep_is_held(&bond->mode_lock));
3847 #endif
3848
3849         new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3850                           GFP_KERNEL);
3851         if (!new_arr) {
3852                 ret = -ENOMEM;
3853                 pr_err("Failed to build slave-array.\n");
3854                 goto out;
3855         }
3856         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3857                 struct ad_info ad_info;
3858
3859                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3860                         pr_debug("bond_3ad_get_active_agg_info failed\n");
3861                         kfree_rcu(new_arr, rcu);
3862                         /* No active aggragator means it's not safe to use
3863                          * the previous array.
3864                          */
3865                         old_arr = rtnl_dereference(bond->slave_arr);
3866                         if (old_arr) {
3867                                 RCU_INIT_POINTER(bond->slave_arr, NULL);
3868                                 kfree_rcu(old_arr, rcu);
3869                         }
3870                         goto out;
3871                 }
3872                 agg_id = ad_info.aggregator_id;
3873         }
3874         bond_for_each_slave(bond, slave, iter) {
3875                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3876                         struct aggregator *agg;
3877
3878                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
3879                         if (!agg || agg->aggregator_identifier != agg_id)
3880                                 continue;
3881                 }
3882                 if (!bond_slave_can_tx(slave))
3883                         continue;
3884                 if (skipslave == slave)
3885                         continue;
3886                 new_arr->arr[new_arr->count++] = slave;
3887         }
3888
3889         old_arr = rtnl_dereference(bond->slave_arr);
3890         rcu_assign_pointer(bond->slave_arr, new_arr);
3891         if (old_arr)
3892                 kfree_rcu(old_arr, rcu);
3893 out:
3894         if (ret != 0 && skipslave) {
3895                 int idx;
3896
3897                 /* Rare situation where caller has asked to skip a specific
3898                  * slave but allocation failed (most likely!). BTW this is
3899                  * only possible when the call is initiated from
3900                  * __bond_release_one(). In this situation; overwrite the
3901                  * skipslave entry in the array with the last entry from the
3902                  * array to avoid a situation where the xmit path may choose
3903                  * this to-be-skipped slave to send a packet out.
3904                  */
3905                 old_arr = rtnl_dereference(bond->slave_arr);
3906                 for (idx = 0; idx < old_arr->count; idx++) {
3907                         if (skipslave == old_arr->arr[idx]) {
3908                                 old_arr->arr[idx] =
3909                                     old_arr->arr[old_arr->count-1];
3910                                 old_arr->count--;
3911                                 break;
3912                         }
3913                 }
3914         }
3915         return ret;
3916 }
3917
3918 /* Use this Xmit function for 3AD as well as XOR modes. The current
3919  * usable slave array is formed in the control path. The xmit function
3920  * just calculates hash and sends the packet out.
3921  */
3922 static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3923 {
3924         struct bonding *bond = netdev_priv(dev);
3925         struct slave *slave;
3926         struct bond_up_slave *slaves;
3927         unsigned int count;
3928
3929         slaves = rcu_dereference(bond->slave_arr);
3930         count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3931         if (likely(count)) {
3932                 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3933                 bond_dev_queue_xmit(bond, skb, slave->dev);
3934         } else {
3935                 bond_tx_drop(dev, skb);
3936         }
3937
3938         return NETDEV_TX_OK;
3939 }
3940
3941 /* in broadcast mode, we send everything to all usable interfaces. */
3942 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3943 {
3944         struct bonding *bond = netdev_priv(bond_dev);
3945         struct slave *slave = NULL;
3946         struct list_head *iter;
3947
3948         bond_for_each_slave_rcu(bond, slave, iter) {
3949                 if (bond_is_last_slave(bond, slave))
3950                         break;
3951                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3952                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3953
3954                         if (!skb2) {
3955                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3956                                                     bond_dev->name, __func__);
3957                                 continue;
3958                         }
