drivers/net/bonding/bond_main.c

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 #include <linux/kernel.h>
  35 #include <linux/module.h>
  36 #include <linux/types.h>
  37 #include <linux/fcntl.h>
  38 #include <linux/interrupt.h>
  39 #include <linux/ptrace.h>
  40 #include <linux/ioport.h>
  41 #include <linux/in.h>
  42 #include <net/ip.h>
  43 #include <linux/ip.h>
  44 #include <linux/tcp.h>
  45 #include <linux/udp.h>
  46 #include <linux/slab.h>
  47 #include <linux/string.h>
  48 #include <linux/init.h>
  49 #include <linux/timer.h>
  50 #include <linux/socket.h>
  51 #include <linux/ctype.h>
  52 #include <linux/inet.h>
  53 #include <linux/bitops.h>
  54 #include <linux/io.h>
  55 #include <asm/dma.h>
  56 #include <linux/uaccess.h>
  57 #include <linux/errno.h>
  58 #include <linux/netdevice.h>
  59 #include <linux/inetdevice.h>
  60 #include <linux/igmp.h>
  61 #include <linux/etherdevice.h>
  62 #include <linux/skbuff.h>
  63 #include <net/sock.h>
  64 #include <linux/rtnetlink.h>
  65 #include <linux/smp.h>
  66 #include <linux/if_ether.h>
  67 #include <net/arp.h>
  68 #include <linux/mii.h>
  69 #include <linux/ethtool.h>
  70 #include <linux/if_vlan.h>
  71 #include <linux/if_bonding.h>
  72 #include <linux/jiffies.h>
  73 #include <linux/preempt.h>
  74 #include <net/route.h>
  75 #include <net/net_namespace.h>
  76 #include <net/netns/generic.h>
  77 #include <net/pkt_sched.h>
  78 #include <linux/rculist.h>
  79 #include <net/flow_dissector.h>
  80 #include <net/switchdev.h>
  81 #include <net/bonding.h>
  82 #include <net/bond_3ad.h>
  83 #include <net/bond_alb.h>
  84
  85 #include "bonding_priv.h"
  86
  87 /*---------------------------- Module parameters ----------------------------*/
  88
  89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  90
  91 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  92 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  93 static int num_peer_notif = 1;
  94 static int miimon;
  95 static int updelay;
  96 static int downdelay;
  97 static int use_carrier  = 1;
  98 static char *mode;
  99 static char *primary;
 100 static char *primary_reselect;
 101 static char *lacp_rate;
 102 static int min_links;
 103 static char *ad_select;
 104 static char *xmit_hash_policy;
 105 static int arp_interval;
 106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 107 static char *arp_validate;
 108 static char *arp_all_targets;
 109 static char *fail_over_mac;
 110 static int all_slaves_active;
 111 static struct bond_params bonding_defaults;
 112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 113 static int packets_per_slave = 1;
 114 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
 115
 116 module_param(max_bonds, int, 0);
 117 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 118 module_param(tx_queues, int, 0);
 119 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 120 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 121 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 122                                "failover event (alias of num_unsol_na)");
 123 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 124 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 125                                "failover event (alias of num_grat_arp)");
 126 module_param(miimon, int, 0);
 127 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 128 module_param(updelay, int, 0);
 129 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 130 module_param(downdelay, int, 0);
 131 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 132                             "in milliseconds");
 133 module_param(use_carrier, int, 0);
 134 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 135                               "0 for off, 1 for on (default)");
 136 module_param(mode, charp, 0);
 137 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 138                        "1 for active-backup, 2 for balance-xor, "
 139                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 140                        "6 for balance-alb");
 141 module_param(primary, charp, 0);
 142 MODULE_PARM_DESC(primary, "Primary network device to use");
 143 module_param(primary_reselect, charp, 0);
 144 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 145                                    "once it comes up; "
 146                                    "0 for always (default), "
 147                                    "1 for only if speed of primary is "
 148                                    "better, "
 149                                    "2 for only on active slave "
 150                                    "failure");
 151 module_param(lacp_rate, charp, 0);
 152 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 153                             "0 for slow, 1 for fast");
 154 module_param(ad_select, charp, 0);
 155 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 156                             "0 for stable (default), 1 for bandwidth, "
 157                             "2 for count");
 158 module_param(min_links, int, 0);
 159 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 160
 161 module_param(xmit_hash_policy, charp, 0);
 162 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 163                                    "0 for layer 2 (default), 1 for layer 3+4, "
 164                                    "2 for layer 2+3, 3 for encap layer 2+3, "
 165                                    "4 for encap layer 3+4");
 166 module_param(arp_interval, int, 0);
 167 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 168 module_param_array(arp_ip_target, charp, NULL, 0);
 169 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 170 module_param(arp_validate, charp, 0);
 171 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 172                                "0 for none (default), 1 for active, "
 173                                "2 for backup, 3 for all");
 174 module_param(arp_all_targets, charp, 0);
 175 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 176 module_param(fail_over_mac, charp, 0);
 177 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 178                                 "the same MAC; 0 for none (default), "
 179                                 "1 for active, 2 for follow");
 180 module_param(all_slaves_active, int, 0);
 181 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
 182                                      "by setting active flag for all slaves; "
 183                                      "0 for never (default), 1 for always.");
 184 module_param(resend_igmp, int, 0);
 185 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 186                               "link failure");
 187 module_param(packets_per_slave, int, 0);
 188 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
 189                                     "mode; 0 for a random slave, 1 packet per "
 190                                     "slave (default), >1 packets per slave.");
 191 module_param(lp_interval, uint, 0);
 192 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
 193                               "the bonding driver sends learning packets to "
 194                               "each slaves peer switch. The default is 1.");
 195
 196 /*----------------------------- Global variables ----------------------------*/
 197
 198 #ifdef CONFIG_NET_POLL_CONTROLLER
 199 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 200 #endif
 201
 202 int bond_net_id __read_mostly;
 203
 204 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 205 static int arp_ip_count;
 206 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 207 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 208 static int lacp_fast;
 209
 210 /*-------------------------- Forward declarations ---------------------------*/
 211
 212 static int bond_init(struct net_device *bond_dev);
 213 static void bond_uninit(struct net_device *bond_dev);
 214 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
 215                                                 struct rtnl_link_stats64 *stats);
 216 static void bond_slave_arr_handler(struct work_struct *work);
 217 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
 218                                   int mod);
 219
 220 /*---------------------------- General routines -----------------------------*/
 221
 222 const char *bond_mode_name(int mode)
 223 {
 224         static const char *names[] = {
 225                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 226                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 227                 [BOND_MODE_XOR] = "load balancing (xor)",
 228                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 229                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 230                 [BOND_MODE_TLB] = "transmit load balancing",
 231                 [BOND_MODE_ALB] = "adaptive load balancing",
 232         };
 233
 234         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 235                 return "unknown";
 236
 237         return names[mode];
 238 }
 239
 240 /*---------------------------------- VLAN -----------------------------------*/
 241
 242 /**
 243  * bond_dev_queue_xmit - Prepare skb for xmit.
 244  *
 245  * @bond: bond device that got this skb for tx.
 246  * @skb: hw accel VLAN tagged skb to transmit
 247  * @slave_dev: slave that is supposed to xmit this skbuff
 248  */
 249 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 250                         struct net_device *slave_dev)
 251 {
 252         skb->dev = slave_dev;
 253
 254         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 255                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 256         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 257
 258         if (unlikely(netpoll_tx_running(bond->dev)))
 259                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 260         else
 261                 dev_queue_xmit(skb);
 262 }
 263
 264 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 265  * We don't protect the slave list iteration with a lock because:
 266  * a. This operation is performed in IOCTL context,
 267  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 268  * c. Holding a lock with BH disabled while directly calling a base driver
 269  *    entry point is generally a BAD idea.
 270  *
 271  * The design of synchronization/protection for this operation in the 8021q
 272  * module is good for one or more VLAN devices over a single physical device
 273  * and cannot be extended for a teaming solution like bonding, so there is a
 274  * potential race condition here where a net device from the vlan group might
 275  * be referenced (either by a base driver or the 8021q code) while it is being
 276  * removed from the system. However, it turns out we're not making matters
 277  * worse, and if it works for regular VLAN usage it will work here too.
 278 */
 279
 280 /**
 281  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 282  * @bond_dev: bonding net device that got called
 283  * @vid: vlan id being added
 284  */
 285 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 286                                 __be16 proto, u16 vid)
 287 {
 288         struct bonding *bond = netdev_priv(bond_dev);
 289         struct slave *slave, *rollback_slave;
 290         struct list_head *iter;
 291         int res;
 292
 293         bond_for_each_slave(bond, slave, iter) {
 294                 res = vlan_vid_add(slave->dev, proto, vid);
 295                 if (res)
 296                         goto unwind;
 297         }
 298
 299         return 0;
 300
 301 unwind:
 302         /* unwind to the slave that failed */
 303         bond_for_each_slave(bond, rollback_slave, iter) {
 304                 if (rollback_slave == slave)
 305                         break;
 306
 307                 vlan_vid_del(rollback_slave->dev, proto, vid);
 308         }
 309
 310         return res;
 311 }
 312
 313 /**
 314  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 315  * @bond_dev: bonding net device that got called
 316  * @vid: vlan id being removed
 317  */
 318 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 319                                  __be16 proto, u16 vid)
 320 {
 321         struct bonding *bond = netdev_priv(bond_dev);
 322         struct list_head *iter;
 323         struct slave *slave;
 324
 325         bond_for_each_slave(bond, slave, iter)
 326                 vlan_vid_del(slave->dev, proto, vid);
 327
 328         if (bond_is_lb(bond))
 329                 bond_alb_clear_vlan(bond, vid);
 330
 331         return 0;
 332 }
 333
 334 /*------------------------------- Link status -------------------------------*/
 335
 336 /* Set the carrier state for the master according to the state of its
 337  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 338  * do special 802.3ad magic.
 339  *
 340  * Returns zero if carrier state does not change, nonzero if it does.
 341  */
 342 int bond_set_carrier(struct bonding *bond)
 343 {
 344         struct list_head *iter;
 345         struct slave *slave;
 346
 347         if (!bond_has_slaves(bond))
 348                 goto down;
 349
 350         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 351                 return bond_3ad_set_carrier(bond);
 352
 353         bond_for_each_slave(bond, slave, iter) {
 354                 if (slave->link == BOND_LINK_UP) {
 355                         if (!netif_carrier_ok(bond->dev)) {
 356                                 netif_carrier_on(bond->dev);
 357                                 return 1;
 358                         }
 359                         return 0;
 360                 }
 361         }
 362
 363 down:
 364         if (netif_carrier_ok(bond->dev)) {
 365                 netif_carrier_off(bond->dev);
 366                 return 1;
 367         }
 368         return 0;
 369 }
 370
 371 /* Get link speed and duplex from the slave's base driver
 372  * using ethtool. If for some reason the call fails or the
 373  * values are invalid, set speed and duplex to -1,
 374  * and return.
 375  */
 376 static void bond_update_speed_duplex(struct slave *slave)
 377 {
 378         struct net_device *slave_dev = slave->dev;
 379         struct ethtool_cmd ecmd;
 380         u32 slave_speed;
 381         int res;
 382
 383         slave->speed = SPEED_UNKNOWN;
 384         slave->duplex = DUPLEX_UNKNOWN;
 385
 386         res = __ethtool_get_settings(slave_dev, &ecmd);
 387         if (res < 0)
 388                 return;
 389
 390         slave_speed = ethtool_cmd_speed(&ecmd);
 391         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 392                 return;
 393
 394         switch (ecmd.duplex) {
 395         case DUPLEX_FULL:
 396         case DUPLEX_HALF:
 397                 break;
 398         default:
 399                 return;
 400         }
 401
 402         slave->speed = slave_speed;
 403         slave->duplex = ecmd.duplex;
 404
 405         return;
 406 }
 407
 408 const char *bond_slave_link_status(s8 link)
 409 {
 410         switch (link) {
 411         case BOND_LINK_UP:
 412                 return "up";
 413         case BOND_LINK_FAIL:
 414                 return "going down";
 415         case BOND_LINK_DOWN:
 416                 return "down";
 417         case BOND_LINK_BACK:
 418                 return "going back";
 419         default:
 420                 return "unknown";
 421         }
 422 }
 423
 424 /* if <dev> supports MII link status reporting, check its link status.