3959                         bond_dev_queue_xmit(bond, skb2, slave->dev);
3960                 }
3961         }
3962         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3963                 bond_dev_queue_xmit(bond, skb, slave->dev);
3964         else
3965                 bond_tx_drop(bond_dev, skb);
3966
3967         return NETDEV_TX_OK;
3968 }
3969
3970 /*------------------------- Device initialization ---------------------------*/
3971
3972 /* Lookup the slave that corresponds to a qid */
3973 static inline int bond_slave_override(struct bonding *bond,
3974                                       struct sk_buff *skb)
3975 {
3976         struct slave *slave = NULL;
3977         struct list_head *iter;
3978
3979         if (!skb->queue_mapping)
3980                 return 1;
3981
3982         /* Find out if any slaves have the same mapping as this skb. */
3983         bond_for_each_slave_rcu(bond, slave, iter) {
3984                 if (slave->queue_id == skb->queue_mapping) {
3985                         if (bond_slave_is_up(slave) &&
3986                             slave->link == BOND_LINK_UP) {
3987                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3988                                 return 0;
3989                         }
3990                         /* If the slave isn't UP, use default transmit policy. */
3991                         break;
3992                 }
3993         }
3994
3995         return 1;
3996 }
3997
3998
3999 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4000                              void *accel_priv, select_queue_fallback_t fallback)
4001 {
4002         /* This helper function exists to help dev_pick_tx get the correct
4003          * destination queue.  Using a helper function skips a call to
4004          * skb_tx_hash and will put the skbs in the queue we expect on their
4005          * way down to the bonding driver.
4006          */
4007         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4008
4009         /* Save the original txq to restore before passing to the driver */
4010         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4011
4012         if (unlikely(txq >= dev->real_num_tx_queues)) {
4013                 do {
4014                         txq -= dev->real_num_tx_queues;
4015                 } while (txq >= dev->real_num_tx_queues);
4016         }
4017         return txq;
4018 }
4019
4020 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4021 {
4022         struct bonding *bond = netdev_priv(dev);
4023
4024         if (bond_should_override_tx_queue(bond) &&
4025             !bond_slave_override(bond, skb))
4026                 return NETDEV_TX_OK;
4027
4028         switch (BOND_MODE(bond)) {
4029         case BOND_MODE_ROUNDROBIN:
4030                 return bond_xmit_roundrobin(skb, dev);
4031         case BOND_MODE_ACTIVEBACKUP:
4032                 return bond_xmit_activebackup(skb, dev);
4033         case BOND_MODE_8023AD:
4034         case BOND_MODE_XOR:
4035                 return bond_3ad_xor_xmit(skb, dev);
4036         case BOND_MODE_BROADCAST:
4037                 return bond_xmit_broadcast(skb, dev);
4038         case BOND_MODE_ALB:
4039                 return bond_alb_xmit(skb, dev);
4040         case BOND_MODE_TLB:
4041                 return bond_tlb_xmit(skb, dev);
4042         default:
4043                 /* Should never happen, mode already checked */
4044                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4045                 WARN_ON_ONCE(1);
4046                 bond_tx_drop(dev, skb);
4047                 return NETDEV_TX_OK;
4048         }
4049 }
4050
4051 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4052 {
4053         struct bonding *bond = netdev_priv(dev);
4054         netdev_tx_t ret = NETDEV_TX_OK;
4055
4056         /* If we risk deadlock from transmitting this in the
4057          * netpoll path, tell netpoll to queue the frame for later tx
4058          */
4059         if (unlikely(is_netpoll_tx_blocked(dev)))
4060                 return NETDEV_TX_BUSY;
4061
4062         rcu_read_lock();
4063         if (bond_has_slaves(bond))
4064                 ret = __bond_start_xmit(skb, dev);
4065         else
4066                 bond_tx_drop(dev, skb);
4067         rcu_read_unlock();
4068
4069         return ret;
4070 }
4071
4072 static int bond_ethtool_get_settings(struct net_device *bond_dev,
4073                                      struct ethtool_cmd *ecmd)
4074 {
4075         struct bonding *bond = netdev_priv(bond_dev);
4076         unsigned long speed = 0;
4077         struct list_head *iter;
4078         struct slave *slave;
4079
4080         ecmd->duplex = DUPLEX_UNKNOWN;
4081         ecmd->port = PORT_OTHER;
4082
4083         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4084          * do not need to check mode.  Though link speed might not represent
4085          * the true receive or transmit bandwidth (not all modes are symmetric)
4086          * this is an accurate maximum.
4087          */
4088         bond_for_each_slave(bond, slave, iter) {
4089                 if (bond_slave_can_tx(slave)) {
4090                         if (slave->speed != SPEED_UNKNOWN)
4091                                 speed += slave->speed;
4092                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
4093                             slave->duplex != DUPLEX_UNKNOWN)
4094                                 ecmd->duplex = slave->duplex;
4095                 }
4096         }
4097         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
4098
4099         return 0;
4100 }
4101
4102 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4103                                      struct ethtool_drvinfo *drvinfo)
4104 {
4105         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4106         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4107         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4108                  BOND_ABI_VERSION);
4109 }
4110
4111 static const struct ethtool_ops bond_ethtool_ops = {
4112         .get_drvinfo            = bond_ethtool_get_drvinfo,
4113         .get_settings           = bond_ethtool_get_settings,
4114         .get_link               = ethtool_op_get_link,
4115 };
4116
4117 static const struct net_device_ops bond_netdev_ops = {
4118         .ndo_init               = bond_init,
4119         .ndo_uninit             = bond_uninit,
4120         .ndo_open               = bond_open,
4121         .ndo_stop               = bond_close,
4122         .ndo_start_xmit         = bond_start_xmit,
4123         .ndo_select_queue       = bond_select_queue,
4124         .ndo_get_stats64        = bond_get_stats,
4125         .ndo_do_ioctl           = bond_do_ioctl,
4126         .ndo_change_rx_flags    = bond_change_rx_flags,
4127         .ndo_set_rx_mode        = bond_set_rx_mode,
4128         .ndo_change_mtu         = bond_change_mtu,
4129         .ndo_set_mac_address    = bond_set_mac_address,
4130         .ndo_neigh_setup        = bond_neigh_setup,
4131         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4132         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4133 #ifdef CONFIG_NET_POLL_CONTROLLER
4134         .ndo_netpoll_setup      = bond_netpoll_setup,
4135         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4136         .ndo_poll_controller    = bond_poll_controller,
4137 #endif
4138         .ndo_add_slave          = bond_enslave,
4139         .ndo_del_slave          = bond_release,
4140         .ndo_fix_features       = bond_fix_features,
4141         .ndo_bridge_setlink     = switchdev_port_bridge_setlink,
4142         .ndo_bridge_getlink     = switchdev_port_bridge_getlink,
4143         .ndo_bridge_dellink     = switchdev_port_bridge_dellink,
4144         .ndo_fdb_add            = switchdev_port_fdb_add,
4145         .ndo_fdb_del            = switchdev_port_fdb_del,
4146         .ndo_fdb_dump           = switchdev_port_fdb_dump,
4147         .ndo_features_check     = passthru_features_check,
4148 };
4149
4150 static const struct device_type bond_type = {
4151         .name = "bond",
4152 };
4153
4154 static void bond_destructor(struct net_device *bond_dev)
4155 {
4156         struct bonding *bond = netdev_priv(bond_dev);
4157         if (bond->wq)
4158                 destroy_workqueue(bond->wq);
4159         free_netdev(bond_dev);
4160 }
4161
4162 void bond_setup(struct net_device *bond_dev)
4163 {
4164         struct bonding *bond = netdev_priv(bond_dev);
4165
4166         spin_lock_init(&bond->mode_lock);
4167         spin_lock_init(&bond->stats_lock);
4168         bond->params = bonding_defaults;
4169
4170         /* Initialize pointers */
4171         bond->dev = bond_dev;
4172
4173         /* Initialize the device entry points */
4174         ether_setup(bond_dev);
4175         bond_dev->netdev_ops = &bond_netdev_ops;
4176         bond_dev->ethtool_ops = &bond_ethtool_ops;
4177
4178         bond_dev->destructor = bond_destructor;
4179
4180         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4181
4182         /* Initialize the device options */
4183         bond_dev->flags |= IFF_MASTER;
4184         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4185         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4186
4187         /* don't acquire bond device's netif_tx_lock when transmitting */
4188         bond_dev->features |= NETIF_F_LLTX;
4189
4190         /* By default, we declare the bond to be fully
4191          * VLAN hardware accelerated capable. Special
4192          * care is taken in the various xmit functions
4193          * when there are slaves that are not hw accel
4194          * capable
4195          */
4196
4197         /* Don't allow bond devices to change network namespaces. */
4198         bond_dev->features |= NETIF_F_NETNS_LOCAL;
4199
4200         bond_dev->hw_features = BOND_VLAN_FEATURES |
4201                                 NETIF_F_HW_VLAN_CTAG_TX |