 425  *
 426  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 427  * depending upon the setting of the use_carrier parameter.
 428  *
 429  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 430  * can't tell and just pretend it is), or 0, meaning that the link is
 431  * down.
 432  *
 433  * If reporting is non-zero, instead of faking link up, return -1 if
 434  * both ETHTOOL and MII ioctls fail (meaning the device does not
 435  * support them).  If use_carrier is set, return whatever it says.
 436  * It'd be nice if there was a good way to tell if a driver supports
 437  * netif_carrier, but there really isn't.
 438  */
 439 static int bond_check_dev_link(struct bonding *bond,
 440                                struct net_device *slave_dev, int reporting)
 441 {
 442         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 443         int (*ioctl)(struct net_device *, struct ifreq *, int);
 444         struct ifreq ifr;
 445         struct mii_ioctl_data *mii;
 446
 447         if (!reporting && !netif_running(slave_dev))
 448                 return 0;
 449
 450         if (bond->params.use_carrier)
 451                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 452
 453         /* Try to get link status using Ethtool first. */
 454         if (slave_dev->ethtool_ops->get_link)
 455                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 456                         BMSR_LSTATUS : 0;
 457
 458         /* Ethtool can't be used, fallback to MII ioctls. */
 459         ioctl = slave_ops->ndo_do_ioctl;
 460         if (ioctl) {
 461                 /* TODO: set pointer to correct ioctl on a per team member
 462                  *       bases to make this more efficient. that is, once
 463                  *       we determine the correct ioctl, we will always
 464                  *       call it and not the others for that team
 465                  *       member.
 466                  */
 467
 468                 /* We cannot assume that SIOCGMIIPHY will also read a
 469                  * register; not all network drivers (e.g., e100)
 470                  * support that.
 471                  */
 472
 473                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 474                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 475                 mii = if_mii(&ifr);
 476                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 477                         mii->reg_num = MII_BMSR;
 478                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 479                                 return mii->val_out & BMSR_LSTATUS;
 480                 }
 481         }
 482
 483         /* If reporting, report that either there's no dev->do_ioctl,
 484          * or both SIOCGMIIREG and get_link failed (meaning that we
 485          * cannot report link status).  If not reporting, pretend
 486          * we're ok.
 487          */
 488         return reporting ? -1 : BMSR_LSTATUS;
 489 }
 490
 491 /*----------------------------- Multicast list ------------------------------*/
 492
 493 /* Push the promiscuity flag down to appropriate slaves */
 494 static int bond_set_promiscuity(struct bonding *bond, int inc)
 495 {
 496         struct list_head *iter;
 497         int err = 0;
 498
 499         if (bond_uses_primary(bond)) {
 500                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 501
 502                 if (curr_active)
 503                         err = dev_set_promiscuity(curr_active->dev, inc);
 504         } else {
 505                 struct slave *slave;
 506
 507                 bond_for_each_slave(bond, slave, iter) {
 508                         err = dev_set_promiscuity(slave->dev, inc);
 509                         if (err)
 510                                 return err;
 511                 }
 512         }
 513         return err;
 514 }
 515
 516 /* Push the allmulti flag down to all slaves */
 517 static int bond_set_allmulti(struct bonding *bond, int inc)
 518 {
 519         struct list_head *iter;
 520         int err = 0;
 521
 522         if (bond_uses_primary(bond)) {
 523                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 524
 525                 if (curr_active)
 526                         err = dev_set_allmulti(curr_active->dev, inc);
 527         } else {
 528                 struct slave *slave;
 529
 530                 bond_for_each_slave(bond, slave, iter) {
 531                         err = dev_set_allmulti(slave->dev, inc);
 532                         if (err)
 533                                 return err;
 534                 }
 535         }
 536         return err;
 537 }
 538
 539 /* Retrieve the list of registered multicast addresses for the bonding
 540  * device and retransmit an IGMP JOIN request to the current active
 541  * slave.
 542  */
 543 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 544 {
 545         struct bonding *bond = container_of(work, struct bonding,
 546                                             mcast_work.work);
 547
 548         if (!rtnl_trylock()) {
 549                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 550                 return;
 551         }
 552         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 553
 554         if (bond->igmp_retrans > 1) {
 555                 bond->igmp_retrans--;
 556                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 557         }
 558         rtnl_unlock();
 559 }
 560
 561 /* Flush bond's hardware addresses from slave */
 562 static void bond_hw_addr_flush(struct net_device *bond_dev,
 563                                struct net_device *slave_dev)
 564 {
 565         struct bonding *bond = netdev_priv(bond_dev);
 566
 567         dev_uc_unsync(slave_dev, bond_dev);
 568         dev_mc_unsync(slave_dev, bond_dev);
 569
 570         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 571                 /* del lacpdu mc addr from mc list */
 572                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 573
 574                 dev_mc_del(slave_dev, lacpdu_multicast);
 575         }
 576 }
 577
 578 /*--------------------------- Active slave change ---------------------------*/
 579
 580 /* Update the hardware address list and promisc/allmulti for the new and
 581  * old active slaves (if any).  Modes that are not using primary keep all
 582  * slaves up date at all times; only the modes that use primary need to call
 583  * this function to swap these settings during a failover.
 584  */
 585 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 586                               struct slave *old_active)
 587 {
 588         if (old_active) {
 589                 if (bond->dev->flags & IFF_PROMISC)
 590                         dev_set_promiscuity(old_active->dev, -1);
 591
 592                 if (bond->dev->flags & IFF_ALLMULTI)
 593                         dev_set_allmulti(old_active->dev, -1);
 594
 595                 bond_hw_addr_flush(bond->dev, old_active->dev);
 596         }
 597
 598         if (new_active) {
 599                 /* FIXME: Signal errors upstream. */
 600                 if (bond->dev->flags & IFF_PROMISC)
 601                         dev_set_promiscuity(new_active->dev, 1);
 602
 603                 if (bond->dev->flags & IFF_ALLMULTI)
 604                         dev_set_allmulti(new_active->dev, 1);
 605
 606                 netif_addr_lock_bh(bond->dev);
 607                 dev_uc_sync(new_active->dev, bond->dev);
 608                 dev_mc_sync(new_active->dev, bond->dev);
 609                 netif_addr_unlock_bh(bond->dev);
 610         }
 611 }
 612
 613 /**
 614  * bond_set_dev_addr - clone slave's address to bond
 615  * @bond_dev: bond net device
 616  * @slave_dev: slave net device
 617  *
 618  * Should be called with RTNL held.
 619  */
 620 static void bond_set_dev_addr(struct net_device *bond_dev,
 621                               struct net_device *slave_dev)
 622 {
 623         netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 624                    bond_dev, slave_dev, slave_dev->addr_len);
 625         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 626         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 627         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 628 }
 629
 630 static struct slave *bond_get_old_active(struct bonding *bond,
 631                                          struct slave *new_active)
 632 {
 633         struct slave *slave;
 634         struct list_head *iter;
 635
 636         bond_for_each_slave(bond, slave, iter) {
 637                 if (slave == new_active)
 638                         continue;
 639
 640                 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
 641                         return slave;
 642         }
 643
 644         return NULL;
 645 }
 646
 647 /* bond_do_fail_over_mac
 648  *
 649  * Perform special MAC address swapping for fail_over_mac settings
 650  *
 651  * Called with RTNL
 652  */
 653 static void bond_do_fail_over_mac(struct bonding *bond,
 654                                   struct slave *new_active,
 655                                   struct slave *old_active)
 656 {
 657         u8 tmp_mac[ETH_ALEN];
 658         struct sockaddr saddr;
 659         int rv;
 660
 661         switch (bond->params.fail_over_mac) {
 662         case BOND_FOM_ACTIVE:
 663                 if (new_active)
 664                         bond_set_dev_addr(bond->dev, new_active->dev);
 665                 break;
 666         case BOND_FOM_FOLLOW:
 667                 /* if new_active && old_active, swap them
 668                  * if just old_active, do nothing (going to no active slave)
 669                  * if just new_active, set new_active to bond's MAC
 670                  */
 671                 if (!new_active)
 672                         return;
 673
 674                 if (!old_active)
 675                         old_active = bond_get_old_active(bond, new_active);
 676
 677                 if (old_active) {
 678                         ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
 679                         ether_addr_copy(saddr.sa_data,
 680                                         old_active->dev->dev_addr);
 681                         saddr.sa_family = new_active->dev->type;
 682                 } else {
 683                         ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
 684                         saddr.sa_family = bond->dev->type;
 685                 }
 686
 687                 rv = dev_set_mac_address(new_active->dev, &saddr);
 688                 if (rv) {
 689                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 690                                    -rv, new_active->dev->name);
 691                         goto out;
 692                 }
 693
 694                 if (!old_active)
 695                         goto out;
 696
 697                 ether_addr_copy(saddr.sa_data, tmp_mac);
 698                 saddr.sa_family = old_active->dev->type;
 699
 700                 rv = dev_set_mac_address(old_active->dev, &saddr);
 701                 if (rv)
 702                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 703                                    -rv, new_active->dev->name);
 704 out:
 705                 break;
 706         default:
 707                 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
 708                            bond->params.fail_over_mac);
 709                 break;
 710         }
 711
 712 }
 713
 714 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
 715 {
 716         struct slave *prim = rtnl_dereference(bond->primary_slave);
 717         struct slave *curr = rtnl_dereference(bond->curr_active_slave);
 718
 719         if (!prim || prim->link != BOND_LINK_UP) {
 720                 if (!curr || curr->link != BOND_LINK_UP)
 721                         return NULL;
 722                 return curr;
 723         }
 724
 725         if (bond->force_primary) {
 726                 bond->force_primary = false;
 727                 return prim;
 728         }
 729
 730         if (!curr || curr->link != BOND_LINK_UP)
 731                 return prim;
 732
 733         /* At this point, prim and curr are both up */
 734         switch (bond->params.primary_reselect) {
 735         case BOND_PRI_RESELECT_ALWAYS:
 736                 return prim;
 737         case BOND_PRI_RESELECT_BETTER:
 738                 if (prim->speed < curr->speed)
 739                         return curr;
 740                 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
 741                         return curr;
 742                 return prim;
 743         case BOND_PRI_RESELECT_FAILURE:
 744                 return curr;
 745         default:
 746                 netdev_err(bond->dev, "impossible primary_reselect %d\n",
 747                            bond->params.primary_reselect);
 748                 return curr;
 749         }
 750 }
 751
 752 /**
 753  * bond_find_best_slave - select the best available slave to be the active one
 754  * @bond: our bonding struct
 755  */
 756 static struct slave *bond_find_best_slave(struct bonding *bond)
 757 {
 758         struct slave *slave, *bestslave = NULL;
 759         struct list_head *iter;
 760         int mintime = bond->params.updelay;
 761
 762         slave = bond_choose_primary_or_current(bond);
 763         if (slave)
 764                 return slave;
 765
 766         bond_for_each_slave(bond, slave, iter) {
 767                 if (slave->link == BOND_LINK_UP)
 768                         return slave;
 769                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
 770                     slave->delay < mintime) {
 771                         mintime = slave->delay;
 772                         bestslave = slave;
 773                 }
 774         }
 775
 776         return bestslave;
 777 }
 778
 779 static bool bond_should_notify_peers(struct bonding *bond)
 780 {
 781         struct slave *slave;
 782
 783         rcu_read_lock();
 784         slave = rcu_dereference(bond->curr_active_slave);
 785         rcu_read_unlock();
 786
 787         netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
 788                    slave ? slave->dev->name : "NULL");
 789
 790         if (!slave || !bond->send_peer_notif ||
 791             !netif_carrier_ok(bond->dev) ||
 792             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 793                 return false;
 794
 795         return true;
 796 }
 797
 798 /**
 799  * change_active_interface - change the active slave into the specified one
 800  * @bond: our bonding struct
 801  * @new: the new slave to make the active one
 802  *
 803  * Set the new slave to the bond's settings and unset them on the old
 804  * curr_active_slave.