4202                                 NETIF_F_HW_VLAN_CTAG_RX |
4203                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4204
4205         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4206         bond_dev->features |= bond_dev->hw_features;
4207 }
4208
4209 /* Destroy a bonding device.
4210  * Must be under rtnl_lock when this function is called.
4211  */
4212 static void bond_uninit(struct net_device *bond_dev)
4213 {
4214         struct bonding *bond = netdev_priv(bond_dev);
4215         struct list_head *iter;
4216         struct slave *slave;
4217         struct bond_up_slave *arr;
4218
4219         bond_netpoll_cleanup(bond_dev);
4220
4221         /* Release the bonded slaves */
4222         bond_for_each_slave(bond, slave, iter)
4223                 __bond_release_one(bond_dev, slave->dev, true);
4224         netdev_info(bond_dev, "Released all slaves\n");
4225
4226         arr = rtnl_dereference(bond->slave_arr);
4227         if (arr) {
4228                 RCU_INIT_POINTER(bond->slave_arr, NULL);
4229                 kfree_rcu(arr, rcu);
4230         }
4231
4232         list_del(&bond->bond_list);
4233
4234         bond_debug_unregister(bond);
4235 }
4236
4237 /*------------------------- Module initialization ---------------------------*/
4238
4239 static int bond_check_params(struct bond_params *params)
4240 {
4241         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4242         struct bond_opt_value newval;
4243         const struct bond_opt_value *valptr;
4244         int arp_all_targets_value;
4245         u16 ad_actor_sys_prio = 0;
4246         u16 ad_user_port_key = 0;
4247
4248         /* Convert string parameters. */
4249         if (mode) {
4250                 bond_opt_initstr(&newval, mode);
4251                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4252                 if (!valptr) {
4253                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4254                         return -EINVAL;
4255                 }
4256                 bond_mode = valptr->value;
4257         }
4258
4259         if (xmit_hash_policy) {
4260                 if ((bond_mode != BOND_MODE_XOR) &&
4261                     (bond_mode != BOND_MODE_8023AD) &&
4262                     (bond_mode != BOND_MODE_TLB)) {
4263                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4264                                 bond_mode_name(bond_mode));
4265                 } else {
4266                         bond_opt_initstr(&newval, xmit_hash_policy);
4267                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4268                                                 &newval);
4269                         if (!valptr) {
4270                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4271                                        xmit_hash_policy);
4272                                 return -EINVAL;
4273                         }
4274                         xmit_hashtype = valptr->value;
4275                 }
4276         }
4277
4278         if (lacp_rate) {
4279                 if (bond_mode != BOND_MODE_8023AD) {
4280                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4281                                 bond_mode_name(bond_mode));
4282                 } else {
4283                         bond_opt_initstr(&newval, lacp_rate);
4284                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4285                                                 &newval);
4286                         if (!valptr) {
4287                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4288                                        lacp_rate);
4289                                 return -EINVAL;
4290                         }
4291                         lacp_fast = valptr->value;
4292                 }
4293         }
4294
4295         if (ad_select) {
4296                 bond_opt_initstr(&newval, ad_select);
4297                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4298                                         &newval);
4299                 if (!valptr) {
4300                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4301                         return -EINVAL;
4302                 }
4303                 params->ad_select = valptr->value;
4304                 if (bond_mode != BOND_MODE_8023AD)
4305                         pr_warn("ad_select param only affects 802.3ad mode\n");
4306         } else {
4307                 params->ad_select = BOND_AD_STABLE;
4308         }
4309
4310         if (max_bonds < 0) {
4311                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4312                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4313                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4314         }
4315
4316         if (miimon < 0) {
4317                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4318                         miimon, INT_MAX);
4319                 miimon = 0;
4320         }
4321
4322         if (updelay < 0) {
4323                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4324                         updelay, INT_MAX);
4325                 updelay = 0;
4326         }
4327
4328         if (downdelay < 0) {
4329                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4330                         downdelay, INT_MAX);
4331                 downdelay = 0;
4332         }
4333
4334         if ((use_carrier != 0) && (use_carrier != 1)) {
4335                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4336                         use_carrier);
4337                 use_carrier = 1;
4338         }
4339
4340         if (num_peer_notif < 0 || num_peer_notif > 255) {
4341                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4342                         num_peer_notif);
4343                 num_peer_notif = 1;
4344         }
4345
4346         /* reset values for 802.3ad/TLB/ALB */
4347         if (!bond_mode_uses_arp(bond_mode)) {
4348                 if (!miimon) {
4349                         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");
4350                         pr_warn("Forcing miimon to 100msec\n");
4351                         miimon = BOND_DEFAULT_MIIMON;
4352                 }
4353         }
4354
4355         if (tx_queues < 1 || tx_queues > 255) {
4356                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4357                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4358                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4359         }
4360
4361         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4362                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4363                         all_slaves_active);
4364                 all_slaves_active = 0;
4365         }
4366
4367         if (resend_igmp < 0 || resend_igmp > 255) {
4368                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4369                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4370                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4371         }
4372
4373         bond_opt_initval(&newval, packets_per_slave);
4374         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4375                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4376                         packets_per_slave, USHRT_MAX);
4377                 packets_per_slave = 1;
4378         }
4379
4380         if (bond_mode == BOND_MODE_ALB) {
4381                 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",
4382                           updelay);
4383         }
4384
4385         if (!miimon) {
4386                 if (updelay || downdelay) {
4387                         /* just warn the user the up/down delay will have
4388                          * no effect since miimon is zero...
4389                          */
4390                         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",
4391                                 updelay, downdelay);
4392                 }
4393         } else {
4394                 /* don't allow arp monitoring */
4395                 if (arp_interval) {
4396                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4397                                 miimon, arp_interval);
4398                         arp_interval = 0;
4399                 }
4400
4401                 if ((updelay % miimon) != 0) {
4402                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4403                                 updelay, miimon, (updelay / miimon) * miimon);
4404                 }
4405
4406                 updelay /= miimon;
4407
4408                 if ((downdelay % miimon) != 0) {
4409                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4410                                 downdelay, miimon,
4411                                 (downdelay / miimon) * miimon);
4412                 }
4413
4414                 downdelay /= miimon;
4415         }
4416
4417         if (arp_interval < 0) {
4418                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4419                         arp_interval, INT_MAX);
4420                 arp_interval = 0;
4421         }
4422
4423         for (arp_ip_count = 0, i = 0;
4424              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4425                 __be32 ip;
4426
4427                 /* not a complete check, but good enough to catch mistakes */
4428                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4429                     !bond_is_ip_target_ok(ip)) {
4430                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4431                                 arp_ip_target[i]);
4432                         arp_interval = 0;
4433                 } else {
4434                         if (bond_get_targets_ip(arp_target, ip) == -1)