 805  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 806  *
 807  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 808  * because it is apparently the best available slave we have, even though its
 809  * updelay hasn't timed out yet.
 810  *
 811  * Caller must hold RTNL.
 812  */
 813 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 814 {
 815         struct slave *old_active;
 816
 817         ASSERT_RTNL();
 818
 819         old_active = rtnl_dereference(bond->curr_active_slave);
 820
 821         if (old_active == new_active)
 822                 return;
 823
 824         if (new_active) {
 825                 new_active->last_link_up = jiffies;
 826
 827                 if (new_active->link == BOND_LINK_BACK) {
 828                         if (bond_uses_primary(bond)) {
 829                                 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
 830                                             new_active->dev->name,
 831                                             (bond->params.updelay - new_active->delay) * bond->params.miimon);
 832                         }
 833
 834                         new_active->delay = 0;
 835                         bond_set_slave_link_state(new_active, BOND_LINK_UP,
 836                                                   BOND_SLAVE_NOTIFY_NOW);
 837
 838                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 839                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 840
 841                         if (bond_is_lb(bond))
 842                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 843                 } else {
 844                         if (bond_uses_primary(bond)) {
 845                                 netdev_info(bond->dev, "making interface %s the new active one\n",
 846                                             new_active->dev->name);
 847                         }
 848                 }
 849         }
 850
 851         if (bond_uses_primary(bond))
 852                 bond_hw_addr_swap(bond, new_active, old_active);
 853
 854         if (bond_is_lb(bond)) {
 855                 bond_alb_handle_active_change(bond, new_active);
 856                 if (old_active)
 857                         bond_set_slave_inactive_flags(old_active,
 858                                                       BOND_SLAVE_NOTIFY_NOW);
 859                 if (new_active)
 860                         bond_set_slave_active_flags(new_active,
 861                                                     BOND_SLAVE_NOTIFY_NOW);
 862         } else {
 863                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 864         }
 865
 866         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
 867                 if (old_active)
 868                         bond_set_slave_inactive_flags(old_active,
 869                                                       BOND_SLAVE_NOTIFY_NOW);
 870
 871                 if (new_active) {
 872                         bool should_notify_peers = false;
 873
 874                         bond_set_slave_active_flags(new_active,
 875                                                     BOND_SLAVE_NOTIFY_NOW);
 876
 877                         if (bond->params.fail_over_mac)
 878                                 bond_do_fail_over_mac(bond, new_active,
 879                                                       old_active);
 880
 881                         if (netif_running(bond->dev)) {
 882                                 bond->send_peer_notif =
 883                                         bond->params.num_peer_notif;
 884                                 should_notify_peers =
 885                                         bond_should_notify_peers(bond);
 886                         }
 887
 888                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 889                         if (should_notify_peers)
 890                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 891                                                          bond->dev);
 892                 }
 893         }
 894
 895         /* resend IGMP joins since active slave has changed or
 896          * all were sent on curr_active_slave.
 897          * resend only if bond is brought up with the affected
 898          * bonding modes and the retransmission is enabled
 899          */
 900         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 901             ((bond_uses_primary(bond) && new_active) ||
 902              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
 903                 bond->igmp_retrans = bond->params.resend_igmp;
 904                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 905         }
 906 }
 907
 908 /**
 909  * bond_select_active_slave - select a new active slave, if needed
 910  * @bond: our bonding struct
 911  *
 912  * This functions should be called when one of the following occurs:
 913  * - The old curr_active_slave has been released or lost its link.
 914  * - The primary_slave has got its link back.
 915  * - A slave has got its link back and there's no old curr_active_slave.
 916  *
 917  * Caller must hold RTNL.
 918  */
 919 void bond_select_active_slave(struct bonding *bond)
 920 {
 921         struct slave *best_slave;
 922         int rv;
 923
 924         ASSERT_RTNL();
 925
 926         best_slave = bond_find_best_slave(bond);
 927         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
 928                 bond_change_active_slave(bond, best_slave);
 929                 rv = bond_set_carrier(bond);
 930                 if (!rv)
 931                         return;
 932
 933                 if (netif_carrier_ok(bond->dev)) {
 934                         netdev_info(bond->dev, "first active interface up!\n");
 935                 } else {
 936                         netdev_info(bond->dev, "now running without any active interface!\n");
 937                 }
 938         }
 939 }
 940
 941 #ifdef CONFIG_NET_POLL_CONTROLLER
 942 static inline int slave_enable_netpoll(struct slave *slave)
 943 {
 944         struct netpoll *np;
 945         int err = 0;
 946
 947         np = kzalloc(sizeof(*np), GFP_KERNEL);
 948         err = -ENOMEM;
 949         if (!np)
 950                 goto out;
 951
 952         err = __netpoll_setup(np, slave->dev);
 953         if (err) {
 954                 kfree(np);
 955                 goto out;
 956         }
 957         slave->np = np;
 958 out:
 959         return err;
 960 }
 961 static inline void slave_disable_netpoll(struct slave *slave)
 962 {
 963         struct netpoll *np = slave->np;
 964
 965         if (!np)
 966                 return;
 967
 968         slave->np = NULL;
 969         __netpoll_free_async(np);
 970 }
 971
 972 static void bond_poll_controller(struct net_device *bond_dev)
 973 {
 974         struct bonding *bond = netdev_priv(bond_dev);
 975         struct slave *slave = NULL;
 976         struct list_head *iter;
 977         struct ad_info ad_info;
 978         struct netpoll_info *ni;
 979         const struct net_device_ops *ops;
 980
 981         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 982                 if (bond_3ad_get_active_agg_info(bond, &ad_info))
 983                         return;
 984
 985         bond_for_each_slave_rcu(bond, slave, iter) {
 986                 ops = slave->dev->netdev_ops;
 987                 if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller)
 988                         continue;
 989
 990                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 991                         struct aggregator *agg =
 992                             SLAVE_AD_INFO(slave)->port.aggregator;
 993
 994                         if (agg &&
 995                             agg->aggregator_identifier != ad_info.aggregator_id)
 996                                 continue;
 997                 }
 998
 999                 ni = rcu_dereference_bh(slave->dev->npinfo);
1000                 if (down_trylock(&ni->dev_lock))
1001                         continue;
1002                 ops->ndo_poll_controller(slave->dev);
1003                 up(&ni->dev_lock);
1004         }
1005 }
1006
1007 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1008 {
1009         struct bonding *bond = netdev_priv(bond_dev);
1010         struct list_head *iter;
1011         struct slave *slave;
1012
1013         bond_for_each_slave(bond, slave, iter)
1014                 if (bond_slave_is_up(slave))
1015                         slave_disable_netpoll(slave);
1016 }
1017
1018 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1019 {
1020         struct bonding *bond = netdev_priv(dev);
1021         struct list_head *iter;
1022         struct slave *slave;
1023         int err = 0;
1024
1025         bond_for_each_slave(bond, slave, iter) {
1026                 err = slave_enable_netpoll(slave);
1027                 if (err) {
1028                         bond_netpoll_cleanup(dev);
1029                         break;
1030                 }
1031         }
1032         return err;
1033 }
1034 #else
1035 static inline int slave_enable_netpoll(struct slave *slave)
1036 {
1037         return 0;
1038 }
1039 static inline void slave_disable_netpoll(struct slave *slave)
1040 {
1041 }
1042 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1043 {
1044 }
1045 #endif
1046
1047 /*---------------------------------- IOCTL ----------------------------------*/
1048
1049 static netdev_features_t bond_fix_features(struct net_device *dev,
1050                                            netdev_features_t features)
1051 {
1052         struct bonding *bond = netdev_priv(dev);
1053         struct list_head *iter;
1054         netdev_features_t mask;
1055         struct slave *slave;
1056
1057         mask = features;
1058
1059         features &= ~NETIF_F_ONE_FOR_ALL;
1060         features |= NETIF_F_ALL_FOR_ALL;
1061
1062         bond_for_each_slave(bond, slave, iter) {
1063                 features = netdev_increment_features(features,
1064                                                      slave->dev->features,
1065                                                      mask);
1066         }
1067         features = netdev_add_tso_features(features, mask);
1068
1069         return features;
1070 }
1071
1072 #define BOND_VLAN_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1073                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1074                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1075
1076 #define BOND_ENC_FEATURES       (NETIF_F_HW_CSUM | NETIF_F_SG | \
1077                                  NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1078
1079 static void bond_compute_features(struct bonding *bond)
1080 {
1081         unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1082                                         IFF_XMIT_DST_RELEASE_PERM;
1083         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1084         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1085         struct net_device *bond_dev = bond->dev;
1086         struct list_head *iter;
1087         struct slave *slave;
1088         unsigned short max_hard_header_len = ETH_HLEN;
1089         unsigned int gso_max_size = GSO_MAX_SIZE;
1090         u16 gso_max_segs = GSO_MAX_SEGS;
1091
1092         if (!bond_has_slaves(bond))
1093                 goto done;
1094         vlan_features &= NETIF_F_ALL_FOR_ALL;
1095
1096         bond_for_each_slave(bond, slave, iter) {
1097                 vlan_features = netdev_increment_features(vlan_features,
1098                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1099
1100                 enc_features = netdev_increment_features(enc_features,
1101                                                          slave->dev->hw_enc_features,
1102                                                          BOND_ENC_FEATURES);
1103                 dst_release_flag &= slave->dev->priv_flags;
1104                 if (slave->dev->hard_header_len > max_hard_header_len)
1105                         max_hard_header_len = slave->dev->hard_header_len;
1106
1107                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1108                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1109         }
1110
1111 done:
1112         bond_dev->vlan_features = vlan_features;
1113         bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
1114         bond_dev->hard_header_len = max_hard_header_len;
1115         bond_dev->gso_max_segs = gso_max_segs;
1116         netif_set_gso_max_size(bond_dev, gso_max_size);
1117
1118         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1119         if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1120             dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1121                 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1122
1123         netdev_change_features(bond_dev);
1124 }
1125
1126 static void bond_setup_by_slave(struct net_device *bond_dev,
1127                                 struct net_device *slave_dev)
1128 {
1129         bond_dev->header_ops        = slave_dev->header_ops;
1130
1131         bond_dev->type              = slave_dev->type;
1132         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1133         bond_dev->addr_len          = slave_dev->addr_len;
1134
1135         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1136                 slave_dev->addr_len);
1137 }
1138
1139 /* On bonding slaves other than the currently active slave, suppress
1140  * duplicates except for alb non-mcast/bcast.
1141  */
1142 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1143                                             struct slave *slave,
1144                                             struct bonding *bond)
1145 {
1146         if (bond_is_slave_inactive(slave)) {
1147                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1148                     skb->pkt_type != PACKET_BROADCAST &&
1149                     skb->pkt_type != PACKET_MULTICAST)
1150                         return false;
1151                 return true;
1152         }
1153         return false;
1154 }
1155
1156 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1157 {
1158         struct sk_buff *skb = *pskb;
1159         struct slave *slave;
1160         struct bonding *bond;
1161         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1162                           struct slave *);
1163         int ret = RX_HANDLER_ANOTHER;
1164
1165         skb = skb_share_check(skb, GFP_ATOMIC);
1166         if (unlikely(!skb))
1167                 return RX_HANDLER_CONSUMED;
1168
1169         *pskb = skb;
1170
1171         slave = bond_slave_get_rcu(skb->dev);
1172         bond = slave->bond;
1173
1174         recv_probe = ACCESS_ONCE(bond->recv_probe);
1175         if (recv_probe) {
1176                 ret = recv_probe(skb, bond, slave);
1177                 if (ret == RX_HANDLER_CONSUMED) {
1178                         consume_skb(skb);
1179                         return ret;
1180                 }
1181         }
1182
1183         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1184                 return RX_HANDLER_EXACT;
1185         }
1186
1187         skb->dev = bond->dev;
1188
1189         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1190             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1191             skb->pkt_type == PACKET_HOST) {
1192
1193                 if (unlikely(skb_cow_head(skb,
1194                                           skb->data - skb_mac_header(skb)))) {
1195                         kfree_skb(skb);
1196                         return RX_HANDLER_CONSUMED;
1197                 }
1198                 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1199         }
1200
1201         return ret;
1202 }
1203
1204 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1205 {
1206         switch (BOND_MODE(bond)) {
1207         case BOND_MODE_ROUNDROBIN:
1208                 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1209         case BOND_MODE_ACTIVEBACKUP:
1210                 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1211         case BOND_MODE_BROADCAST:
1212                 return NETDEV_LAG_TX_TYPE_BROADCAST;
1213         case BOND_MODE_XOR:
1214         case BOND_MODE_8023AD:
1215                 return NETDEV_LAG_TX_TYPE_HASH;
1216         default:
1217                 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1218         }
1219 }
1220
1221 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave)
1222 {
1223         struct netdev_lag_upper_info lag_upper_info;
1224         int err;
1225
1226         lag_upper_info.tx_type = bond_lag_tx_type(bond);
1227         err = netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1228                                            &lag_upper_info);
1229         if (err)
1230                 return err;
1231         rtmsg_ifinfo(RTM_NEWLINK, slave->dev, IFF_SLAVE, GFP_KERNEL);
1232         return 0;
1233 }
1234
1235 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1236 {
1237         netdev_upper_dev_unlink(slave->dev, bond->dev);
1238         slave->dev->flags &= ~IFF_SLAVE;
1239         rtmsg_ifinfo(RTM_NEWLINK, slave->dev, IFF_SLAVE, GFP_KERNEL);
1240 }
1241
1242 static struct slave *bond_alloc_slave(struct bonding *bond)
1243 {
1244         struct slave *slave = NULL;
1245
1246         slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1247         if (!slave)
1248                 return NULL;
1249
1250         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1251                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1252                                                GFP_KERNEL);
1253                 if (!SLAVE_AD_INFO(slave)) {
1254                         kfree(slave);
1255                         return NULL;
1256                 }
1257         }
1258         return slave;
1259 }
1260
1261 static void bond_free_slave(struct slave *slave)
1262 {
1263         struct bonding *bond = bond_get_bond_by_slave(slave);
1264
1265         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1266                 kfree(SLAVE_AD_INFO(slave));
1267
1268         kfree(slave);
1269 }
1270
1271 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1272 {
1273         info->bond_mode = BOND_MODE(bond);
1274         info->miimon = bond->params.miimon;
1275         info->num_slaves = bond->slave_cnt;
1276 }
1277
1278 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1279 {
1280         strcpy(info->slave_name, slave->dev->name);
1281         info->link = slave->link;
1282         info->state = bond_slave_state(slave);
1283         info->link_failure_count = slave->link_failure_count;
1284 }
1285
1286 static void bond_netdev_notify(struct net_device *dev,
1287                                struct netdev_bonding_info *info)
1288 {
1289         rtnl_lock();
1290         netdev_bonding_info_change(dev, info);
1291         rtnl_unlock();
1292 }
1293
1294 static void bond_netdev_notify_work(struct work_struct *_work)
1295 {
1296         struct netdev_notify_work *w =
1297                 container_of(_work, struct netdev_notify_work, work.work);
1298
1299         bond_netdev_notify(w->dev, &w->bonding_info);
1300         dev_put(w->dev);
1301         kfree(w);
1302 }
1303
1304 void bond_queue_slave_event(struct slave *slave)
1305 {
1306         struct bonding *bond = slave->bond;
1307         struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1308
1309         if (!nnw)
1310                 return;
1311
1312         dev_hold(slave->dev);
1313         nnw->dev = slave->dev;
1314         bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1315         bond_fill_ifbond(bond, &nnw->bonding_info.master);
1316         INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1317
1318         queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1319 }
1320
1321 void bond_lower_state_changed(struct slave *slave)
1322 {
1323         struct netdev_lag_lower_state_info info;
1324
1325         info.link_up = slave->link == BOND_LINK_UP ||
1326                        slave->link == BOND_LINK_FAIL;
1327         info.tx_enabled = bond_is_active_slave(slave);
1328         netdev_lower_state_changed(slave->dev, &info);
1329 }
1330
1331 /* enslave device <slave> to bond device <master> */
1332 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1333 {
1334         struct bonding *bond = netdev_priv(bond_dev);
1335         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1336         struct slave *new_slave = NULL, *prev_slave;
1337         struct sockaddr addr;
1338         int link_reporting;
1339         int res = 0, i;
1340
1341         if (!bond->params.use_carrier &&
1342             slave_dev->ethtool_ops->get_link == NULL &&
1343             slave_ops->ndo_do_ioctl == NULL) {
1344                 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1345                             slave_dev->name);
1346         }
1347
1348         /* already enslaved */
1349         if (slave_dev->flags & IFF_SLAVE) {
1350                 netdev_dbg(bond_dev, "Error: Device was already enslaved\n");
1351                 return -EBUSY;
1352         }
1353
1354         if (bond_dev == slave_dev) {
1355                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1356                 return -EPERM;
1357         }
1358
1359         /* vlan challenged mutual exclusion */
1360         /* no need to lock since we're protected by rtnl_lock */
1361         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1362                 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1363                            slave_dev->name);
1364                 if (vlan_uses_dev(bond_dev)) {
1365                         netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1366                                    slave_dev->name, bond_dev->name);
1367                         return -EPERM;
1368                 } else {
1369                         netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1370                                     slave_dev->name, slave_dev->name,
1371                                     bond_dev->name);
1372                 }
1373         } else {
1374                 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1375                            slave_dev->name);
1376         }
1377
1378         /* Old ifenslave binaries are no longer supported.  These can
1379          * be identified with moderate accuracy by the state of the slave:
1380          * the current ifenslave will set the interface down prior to
1381          * enslaving it; the old ifenslave will not.
1382          */
1383         if (slave_dev->flags & IFF_UP) {
1384                 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1385                            slave_dev->name);
1386                 res = -EPERM;
1387                 goto err_undo_flags;
1388         }
1389
1390         /* set bonding device ether type by slave - bonding netdevices are
1391          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1392          * there is a need to override some of the type dependent attribs/funcs.
1393          *
1394          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1395          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1396          */
1397         if (!bond_has_slaves(bond)) {
1398                 if (bond_dev->type != slave_dev->type) {
1399                         netdev_dbg(bond_dev, "change device type from %d to %d\n",
1400                                    bond_dev->type, slave_dev->type);
1401
1402                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1403                                                        bond_dev);
1404                         res = notifier_to_errno(res);
1405                         if (res) {
1406                                 netdev_err(bond_dev, "refused to change device type\n");
1407                                 res = -EBUSY;
1408                                 goto err_undo_flags;
1409                         }
1410
1411                         /* Flush unicast and multicast addresses */
1412                         dev_uc_flush(bond_dev);
1413                         dev_mc_flush(bond_dev);
1414
1415                         if (slave_dev->type != ARPHRD_ETHER)
1416                                 bond_setup_by_slave(bond_dev, slave_dev);
1417                         else {
1418                                 ether_setup(bond_dev);
1419                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1420                         }
1421
1422                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1423                                                  bond_dev);
1424                 }
1425         } else if (bond_dev->type != slave_dev->type) {
1426                 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1427                            slave_dev->name, slave_dev->type, bond_dev->type);
1428                 res = -EINVAL;
1429                 goto err_undo_flags;
1430         }
1431
1432         if (slave_ops->ndo_set_mac_address == NULL) {
1433                 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1434                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1435                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1436                         if (!bond_has_slaves(bond)) {
1437                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1438                                 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1439                         } else {
1440                                 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");
1441                                 res = -EOPNOTSUPP;
1442                                 goto err_undo_flags;
1443                         }
1444                 }
1445         }
1446
1447         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1448
1449         /* If this is the first slave, then we need to set the master's hardware
1450          * address to be the same as the slave's.
1451          */
1452         if (!bond_has_slaves(bond) &&
1453             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1454                 bond_set_dev_addr(bond->dev, slave_dev);
1455
1456         new_slave = bond_alloc_slave(bond);
1457         if (!new_slave) {
1458                 res = -ENOMEM;
1459                 goto err_undo_flags;
1460         }
1461
1462         new_slave->bond = bond;
1463         new_slave->dev = slave_dev;
1464         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1465          * is set via sysfs or module option if desired.
1466          */
1467         new_slave->queue_id = 0;
1468
1469         /* Save slave's original mtu and then set it to match the bond */
1470         new_slave->original_mtu = slave_dev->mtu;
1471         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1472         if (res) {
1473                 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1474                 goto err_free;
1475         }
1476
1477         /* Save slave's original ("permanent") mac address for modes
1478          * that need it, and for restoring it upon release, and then
1479          * set it to the master's address
1480          */
1481         ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1482
1483         if (!bond->params.fail_over_mac ||
1484             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1485                 /* Set slave to master's mac address.  The application already
1486                  * set the master's mac address to that of the first slave
1487                  */
1488                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1489                 addr.sa_family = slave_dev->type;
1490                 res = dev_set_mac_address(slave_dev, &addr);
1491                 if (res) {
1492                         netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1493                         goto err_restore_mtu;
1494                 }
1495         }
1496
1497         /* set slave flag before open to prevent IPv6 addrconf */
1498         slave_dev->flags |= IFF_SLAVE;
1499
1500         /* open the slave since the application closed it */
1501         res = dev_open(slave_dev);
1502         if (res) {
1503                 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1504                 goto err_restore_mac;
1505         }
1506
1507         slave_dev->priv_flags |= IFF_BONDING;
1508         /* initialize slave stats */
1509         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1510
1511         if (bond_is_lb(bond)) {
1512                 /* bond_alb_init_slave() must be called before all other stages since
1513                  * it might fail and we do not want to have to undo everything
1514                  */
1515                 res = bond_alb_init_slave(bond, new_slave);
1516                 if (res)
1517                         goto err_close;
1518         }
1519
1520         /* If the mode uses primary, then the following is handled by
1521          * bond_change_active_slave().
1522          */
1523         if (!bond_uses_primary(bond)) {
1524                 /* set promiscuity level to new slave */
1525                 if (bond_dev->flags & IFF_PROMISC) {
1526                         res = dev_set_promiscuity(slave_dev, 1);
1527                         if (res)
1528                                 goto err_close;
1529                 }
1530
1531                 /* set allmulti level to new slave */
1532                 if (bond_dev->flags & IFF_ALLMULTI) {
1533                         res = dev_set_allmulti(slave_dev, 1);
1534                         if (res)
1535                                 goto err_close;
1536                 }
1537
1538                 netif_addr_lock_bh(bond_dev);
1539
1540                 dev_mc_sync_multiple(slave_dev, bond_dev);
1541                 dev_uc_sync_multiple(slave_dev, bond_dev);
1542
1543                 netif_addr_unlock_bh(bond_dev);
1544         }
1545
1546         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1547                 /* add lacpdu mc addr to mc list */
1548                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1549
1550                 dev_mc_add(slave_dev, lacpdu_multicast);
1551         }
1552
1553         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1554         if (res) {
1555                 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1556                            slave_dev->name);
1557                 goto err_close;
1558         }
1559
1560         prev_slave = bond_last_slave(bond);
1561
1562         new_slave->delay = 0;
1563         new_slave->link_failure_count = 0;
1564
1565         bond_update_speed_duplex(new_slave);
1566
1567         new_slave->last_rx = jiffies -
1568                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1569         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1570                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1571
1572         if (bond->params.miimon && !bond->params.use_carrier) {
1573                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1574
1575                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1576                         /* miimon is set but a bonded network driver
1577                          * does not support ETHTOOL/MII and
1578                          * arp_interval is not set.  Note: if
1579                          * use_carrier is enabled, we will never go
1580                          * here (because netif_carrier is always
1581                          * supported); thus, we don't need to change
1582                          * the messages for netif_carrier.