4435                                 arp_target[arp_ip_count++] = ip;
4436                         else
4437                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4438                                         &ip);
4439                 }
4440         }
4441
4442         if (arp_interval && !arp_ip_count) {
4443                 /* don't allow arping if no arp_ip_target given... */
4444                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4445                         arp_interval);
4446                 arp_interval = 0;
4447         }
4448
4449         if (arp_validate) {
4450                 if (!arp_interval) {
4451                         pr_err("arp_validate requires arp_interval\n");
4452                         return -EINVAL;
4453                 }
4454
4455                 bond_opt_initstr(&newval, arp_validate);
4456                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4457                                         &newval);
4458                 if (!valptr) {
4459                         pr_err("Error: invalid arp_validate \"%s\"\n",
4460                                arp_validate);
4461                         return -EINVAL;
4462                 }
4463                 arp_validate_value = valptr->value;
4464         } else {
4465                 arp_validate_value = 0;
4466         }
4467
4468         arp_all_targets_value = 0;
4469         if (arp_all_targets) {
4470                 bond_opt_initstr(&newval, arp_all_targets);
4471                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4472                                         &newval);
4473                 if (!valptr) {
4474                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4475                                arp_all_targets);
4476                         arp_all_targets_value = 0;
4477                 } else {
4478                         arp_all_targets_value = valptr->value;
4479                 }
4480         }
4481
4482         if (miimon) {
4483                 pr_info("MII link monitoring set to %d ms\n", miimon);
4484         } else if (arp_interval) {
4485                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4486                                           arp_validate_value);
4487                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4488                         arp_interval, valptr->string, arp_ip_count);
4489
4490                 for (i = 0; i < arp_ip_count; i++)
4491                         pr_cont(" %s", arp_ip_target[i]);
4492
4493                 pr_cont("\n");
4494
4495         } else if (max_bonds) {
4496                 /* miimon and arp_interval not set, we need one so things
4497                  * work as expected, see bonding.txt for details
4498                  */
4499                 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");
4500         }
4501
4502         if (primary && !bond_mode_uses_primary(bond_mode)) {
4503                 /* currently, using a primary only makes sense
4504                  * in active backup, TLB or ALB modes
4505                  */
4506                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4507                         primary, bond_mode_name(bond_mode));
4508                 primary = NULL;
4509         }
4510
4511         if (primary && primary_reselect) {
4512                 bond_opt_initstr(&newval, primary_reselect);
4513                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4514                                         &newval);
4515                 if (!valptr) {
4516                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4517                                primary_reselect);
4518                         return -EINVAL;
4519                 }
4520                 primary_reselect_value = valptr->value;
4521         } else {
4522                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4523         }
4524
4525         if (fail_over_mac) {
4526                 bond_opt_initstr(&newval, fail_over_mac);
4527                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4528                                         &newval);
4529                 if (!valptr) {
4530                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4531                                fail_over_mac);
4532                         return -EINVAL;
4533                 }
4534                 fail_over_mac_value = valptr->value;
4535                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4536                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4537         } else {
4538                 fail_over_mac_value = BOND_FOM_NONE;
4539         }
4540
4541         bond_opt_initstr(&newval, "default");
4542         valptr = bond_opt_parse(
4543                         bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4544                                      &newval);