1583                          */
1584                         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",
1585                                     slave_dev->name);
1586                 } else if (link_reporting == -1) {
1587                         /* unable get link status using mii/ethtool */
1588                         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",
1589                                     slave_dev->name);
1590                 }
1591         }
1592
1593         /* check for initial state */
1594         if (bond->params.miimon) {
1595                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1596                         if (bond->params.updelay) {
1597                                 bond_set_slave_link_state(new_slave,
1598                                                           BOND_LINK_BACK,
1599                                                           BOND_SLAVE_NOTIFY_NOW);
1600                                 new_slave->delay = bond->params.updelay;
1601                         } else {
1602                                 bond_set_slave_link_state(new_slave,
1603                                                           BOND_LINK_UP,
1604                                                           BOND_SLAVE_NOTIFY_NOW);
1605                         }
1606                 } else {
1607                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1608                                                   BOND_SLAVE_NOTIFY_NOW);
1609                 }
1610         } else if (bond->params.arp_interval) {
1611                 bond_set_slave_link_state(new_slave,
1612                                           (netif_carrier_ok(slave_dev) ?
1613                                           BOND_LINK_UP : BOND_LINK_DOWN),
1614                                           BOND_SLAVE_NOTIFY_NOW);
1615         } else {
1616                 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1617                                           BOND_SLAVE_NOTIFY_NOW);
1618         }
1619
1620         if (new_slave->link != BOND_LINK_DOWN)
1621                 new_slave->last_link_up = jiffies;
1622         netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1623                    new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1624                    (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1625
1626         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1627                 /* if there is a primary slave, remember it */
1628                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1629                         rcu_assign_pointer(bond->primary_slave, new_slave);
1630                         bond->force_primary = true;
1631                 }
1632         }
1633
1634         switch (BOND_MODE(bond)) {
1635         case BOND_MODE_ACTIVEBACKUP:
1636                 bond_set_slave_inactive_flags(new_slave,
1637                                               BOND_SLAVE_NOTIFY_NOW);
1638                 break;
1639         case BOND_MODE_8023AD:
1640                 /* in 802.3ad mode, the internal mechanism
1641                  * will activate the slaves in the selected
1642                  * aggregator
1643                  */
1644                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1645                 /* if this is the first slave */
1646                 if (!prev_slave) {
1647                         SLAVE_AD_INFO(new_slave)->id = 1;
1648                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1649                          * can be called only after the mac address of the bond is set
1650                          */
1651                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1652                 } else {
1653                         SLAVE_AD_INFO(new_slave)->id =
1654                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1655                 }
1656
1657                 bond_3ad_bind_slave(new_slave);
1658                 break;
1659         case BOND_MODE_TLB:
1660         case BOND_MODE_ALB:
1661                 bond_set_active_slave(new_slave);
1662                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1663                 break;
1664         default:
1665                 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1666
1667                 /* always active in trunk mode */
1668                 bond_set_active_slave(new_slave);
1669
1670                 /* In trunking mode there is little meaning to curr_active_slave
1671                  * anyway (it holds no special properties of the bond device),
1672                  * so we can change it without calling change_active_interface()
1673                  */
1674                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1675                     new_slave->link == BOND_LINK_UP)
1676                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1677
1678                 break;
1679         } /* switch(bond_mode) */
1680
1681 #ifdef CONFIG_NET_POLL_CONTROLLER
1682         slave_dev->npinfo = bond->dev->npinfo;
1683         if (slave_dev->npinfo) {
1684                 if (slave_enable_netpoll(new_slave)) {
1685                         netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1686                         res = -EBUSY;
1687                         goto err_detach;
1688                 }
1689         }
1690 #endif
1691
1692         if (!(bond_dev->features & NETIF_F_LRO))
1693                 dev_disable_lro(slave_dev);
1694
1695         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1696                                          new_slave);
1697         if (res) {
1698                 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1699                 goto err_detach;
1700         }
1701
1702         res = bond_master_upper_dev_link(bond, new_slave);
1703         if (res) {
1704                 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1705                 goto err_unregister;
1706         }
1707
1708         res = bond_sysfs_slave_add(new_slave);
1709         if (res) {
1710                 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1711                 goto err_upper_unlink;
1712         }
1713
1714         bond->slave_cnt++;
1715         bond_compute_features(bond);
1716         bond_set_carrier(bond);
1717
1718         if (bond_uses_primary(bond)) {
1719                 block_netpoll_tx();
1720                 bond_select_active_slave(bond);
1721                 unblock_netpoll_tx();
1722         }
1723
1724         if (bond_mode_uses_xmit_hash(bond))
1725                 bond_update_slave_arr(bond, NULL);
1726
1727         netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1728                     slave_dev->name,
1729                     bond_is_active_slave(new_slave) ? "an active" : "a backup",
1730                     new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1731
1732         /* enslave is successful */
1733         bond_queue_slave_event(new_slave);
1734         return 0;
1735
1736 /* Undo stages on error */
1737 err_upper_unlink:
1738         bond_upper_dev_unlink(bond, new_slave);
1739
1740 err_unregister:
1741         netdev_rx_handler_unregister(slave_dev);
1742
1743 err_detach:
1744         if (!bond_uses_primary(bond))
1745                 bond_hw_addr_flush(bond_dev, slave_dev);
1746
1747         vlan_vids_del_by_dev(slave_dev, bond_dev);
1748         if (rcu_access_pointer(bond->primary_slave) == new_slave)
1749                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1750         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1751                 block_netpoll_tx();
1752                 bond_change_active_slave(bond, NULL);
1753                 bond_select_active_slave(bond);
1754                 unblock_netpoll_tx();
1755         }
1756         /* either primary_slave or curr_active_slave might've changed */
1757         synchronize_rcu();
1758         slave_disable_netpoll(new_slave);
1759
1760 err_close:
1761         slave_dev->priv_flags &= ~IFF_BONDING;
1762         dev_close(slave_dev);
1763
1764 err_restore_mac:
1765         slave_dev->flags &= ~IFF_SLAVE;
1766         if (!bond->params.fail_over_mac ||
1767             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1768                 /* XXX TODO - fom follow mode needs to change master's
1769                  * MAC if this slave's MAC is in use by the bond, or at
1770                  * least print a warning.
1771                  */
1772                 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1773                 addr.sa_family = slave_dev->type;
1774                 dev_set_mac_address(slave_dev, &addr);
1775         }
1776
1777 err_restore_mtu:
1778         dev_set_mtu(slave_dev, new_slave->original_mtu);
1779
1780 err_free:
1781         bond_free_slave(new_slave);
1782
1783 err_undo_flags:
1784         /* Enslave of first slave has failed and we need to fix master's mac */
1785         if (!bond_has_slaves(bond)) {
1786                 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1787                                             slave_dev->dev_addr))
1788                         eth_hw_addr_random(bond_dev);
1789                 if (bond_dev->type != ARPHRD_ETHER) {
1790                         dev_close(bond_dev);
1791                         ether_setup(bond_dev);
1792                         bond_dev->flags |= IFF_MASTER;
1793                         bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1794                 }
1795         }
1796
1797         return res;
1798 }
1799
1800 /* Try to release the slave device <slave> from the bond device <master>
1801  * It is legal to access curr_active_slave without a lock because all the function
1802  * is RTNL-locked. If "all" is true it means that the function is being called
1803  * while destroying a bond interface and all slaves are being released.
1804  *
1805  * The rules for slave state should be:
1806  *   for Active/Backup:
1807  *     Active stays on all backups go down
1808  *   for Bonded connections:
1809  *     The first up interface should be left on and all others downed.
1810  */
1811 static int __bond_release_one(struct net_device *bond_dev,
1812                               struct net_device *slave_dev,
1813                               bool all)
1814 {
1815         struct bonding *bond = netdev_priv(bond_dev);
1816         struct slave *slave, *oldcurrent;
1817         struct sockaddr addr;
1818         int old_flags = bond_dev->flags;
1819         netdev_features_t old_features = bond_dev->features;
1820
1821         /* slave is not a slave or master is not master of this slave */
1822         if (!(slave_dev->flags & IFF_SLAVE) ||
1823             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1824                 netdev_dbg(bond_dev, "cannot release %s\n",
1825                            slave_dev->name);
1826                 return -EINVAL;
1827         }
1828
1829         block_netpoll_tx();
1830
1831         slave = bond_get_slave_by_dev(bond, slave_dev);
1832         if (!slave) {
1833                 /* not a slave of this bond */
1834                 netdev_info(bond_dev, "%s not enslaved\n",
1835                             slave_dev->name);
1836                 unblock_netpoll_tx();
1837                 return -EINVAL;
1838         }
1839
1840         bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1841
1842         bond_sysfs_slave_del(slave);
1843
1844         /* recompute stats just before removing the slave */
1845         bond_get_stats(bond->dev, &bond->bond_stats);
1846
1847         bond_upper_dev_unlink(bond, slave);
1848         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1849          * for this slave anymore.
1850          */
1851         netdev_rx_handler_unregister(slave_dev);
1852
1853         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1854                 bond_3ad_unbind_slave(slave);
1855
1856         if (bond_mode_uses_xmit_hash(bond))
1857                 bond_update_slave_arr(bond, slave);
1858
1859         netdev_info(bond_dev, "Releasing %s interface %s\n",
1860                     bond_is_active_slave(slave) ? "active" : "backup",
1861                     slave_dev->name);
1862
1863         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1864
1865         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1866
1867         if (!all && (!bond->params.fail_over_mac ||
1868                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1869                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1870                     bond_has_slaves(bond))
1871                         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",
1872                                     slave_dev->name, slave->perm_hwaddr,
1873                                     bond_dev->name, slave_dev->name);
1874         }
1875
1876         if (rtnl_dereference(bond->primary_slave) == slave)
1877                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1878
1879         if (oldcurrent == slave)
1880                 bond_change_active_slave(bond, NULL);
1881
1882         if (bond_is_lb(bond)) {
1883                 /* Must be called only after the slave has been
1884                  * detached from the list and the curr_active_slave
1885                  * has been cleared (if our_slave == old_current),
1886                  * but before a new active slave is selected.
1887                  */
1888                 bond_alb_deinit_slave(bond, slave);
1889         }
1890
1891         if (all) {
1892                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1893         } else if (oldcurrent == slave) {
1894                 /* Note that we hold RTNL over this sequence, so there
1895                  * is no concern that another slave add/remove event
1896                  * will interfere.
1897                  */
1898                 bond_select_active_slave(bond);
1899         }
1900
1901         if (!bond_has_slaves(bond)) {
1902                 bond_set_carrier(bond);
1903                 eth_hw_addr_random(bond_dev);
1904         }
1905
1906         unblock_netpoll_tx();
1907         synchronize_rcu();
1908         bond->slave_cnt--;
1909
1910         if (!bond_has_slaves(bond)) {
1911                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1912                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1913         }
1914
1915         bond_compute_features(bond);
1916         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1917             (old_features & NETIF_F_VLAN_CHALLENGED))
1918                 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1919                             slave_dev->name, bond_dev->name);
1920
1921         vlan_vids_del_by_dev(slave_dev, bond_dev);
1922
1923         /* If the mode uses primary, then this case was handled above by
1924          * bond_change_active_slave(..., NULL)
1925          */
1926         if (!bond_uses_primary(bond)) {
1927                 /* unset promiscuity level from slave
1928                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1929                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1930                  * value of that flag before that change, as that was the value
1931                  * when this slave was attached, so we cache at the start of the
1932                  * function and use it here. Same goes for ALLMULTI below
1933                  */
1934                 if (old_flags & IFF_PROMISC)
1935                         dev_set_promiscuity(slave_dev, -1);
1936
1937                 /* unset allmulti level from slave */
1938                 if (old_flags & IFF_ALLMULTI)
1939                         dev_set_allmulti(slave_dev, -1);
1940
1941                 bond_hw_addr_flush(bond_dev, slave_dev);
1942         }
1943
1944         slave_disable_netpoll(slave);
1945
1946         /* close slave before restoring its mac address */
1947         dev_close(slave_dev);
1948
1949         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1950             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1951                 /* restore original ("permanent") mac address */
1952                 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1953                 addr.sa_family = slave_dev->type;
1954                 dev_set_mac_address(slave_dev, &addr);
1955         }
1956
1957         dev_set_mtu(slave_dev, slave->original_mtu);
1958
1959         slave_dev->priv_flags &= ~IFF_BONDING;
1960
1961         bond_free_slave(slave);
1962
1963         return 0;
1964 }
1965
1966 /* A wrapper used because of ndo_del_link */
1967 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1968 {
1969         return __bond_release_one(bond_dev, slave_dev, false);
1970 }
1971
1972 /* First release a slave and then destroy the bond if no more slaves are left.
1973  * Must be under rtnl_lock when this function is called.
1974  */
1975 static int  bond_release_and_destroy(struct net_device *bond_dev,
1976                                      struct net_device *slave_dev)
1977 {
1978         struct bonding *bond = netdev_priv(bond_dev);
1979         int ret;
1980
1981         ret = bond_release(bond_dev, slave_dev);
1982         if (ret == 0 && !bond_has_slaves(bond)) {
1983                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1984                 netdev_info(bond_dev, "Destroying bond %s\n",
1985                             bond_dev->name);
1986                 bond_remove_proc_entry(bond);
1987                 unregister_netdevice(bond_dev);
1988         }
1989         return ret;
1990 }
1991
1992 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1993 {
1994         struct bonding *bond = netdev_priv(bond_dev);
1995         bond_fill_ifbond(bond, info);
1996         return 0;
1997 }
1998
1999 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2000 {
2001         struct bonding *bond = netdev_priv(bond_dev);
2002         struct list_head *iter;
2003         int i = 0, res = -ENODEV;
2004         struct slave *slave;
2005
2006         bond_for_each_slave(bond, slave, iter) {
2007                 if (i++ == (int)info->slave_id) {
2008                         res = 0;
2009                         bond_fill_ifslave(slave, info);
2010                         break;
2011                 }
2012         }
2013
2014         return res;
2015 }
2016
2017 /*-------------------------------- Monitoring -------------------------------*/
2018
2019 /* called with rcu_read_lock() */
2020 static int bond_miimon_inspect(struct bonding *bond)
2021 {
2022         int link_state, commit = 0;
2023         struct list_head *iter;
2024         struct slave *slave;
2025         bool ignore_updelay;
2026
2027         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2028
2029         bond_for_each_slave_rcu(bond, slave, iter) {
2030                 slave->new_link = BOND_LINK_NOCHANGE;
2031
2032                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2033
2034                 switch (slave->link) {
2035                 case BOND_LINK_UP:
2036                         if (link_state)
2037                                 continue;
2038
2039                         bond_set_slave_link_state(slave, BOND_LINK_FAIL,
2040                                                   BOND_SLAVE_NOTIFY_LATER);
2041                         slave->delay = bond->params.downdelay;
2042                         if (slave->delay) {
2043                                 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2044                                             (BOND_MODE(bond) ==
2045                                              BOND_MODE_ACTIVEBACKUP) ?
2046                                              (bond_is_active_slave(slave) ?
2047                                               "active " : "backup ") : "",
2048                                             slave->dev->name,
2049                                             bond->params.downdelay * bond->params.miimon);
2050                         }
2051                         /*FALLTHRU*/
2052                 case BOND_LINK_FAIL:
2053                         if (link_state) {
2054                                 /* recovered before downdelay expired */
2055                                 bond_set_slave_link_state(slave, BOND_LINK_UP,
2056                                                           BOND_SLAVE_NOTIFY_LATER);
2057                                 slave->last_link_up = jiffies;
2058                                 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2059                                             (bond->params.downdelay - slave->delay) *
2060                                             bond->params.miimon,
2061                                             slave->dev->name);
2062                                 continue;
2063                         }
2064
2065                         if (slave->delay <= 0) {
2066                                 slave->new_link = BOND_LINK_DOWN;
2067                                 commit++;
2068                                 continue;
2069                         }
2070
2071                         slave->delay--;
2072                         break;
2073
2074                 case BOND_LINK_DOWN:
2075                         if (!link_state)
2076                                 continue;
2077
2078                         bond_set_slave_link_state(slave, BOND_LINK_BACK,
2079                                                   BOND_SLAVE_NOTIFY_LATER);
2080                         slave->delay = bond->params.updelay;
2081
2082                         if (slave->delay) {
2083                                 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2084                                             slave->dev->name,
2085                                             ignore_updelay ? 0 :
2086                                             bond->params.updelay *
2087                                             bond->params.miimon);
2088                         }
2089                         /*FALLTHRU*/
2090                 case BOND_LINK_BACK:
2091                         if (!link_state) {
2092                                 bond_set_slave_link_state(slave,
2093                                                           BOND_LINK_DOWN,
2094                                                           BOND_SLAVE_NOTIFY_LATER);
2095                                 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2096                                             (bond->params.updelay - slave->delay) *
2097                                             bond->params.miimon,
2098                                             slave->dev->name);
2099
2100                                 continue;
2101                         }
2102
2103                         if (ignore_updelay)
2104                                 slave->delay = 0;
2105
2106                         if (slave->delay <= 0) {
2107                                 slave->new_link = BOND_LINK_UP;
2108                                 commit++;
2109                                 ignore_updelay = false;
2110                                 continue;
2111                         }
2112
2113                         slave->delay--;
2114                         break;
2115                 }
2116         }
2117
2118         return commit;
2119 }
2120
2121 static void bond_miimon_commit(struct bonding *bond)
2122 {
2123         struct list_head *iter;
2124         struct slave *slave, *primary;
2125
2126         bond_for_each_slave(bond, slave, iter) {
2127                 switch (slave->new_link) {
2128                 case BOND_LINK_NOCHANGE:
2129                         continue;
2130
2131                 case BOND_LINK_UP:
2132                         bond_update_speed_duplex(slave);
2133                         bond_set_slave_link_state(slave, BOND_LINK_UP,
2134                                                   BOND_SLAVE_NOTIFY_NOW);
2135                         slave->last_link_up = jiffies;
2136
2137                         primary = rtnl_dereference(bond->primary_slave);
2138                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2139                                 /* prevent it from being the active one */
2140                                 bond_set_backup_slave(slave);
2141                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2142                                 /* make it immediately active */
2143                                 bond_set_active_slave(slave);
2144                         } else if (slave != primary) {
2145                                 /* prevent it from being the active one */
2146                                 bond_set_backup_slave(slave);
2147                         }
2148
2149                         netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2150                                     slave->dev->name,
2151                                     slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2152                                     slave->duplex ? "full" : "half");
2153
2154                         /* notify ad that the link status has changed */
2155                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2156                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2157
2158                         if (bond_is_lb(bond))
2159                                 bond_alb_handle_link_change(bond, slave,
2160                                                             BOND_LINK_UP);
2161
2162                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2163                                 bond_update_slave_arr(bond, NULL);
2164
2165                         if (!bond->curr_active_slave || slave == primary)
2166                                 goto do_failover;
2167
2168                         continue;
2169
2170                 case BOND_LINK_DOWN:
2171                         if (slave->link_failure_count < UINT_MAX)
2172                                 slave->link_failure_count++;
2173
2174                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2175                                                   BOND_SLAVE_NOTIFY_NOW);
2176
2177                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2178                             BOND_MODE(bond) == BOND_MODE_8023AD)
2179                                 bond_set_slave_inactive_flags(slave,
2180                                                               BOND_SLAVE_NOTIFY_NOW);
2181
2182                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2183                                     slave->dev->name);
2184
2185                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2186                                 bond_3ad_handle_link_change(slave,
2187                                                             BOND_LINK_DOWN);
2188
2189                         if (bond_is_lb(bond))
2190                                 bond_alb_handle_link_change(bond, slave,
2191                                                             BOND_LINK_DOWN);
2192
2193                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2194                                 bond_update_slave_arr(bond, NULL);
2195
2196                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2197                                 goto do_failover;
2198
2199                         continue;
2200
2201                 default:
2202                         netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2203                                    slave->new_link, slave->dev->name);
2204                         slave->new_link = BOND_LINK_NOCHANGE;
2205
2206                         continue;
2207                 }
2208
2209 do_failover:
2210                 block_netpoll_tx();
2211                 bond_select_active_slave(bond);
2212                 unblock_netpoll_tx();
2213         }
2214
2215         bond_set_carrier(bond);
2216 }
2217
2218 /* bond_mii_monitor
2219  *
2220  * Really a wrapper that splits the mii monitor into two phases: an
2221  * inspection, then (if inspection indicates something needs to be done)
2222  * an acquisition of appropriate locks followed by a commit phase to
2223  * implement whatever link state changes are indicated.
2224  */
2225 static void bond_mii_monitor(struct work_struct *work)
2226 {
2227         struct bonding *bond = container_of(work, struct bonding,
2228                                             mii_work.work);
2229         bool should_notify_peers = false;
2230         unsigned long delay;
2231
2232         delay = msecs_to_jiffies(bond->params.miimon);
2233
2234         if (!bond_has_slaves(bond))
2235                 goto re_arm;
2236
2237         rcu_read_lock();
2238
2239         should_notify_peers = bond_should_notify_peers(bond);
2240
2241         if (bond_miimon_inspect(bond)) {
2242                 rcu_read_unlock();
2243
2244                 /* Race avoidance with bond_close cancel of workqueue */
2245                 if (!rtnl_trylock()) {
2246                         delay = 1;
2247                         should_notify_peers = false;
2248                         goto re_arm;
2249                 }
2250
2251                 bond_miimon_commit(bond);
2252
2253                 rtnl_unlock();  /* might sleep, hold no other locks */
2254         } else
2255                 rcu_read_unlock();
2256
2257 re_arm:
2258         if (bond->params.miimon)
2259                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2260
2261         if (should_notify_peers) {
2262                 if (!rtnl_trylock())
2263                         return;
2264                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2265                 rtnl_unlock();
2266         }
2267 }
2268
2269 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2270 {
2271         struct net_device *upper;
2272         struct list_head *iter;
2273         bool ret = false;
2274
2275         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2276                 return true;
2277
2278         rcu_read_lock();
2279         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2280                 if (ip == bond_confirm_addr(upper, 0, ip)) {
2281                         ret = true;
2282                         break;
2283                 }
2284         }
2285         rcu_read_unlock();
2286
2287         return ret;
2288 }
2289
2290 /* We go to the (large) trouble of VLAN tagging ARP frames because
2291  * switches in VLAN mode (especially if ports are configured as
2292  * "native" to a VLAN) might not pass non-tagged frames.
2293  */
2294 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2295                           __be32 dest_ip, __be32 src_ip,
2296                           struct bond_vlan_tag *tags)
2297 {
2298         struct sk_buff *skb;
2299         struct bond_vlan_tag *outer_tag = tags;
2300
2301         netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2302                    arp_op, slave_dev->name, &dest_ip, &src_ip);
2303
2304         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2305                          NULL, slave_dev->dev_addr, NULL);
2306
2307         if (!skb) {
2308                 net_err_ratelimited("ARP packet allocation failed\n");
2309                 return;
2310         }
2311
2312         if (!tags || tags->vlan_proto == VLAN_N_VID)
2313                 goto xmit;
2314
2315         tags++;
2316
2317         /* Go through all the tags backwards and add them to the packet */
2318         while (tags->vlan_proto != VLAN_N_VID) {
2319                 if (!tags->vlan_id) {
2320                         tags++;
2321                         continue;
2322                 }
2323
2324                 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2325                            ntohs(outer_tag->vlan_proto), tags->vlan_id);
2326                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2327                                                 tags->vlan_id);
2328                 if (!skb) {
2329                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2330                         return;
2331                 }
2332
2333                 tags++;
2334         }
2335         /* Set the outer tag */
2336         if (outer_tag->vlan_id) {
2337                 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2338                            ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2339                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2340                                        outer_tag->vlan_id);
2341         }
2342
2343 xmit:
2344         arp_xmit(skb);
2345 }
2346
2347 /* Validate the device path between the @start_dev and the @end_dev.