4545         if (!valptr) {
4546                 pr_err("Error: No ad_actor_sys_prio default value");
4547                 return -EINVAL;
4548         }
4549         ad_actor_sys_prio = valptr->value;
4550
4551         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4552                                 &newval);
4553         if (!valptr) {
4554                 pr_err("Error: No ad_user_port_key default value");
4555                 return -EINVAL;
4556         }
4557         ad_user_port_key = valptr->value;
4558
4559         if (lp_interval == 0) {
4560                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4561                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4562                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4563         }
4564
4565         /* fill params struct with the proper values */
4566         params->mode = bond_mode;
4567         params->xmit_policy = xmit_hashtype;
4568         params->miimon = miimon;
4569         params->num_peer_notif = num_peer_notif;
4570         params->arp_interval = arp_interval;
4571         params->arp_validate = arp_validate_value;
4572         params->arp_all_targets = arp_all_targets_value;
4573         params->updelay = updelay;
4574         params->downdelay = downdelay;
4575         params->use_carrier = use_carrier;
4576         params->lacp_fast = lacp_fast;
4577         params->primary[0] = 0;
4578         params->primary_reselect = primary_reselect_value;
4579         params->fail_over_mac = fail_over_mac_value;
4580         params->tx_queues = tx_queues;
4581         params->all_slaves_active = all_slaves_active;
4582         params->resend_igmp = resend_igmp;
4583         params->min_links = min_links;
4584         params->lp_interval = lp_interval;
4585         params->packets_per_slave = packets_per_slave;
4586         params->tlb_dynamic_lb = 1; /* Default value */
4587         params->ad_actor_sys_prio = ad_actor_sys_prio;
4588         eth_zero_addr(params->ad_actor_system);
4589         params->ad_user_port_key = ad_user_port_key;
4590         if (packets_per_slave > 0) {
4591                 params->reciprocal_packets_per_slave =
4592                         reciprocal_value(packets_per_slave);
4593         } else {
4594                 /* reciprocal_packets_per_slave is unused if
4595                  * packets_per_slave is 0 or 1, just initialize it
4596                  */
4597                 params->reciprocal_packets_per_slave =
4598                         (struct reciprocal_value) { 0 };
4599         }
4600
4601         if (primary) {
4602                 strncpy(params->primary, primary, IFNAMSIZ);
4603                 params->primary[IFNAMSIZ - 1] = 0;
4604         }
4605
4606         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4607
4608         return 0;
4609 }
4610
4611 static struct lock_class_key bonding_netdev_xmit_lock_key;
4612 static struct lock_class_key bonding_netdev_addr_lock_key;
4613 static struct lock_class_key bonding_tx_busylock_key;
4614
4615 static void bond_set_lockdep_class_one(struct net_device *dev,
4616                                        struct netdev_queue *txq,
4617                                        void *_unused)
4618 {
4619         lockdep_set_class(&txq->_xmit_lock,
4620                           &bonding_netdev_xmit_lock_key);
4621 }
4622
4623 static void bond_set_lockdep_class(struct net_device *dev)
4624 {
4625         lockdep_set_class(&dev->addr_list_lock,
4626                           &bonding_netdev_addr_lock_key);
4627         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4628         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4629 }
4630
4631 /* Called from registration process */
4632 static int bond_init(struct net_device *bond_dev)
4633 {
4634         struct bonding *bond = netdev_priv(bond_dev);
4635         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4636
4637         netdev_dbg(bond_dev, "Begin bond_init\n");
4638
4639         bond->wq = create_singlethread_workqueue(bond_dev->name);
4640         if (!bond->wq)
4641                 return -ENOMEM;
4642
4643         bond_set_lockdep_class(bond_dev);
4644
4645         list_add_tail(&bond->bond_list, &bn->dev_list);
4646
4647         bond_prepare_sysfs_group(bond);
4648
4649         bond_debug_register(bond);
4650
4651         /* Ensure valid dev_addr */
4652         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4653             bond_dev->addr_assign_type == NET_ADDR_PERM)
4654                 eth_hw_addr_random(bond_dev);
4655
4656         return 0;
4657 }
4658
4659 unsigned int bond_get_num_tx_queues(void)
4660 {
4661         return tx_queues;
4662 }
4663
4664 /* Create a new bond based on the specified name and bonding parameters.
4665  * If name is NULL, obtain a suitable "bond%d" name for us.
4666  * Caller must NOT hold rtnl_lock; we need to release it here before we
4667  * set up our sysfs entries.