2348  * The path is valid if the @end_dev is reachable through device
2349  * stacking.
2350  * When the path is validated, collect any vlan information in the
2351  * path.
2352  */
2353 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2354                                               struct net_device *end_dev,
2355                                               int level)
2356 {
2357         struct bond_vlan_tag *tags;
2358         struct net_device *upper;
2359         struct list_head  *iter;
2360
2361         if (start_dev == end_dev) {
2362                 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2363                 if (!tags)
2364                         return ERR_PTR(-ENOMEM);
2365                 tags[level].vlan_proto = VLAN_N_VID;
2366                 return tags;
2367         }
2368
2369         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2370                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2371                 if (IS_ERR_OR_NULL(tags)) {
2372                         if (IS_ERR(tags))
2373                                 return tags;
2374                         continue;
2375                 }
2376                 if (is_vlan_dev(upper)) {
2377                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2378                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2379                 }
2380
2381                 return tags;
2382         }
2383
2384         return NULL;
2385 }
2386
2387 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2388 {
2389         struct rtable *rt;
2390         struct bond_vlan_tag *tags;
2391         __be32 *targets = bond->params.arp_targets, addr;
2392         int i;
2393
2394         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2395                 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2396                 tags = NULL;
2397
2398                 /* Find out through which dev should the packet go */
2399                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2400                                      RTO_ONLINK, 0);
2401                 if (IS_ERR(rt)) {
2402                         /* there's no route to target - try to send arp
2403                          * probe to generate any traffic (arp_validate=0)
2404                          */
2405                         if (bond->params.arp_validate)
2406                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2407                                                      bond->dev->name,
2408                                                      &targets[i]);
2409                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2410                                       0, tags);
2411                         continue;
2412                 }
2413
2414                 /* bond device itself */
2415                 if (rt->dst.dev == bond->dev)
2416                         goto found;
2417
2418                 rcu_read_lock();
2419                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2420                 rcu_read_unlock();
2421
2422                 if (!IS_ERR_OR_NULL(tags))
2423                         goto found;
2424
2425                 /* Not our device - skip */
2426                 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2427                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2428
2429                 ip_rt_put(rt);
2430                 continue;
2431
2432 found:
2433                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2434                 ip_rt_put(rt);
2435                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2436                               addr, tags);
2437                 kfree(tags);
2438         }
2439 }
2440
2441 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2442 {
2443         int i;
2444
2445         if (!sip || !bond_has_this_ip(bond, tip)) {
2446                 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2447                            &sip, &tip);
2448                 return;
2449         }
2450
2451         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2452         if (i == -1) {
2453                 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2454                            &sip);
2455                 return;
2456         }
2457         slave->last_rx = jiffies;
2458         slave->target_last_arp_rx[i] = jiffies;
2459 }
2460
2461 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2462                  struct slave *slave)
2463 {
2464         struct arphdr *arp = (struct arphdr *)skb->data;
2465         struct slave *curr_active_slave, *curr_arp_slave;
2466         unsigned char *arp_ptr;
2467         __be32 sip, tip;
2468         int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2469
2470         if (!slave_do_arp_validate(bond, slave)) {
2471                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2472                     !slave_do_arp_validate_only(bond))
2473                         slave->last_rx = jiffies;
2474                 return RX_HANDLER_ANOTHER;
2475         } else if (!is_arp) {
2476                 return RX_HANDLER_ANOTHER;
2477         }
2478
2479         alen = arp_hdr_len(bond->dev);
2480
2481         netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2482                    skb->dev->name);
2483
2484         if (alen > skb_headlen(skb)) {
2485                 arp = kmalloc(alen, GFP_ATOMIC);
2486                 if (!arp)
2487                         goto out_unlock;
2488                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2489                         goto out_unlock;
2490         }
2491
2492         if (arp->ar_hln != bond->dev->addr_len ||
2493             skb->pkt_type == PACKET_OTHERHOST ||
2494             skb->pkt_type == PACKET_LOOPBACK ||
2495             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2496             arp->ar_pro != htons(ETH_P_IP) ||
2497             arp->ar_pln != 4)
2498                 goto out_unlock;
2499
2500         arp_ptr = (unsigned char *)(arp + 1);
2501         arp_ptr += bond->dev->addr_len;
2502         memcpy(&sip, arp_ptr, 4);
2503         arp_ptr += 4 + bond->dev->addr_len;
2504         memcpy(&tip, arp_ptr, 4);
2505
2506         netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2507                    slave->dev->name, bond_slave_state(slave),
2508                      bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2509                      &sip, &tip);
2510
2511         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2512         curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2513
2514         /* We 'trust' the received ARP enough to validate it if:
2515          *
2516          * (a) the slave receiving the ARP is active (which includes the
2517          * current ARP slave, if any), or
2518          *
2519          * (b) the receiving slave isn't active, but there is a currently
2520          * active slave and it received valid arp reply(s) after it became
2521          * the currently active slave, or
2522          *
2523          * (c) there is an ARP slave that sent an ARP during the prior ARP
2524          * interval, and we receive an ARP reply on any slave.  We accept
2525          * these because switch FDB update delays may deliver the ARP
2526          * reply to a slave other than the sender of the ARP request.
2527          *
2528          * Note: for (b), backup slaves are receiving the broadcast ARP
2529          * request, not a reply.  This request passes from the sending
2530          * slave through the L2 switch(es) to the receiving slave.  Since
2531          * this is checking the request, sip/tip are swapped for
2532          * validation.
2533          *
2534          * This is done to avoid endless looping when we can't reach the
2535          * arp_ip_target and fool ourselves with our own arp requests.
2536          */
2537         if (bond_is_active_slave(slave))
2538                 bond_validate_arp(bond, slave, sip, tip);
2539         else if (curr_active_slave &&
2540                  time_after(slave_last_rx(bond, curr_active_slave),
2541                             curr_active_slave->last_link_up))
2542                 bond_validate_arp(bond, slave, tip, sip);
2543         else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2544                  bond_time_in_interval(bond,
2545                                        dev_trans_start(curr_arp_slave->dev), 1))
2546                 bond_validate_arp(bond, slave, sip, tip);
2547
2548 out_unlock:
2549         if (arp != (struct arphdr *)skb->data)
2550                 kfree(arp);
2551         return RX_HANDLER_ANOTHER;
2552 }
2553
2554 /* function to verify if we're in the arp_interval timeslice, returns true if
2555  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2556  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2557  */
2558 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2559                                   int mod)
2560 {
2561         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2562
2563         return time_in_range(jiffies,
2564                              last_act - delta_in_ticks,
2565                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2566 }
2567
2568 /* This function is called regularly to monitor each slave's link
2569  * ensuring that traffic is being sent and received when arp monitoring
2570  * is used in load-balancing mode. if the adapter has been dormant, then an
2571  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2572  * arp monitoring in active backup mode.
2573  */
2574 static void bond_loadbalance_arp_mon(struct work_struct *work)
2575 {
2576         struct bonding *bond = container_of(work, struct bonding,
2577                                             arp_work.work);
2578         struct slave *slave, *oldcurrent;
2579         struct list_head *iter;
2580         int do_failover = 0, slave_state_changed = 0;
2581
2582         if (!bond_has_slaves(bond))
2583                 goto re_arm;
2584
2585         rcu_read_lock();
2586
2587         oldcurrent = rcu_dereference(bond->curr_active_slave);
2588         /* see if any of the previous devices are up now (i.e. they have
2589          * xmt and rcv traffic). the curr_active_slave does not come into
2590          * the picture unless it is null. also, slave->last_link_up is not
2591          * needed here because we send an arp on each slave and give a slave
2592          * as long as it needs to get the tx/rx within the delta.
2593          * TODO: what about up/down delay in arp mode? it wasn't here before
2594          *       so it can wait
2595          */
2596         bond_for_each_slave_rcu(bond, slave, iter) {
2597                 unsigned long trans_start = dev_trans_start(slave->dev);
2598
2599                 if (slave->link != BOND_LINK_UP) {
2600                         if (bond_time_in_interval(bond, trans_start, 1) &&
2601                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2602
2603                                 slave->link  = BOND_LINK_UP;
2604                                 slave_state_changed = 1;
2605
2606                                 /* primary_slave has no meaning in round-robin
2607                                  * mode. the window of a slave being up and
2608                                  * curr_active_slave being null after enslaving
2609                                  * is closed.
2610                                  */
2611                                 if (!oldcurrent) {
2612                                         netdev_info(bond->dev, "link status definitely up for interface %s\n",
2613                                                     slave->dev->name);
2614                                         do_failover = 1;
2615                                 } else {
2616                                         netdev_info(bond->dev, "interface %s is now up\n",
2617                                                     slave->dev->name);
2618                                 }
2619                         }
2620                 } else {
2621                         /* slave->link == BOND_LINK_UP */
2622
2623                         /* not all switches will respond to an arp request
2624                          * when the source ip is 0, so don't take the link down
2625                          * if we don't know our ip yet
2626                          */
2627                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2628                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2629
2630                                 slave->link  = BOND_LINK_DOWN;
2631                                 slave_state_changed = 1;
2632
2633                                 if (slave->link_failure_count < UINT_MAX)
2634                                         slave->link_failure_count++;
2635
2636                                 netdev_info(bond->dev, "interface %s is now down\n",
2637                                             slave->dev->name);
2638
2639                                 if (slave == oldcurrent)
2640                                         do_failover = 1;
2641                         }
2642                 }
2643
2644                 /* note: if switch is in round-robin mode, all links
2645                  * must tx arp to ensure all links rx an arp - otherwise
2646                  * links may oscillate or not come up at all; if switch is
2647                  * in something like xor mode, there is nothing we can
2648                  * do - all replies will be rx'ed on same link causing slaves
2649                  * to be unstable during low/no traffic periods
2650                  */
2651                 if (bond_slave_is_up(slave))
2652                         bond_arp_send_all(bond, slave);
2653         }
2654
2655         rcu_read_unlock();
2656
2657         if (do_failover || slave_state_changed) {
2658                 if (!rtnl_trylock())
2659                         goto re_arm;
2660
2661                 if (slave_state_changed) {
2662                         bond_slave_state_change(bond);
2663                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2664                                 bond_update_slave_arr(bond, NULL);
2665                 }
2666                 if (do_failover) {
2667                         block_netpoll_tx();
2668                         bond_select_active_slave(bond);
2669                         unblock_netpoll_tx();
2670                 }
2671                 rtnl_unlock();
2672         }
2673
2674 re_arm:
2675         if (bond->params.arp_interval)
2676                 queue_delayed_work(bond->wq, &bond->arp_work,
2677                                    msecs_to_jiffies(bond->params.arp_interval));
2678 }
2679
2680 /* Called to inspect slaves for active-backup mode ARP monitor link state
2681  * changes.  Sets new_link in slaves to specify what action should take
2682  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2683  * to link states must be committed.
2684  *
2685  * Called with rcu_read_lock held.
2686  */
2687 static int bond_ab_arp_inspect(struct bonding *bond)
2688 {
2689         unsigned long trans_start, last_rx;
2690         struct list_head *iter;
2691         struct slave *slave;
2692         int commit = 0;
2693
2694         bond_for_each_slave_rcu(bond, slave, iter) {
2695                 slave->new_link = BOND_LINK_NOCHANGE;
2696                 last_rx = slave_last_rx(bond, slave);
2697
2698                 if (slave->link != BOND_LINK_UP) {
2699                         if (bond_time_in_interval(bond, last_rx, 1)) {
2700                                 slave->new_link = BOND_LINK_UP;
2701                                 commit++;
2702                         }
2703                         continue;
2704                 }
2705
2706                 /* Give slaves 2*delta after being enslaved or made
2707                  * active.  This avoids bouncing, as the last receive
2708                  * times need a full ARP monitor cycle to be updated.