4668  */
4669 int bond_create(struct net *net, const char *name)
4670 {
4671         struct net_device *bond_dev;
4672         struct bonding *bond;
4673         struct alb_bond_info *bond_info;
4674         int res;
4675
4676         rtnl_lock();
4677
4678         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4679                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
4680                                    bond_setup, tx_queues);
4681         if (!bond_dev) {
4682                 pr_err("%s: eek! can't alloc netdev!\n", name);
4683                 rtnl_unlock();
4684                 return -ENOMEM;
4685         }
4686
4687         /*
4688          * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4689          * It is set to 0 by default which is wrong.
4690          */
4691         bond = netdev_priv(bond_dev);
4692         bond_info = &(BOND_ALB_INFO(bond));
4693         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4694
4695         dev_net_set(bond_dev, net);
4696         bond_dev->rtnl_link_ops = &bond_link_ops;
4697
4698         res = register_netdevice(bond_dev);
4699
4700         netif_carrier_off(bond_dev);
4701
4702         rtnl_unlock();
4703         if (res < 0)
4704                 bond_destructor(bond_dev);
4705         return res;
4706 }
4707
4708 static int __net_init bond_net_init(struct net *net)
4709 {
4710         struct bond_net *bn = net_generic(net, bond_net_id);
4711
4712         bn->net = net;
4713         INIT_LIST_HEAD(&bn->dev_list);
4714
4715         bond_create_proc_dir(bn);
4716         bond_create_sysfs(bn);
4717
4718         return 0;
4719 }
4720
4721 static void __net_exit bond_net_exit(struct net *net)
4722 {
4723         struct bond_net *bn = net_generic(net, bond_net_id);
4724         struct bonding *bond, *tmp_bond;
4725         LIST_HEAD(list);
4726
4727         bond_destroy_sysfs(bn);
4728
4729         /* Kill off any bonds created after unregistering bond rtnl ops */
4730         rtnl_lock();
4731         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4732                 unregister_netdevice_queue(bond->dev, &list);
4733         unregister_netdevice_many(&list);
4734         rtnl_unlock();
4735
4736         bond_destroy_proc_dir(bn);
4737 }
4738
4739 static struct pernet_operations bond_net_ops = {
4740         .init = bond_net_init,
4741         .exit = bond_net_exit,
4742         .id   = &bond_net_id,
4743         .size = sizeof(struct bond_net),
4744 };
4745
4746 static int __init bonding_init(void)
4747 {
4748         int i;
4749         int res;
4750
4751         pr_info("%s", bond_version);
4752
4753         res = bond_check_params(&bonding_defaults);
4754         if (res)
4755                 goto out;
4756
4757         res = register_pernet_subsys(&bond_net_ops);
4758         if (res)
4759                 goto out;
4760
4761         res = bond_netlink_init();
4762         if (res)
4763                 goto err_link;
4764
4765         bond_create_debugfs();
4766
4767         for (i = 0; i < max_bonds; i++) {
4768                 res = bond_create(&init_net, NULL);
4769                 if (res)
4770                         goto err;
4771         }
4772
4773         register_netdevice_notifier(&bond_netdev_notifier);
4774 out:
4775         return res;
4776 err:
4777         bond_destroy_debugfs();
4778         bond_netlink_fini();
4779 err_link:
4780         unregister_pernet_subsys(&bond_net_ops);
4781         goto out;
4782
4783 }
4784
4785 static void __exit bonding_exit(void)
4786 {
4787         unregister_netdevice_notifier(&bond_netdev_notifier);
4788
4789         bond_destroy_debugfs();
4790
4791         bond_netlink_fini();
4792         unregister_pernet_subsys(&bond_net_ops);
4793
4794 #ifdef CONFIG_NET_POLL_CONTROLLER
4795         /* Make sure we don't have an imbalance on our netpoll blocking */
4796         WARN_ON(atomic_read(&netpoll_block_tx));
4797 #endif
4798 }
4799
4800 module_init(bonding_init);
4801 module_exit(bonding_exit);
4802 MODULE_LICENSE("GPL");
4803 MODULE_VERSION(DRV_VERSION);
4804 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4805 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");