2709                  */
2710                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2711                         continue;
2712
2713                 /* Backup slave is down if:
2714                  * - No current_arp_slave AND
2715                  * - more than 3*delta since last receive AND
2716                  * - the bond has an IP address
2717                  *
2718                  * Note: a non-null current_arp_slave indicates
2719                  * the curr_active_slave went down and we are
2720                  * searching for a new one; under this condition
2721                  * we only take the curr_active_slave down - this
2722                  * gives each slave a chance to tx/rx traffic
2723                  * before being taken out
2724                  */
2725                 if (!bond_is_active_slave(slave) &&
2726                     !rcu_access_pointer(bond->current_arp_slave) &&
2727                     !bond_time_in_interval(bond, last_rx, 3)) {
2728                         slave->new_link = BOND_LINK_DOWN;
2729                         commit++;
2730                 }
2731
2732                 /* Active slave is down if:
2733                  * - more than 2*delta since transmitting OR
2734                  * - (more than 2*delta since receive AND
2735                  *    the bond has an IP address)
2736                  */
2737                 trans_start = dev_trans_start(slave->dev);
2738                 if (bond_is_active_slave(slave) &&
2739                     (!bond_time_in_interval(bond, trans_start, 2) ||
2740                      !bond_time_in_interval(bond, last_rx, 2))) {
2741                         slave->new_link = BOND_LINK_DOWN;
2742                         commit++;
2743                 }
2744         }
2745
2746         return commit;
2747 }
2748
2749 /* Called to commit link state changes noted by inspection step of
2750  * active-backup mode ARP monitor.
2751  *
2752  * Called with RTNL hold.
2753  */
2754 static void bond_ab_arp_commit(struct bonding *bond)
2755 {
2756         unsigned long trans_start;
2757         struct list_head *iter;
2758         struct slave *slave;
2759
2760         bond_for_each_slave(bond, slave, iter) {
2761                 switch (slave->new_link) {
2762                 case BOND_LINK_NOCHANGE:
2763                         continue;
2764
2765                 case BOND_LINK_UP:
2766                         trans_start = dev_trans_start(slave->dev);
2767                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
2768                             (!rtnl_dereference(bond->curr_active_slave) &&
2769                              bond_time_in_interval(bond, trans_start, 1))) {
2770                                 struct slave *current_arp_slave;
2771
2772                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2773                                 bond_set_slave_link_state(slave, BOND_LINK_UP,
2774                                                           BOND_SLAVE_NOTIFY_NOW);
2775                                 if (current_arp_slave) {
2776                                         bond_set_slave_inactive_flags(
2777                                                 current_arp_slave,
2778                                                 BOND_SLAVE_NOTIFY_NOW);
2779                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2780                                 }
2781
2782                                 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2783                                             slave->dev->name);
2784
2785                                 if (!rtnl_dereference(bond->curr_active_slave) ||
2786                                     slave == rtnl_dereference(bond->primary_slave))
2787                                         goto do_failover;
2788
2789                         }
2790
2791                         continue;
2792
2793                 case BOND_LINK_DOWN:
2794                         if (slave->link_failure_count < UINT_MAX)
2795                                 slave->link_failure_count++;
2796
2797                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2798                                                   BOND_SLAVE_NOTIFY_NOW);
2799                         bond_set_slave_inactive_flags(slave,
2800                                                       BOND_SLAVE_NOTIFY_NOW);
2801
2802                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2803                                     slave->dev->name);
2804
2805                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
2806                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2807                                 goto do_failover;
2808                         }
2809
2810                         continue;
2811
2812                 default:
2813                         netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2814                                    slave->new_link, slave->dev->name);
2815                         continue;
2816                 }
2817
2818 do_failover:
2819                 block_netpoll_tx();
2820                 bond_select_active_slave(bond);
2821                 unblock_netpoll_tx();
2822         }
2823
2824         bond_set_carrier(bond);
2825 }
2826
2827 /* Send ARP probes for active-backup mode ARP monitor.
2828  *
2829  * Called with rcu_read_lock held.
2830  */
2831 static bool bond_ab_arp_probe(struct bonding *bond)
2832 {
2833         struct slave *slave, *before = NULL, *new_slave = NULL,
2834                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2835                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2836         struct list_head *iter;
2837         bool found = false;
2838         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2839
2840         if (curr_arp_slave && curr_active_slave)
2841                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2842                             curr_arp_slave->dev->name,
2843                             curr_active_slave->dev->name);
2844
2845         if (curr_active_slave) {
2846                 bond_arp_send_all(bond, curr_active_slave);
2847                 return should_notify_rtnl;
2848         }
2849
2850         /* if we don't have a curr_active_slave, search for the next available
2851          * backup slave from the current_arp_slave and make it the candidate
2852          * for becoming the curr_active_slave
2853          */
2854
2855         if (!curr_arp_slave) {
2856                 curr_arp_slave = bond_first_slave_rcu(bond);
2857                 if (!curr_arp_slave)
2858                         return should_notify_rtnl;
2859         }
2860
2861         bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2862
2863         bond_for_each_slave_rcu(bond, slave, iter) {
2864                 if (!found && !before && bond_slave_is_up(slave))
2865                         before = slave;
2866
2867                 if (found && !new_slave && bond_slave_is_up(slave))
2868                         new_slave = slave;
2869                 /* if the link state is up at this point, we
2870                  * mark it down - this can happen if we have
2871                  * simultaneous link failures and
2872                  * reselect_active_interface doesn't make this
2873                  * one the current slave so it is still marked
2874                  * up when it is actually down
2875                  */
2876                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2877                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2878                                                   BOND_SLAVE_NOTIFY_LATER);
2879                         if (slave->link_failure_count < UINT_MAX)
2880                                 slave->link_failure_count++;
2881
2882                         bond_set_slave_inactive_flags(slave,
2883                                                       BOND_SLAVE_NOTIFY_LATER);
2884
2885                         netdev_info(bond->dev, "backup interface %s is now down\n",
2886                                     slave->dev->name);
2887                 }
2888                 if (slave == curr_arp_slave)
2889                         found = true;
2890         }
2891
2892         if (!new_slave && before)
2893                 new_slave = before;
2894
2895         if (!new_slave)
2896                 goto check_state;
2897
2898         bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2899                                   BOND_SLAVE_NOTIFY_LATER);
2900         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2901         bond_arp_send_all(bond, new_slave);
2902         new_slave->last_link_up = jiffies;
2903         rcu_assign_pointer(bond->current_arp_slave, new_slave);
2904
2905 check_state:
2906         bond_for_each_slave_rcu(bond, slave, iter) {
2907                 if (slave->should_notify || slave->should_notify_link) {
2908                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2909                         break;
2910                 }
2911         }
2912         return should_notify_rtnl;
2913 }
2914
2915 static void bond_activebackup_arp_mon(struct work_struct *work)
2916 {
2917         struct bonding *bond = container_of(work, struct bonding,
2918                                             arp_work.work);
2919         bool should_notify_peers = false;
2920         bool should_notify_rtnl = false;
2921         int delta_in_ticks;
2922
2923         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2924
2925         if (!bond_has_slaves(bond))
2926                 goto re_arm;
2927
2928         rcu_read_lock();
2929
2930         should_notify_peers = bond_should_notify_peers(bond);
2931
2932         if (bond_ab_arp_inspect(bond)) {
2933                 rcu_read_unlock();
2934
2935                 /* Race avoidance with bond_close flush of workqueue */
2936                 if (!rtnl_trylock()) {
2937                         delta_in_ticks = 1;
2938                         should_notify_peers = false;
2939                         goto re_arm;
2940                 }
2941
2942                 bond_ab_arp_commit(bond);
2943
2944                 rtnl_unlock();
2945                 rcu_read_lock();
2946         }
2947
2948         should_notify_rtnl = bond_ab_arp_probe(bond);
2949         rcu_read_unlock();
2950
2951 re_arm:
2952         if (bond->params.arp_interval)
2953                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2954
2955         if (should_notify_peers || should_notify_rtnl) {
2956                 if (!rtnl_trylock())
2957                         return;
2958
2959                 if (should_notify_peers)
2960                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2961                                                  bond->dev);
2962                 if (should_notify_rtnl) {
2963                         bond_slave_state_notify(bond);
2964                         bond_slave_link_notify(bond);
2965                 }
2966
2967                 rtnl_unlock();
2968         }
2969 }
2970
2971 /*-------------------------- netdev event handling --------------------------*/
2972
2973 /* Change device name */
2974 static int bond_event_changename(struct bonding *bond)
2975 {
2976         bond_remove_proc_entry(bond);
2977         bond_create_proc_entry(bond);
2978
2979         bond_debug_reregister(bond);
2980
2981         return NOTIFY_DONE;
2982 }
2983
2984 static int bond_master_netdev_event(unsigned long event,
2985                                     struct net_device *bond_dev)
2986 {
2987         struct bonding *event_bond = netdev_priv(bond_dev);
2988
2989         switch (event) {
2990         case NETDEV_CHANGENAME:
2991                 return bond_event_changename(event_bond);
2992         case NETDEV_UNREGISTER:
2993                 bond_remove_proc_entry(event_bond);
2994                 break;
2995         case NETDEV_REGISTER:
2996                 bond_create_proc_entry(event_bond);
2997                 break;
2998         case NETDEV_NOTIFY_PEERS:
2999                 if (event_bond->send_peer_notif)
3000                         event_bond->send_peer_notif--;
3001                 break;
3002         default:
3003                 break;
3004         }
3005
3006         return NOTIFY_DONE;
3007 }
3008
3009 static int bond_slave_netdev_event(unsigned long event,
3010                                    struct net_device *slave_dev)
3011 {
3012         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3013         struct bonding *bond;
3014         struct net_device *bond_dev;
3015
3016         /* A netdev event can be generated while enslaving a device
3017          * before netdev_rx_handler_register is called in which case
3018          * slave will be NULL
3019          */
3020         if (!slave)
3021                 return NOTIFY_DONE;
3022         bond_dev = slave->bond->dev;
3023         bond = slave->bond;
3024         primary = rtnl_dereference(bond->primary_slave);
3025
3026         switch (event) {
3027         case NETDEV_UNREGISTER:
3028                 if (bond_dev->type != ARPHRD_ETHER)
3029                         bond_release_and_destroy(bond_dev, slave_dev);
3030                 else
3031                         bond_release(bond_dev, slave_dev);
3032                 break;
3033         case NETDEV_UP:
3034         case NETDEV_CHANGE:
3035                 bond_update_speed_duplex(slave);
3036                 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3037                         bond_3ad_adapter_speed_duplex_changed(slave);
3038                 /* Fallthrough */
3039         case NETDEV_DOWN:
3040                 /* Refresh slave-array if applicable!
3041                  * If the setup does not use miimon or arpmon (mode-specific!),
3042                  * then these events will not cause the slave-array to be
3043                  * refreshed. This will cause xmit to use a slave that is not
3044                  * usable. Avoid such situation by refeshing the array at these
3045                  * events. If these (miimon/arpmon) parameters are configured
3046                  * then array gets refreshed twice and that should be fine!
3047                  */
3048                 if (bond_mode_uses_xmit_hash(bond))
3049                         bond_update_slave_arr(bond, NULL);
3050                 break;
3051         case NETDEV_CHANGEMTU:
3052                 /* TODO: Should slaves be allowed to
3053                  * independently alter their MTU?  For
3054                  * an active-backup bond, slaves need
3055                  * not be the same type of device, so
3056                  * MTUs may vary.  For other modes,
3057                  * slaves arguably should have the
3058                  * same MTUs. To do this, we'd need to
3059                  * take over the slave's change_mtu
3060                  * function for the duration of their
3061                  * servitude.
3062                  */
3063                 break;
3064         case NETDEV_CHANGENAME:
3065                 /* we don't care if we don't have primary set */
3066                 if (!bond_uses_primary(bond) ||
3067                     !bond->params.primary[0])
3068                         break;
3069
3070                 if (slave == primary) {
3071                         /* slave's name changed - he's no longer primary */
3072                         RCU_INIT_POINTER(bond->primary_slave, NULL);
3073                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3074                         /* we have a new primary slave */
3075                         rcu_assign_pointer(bond->primary_slave, slave);
3076                 } else { /* we didn't change primary - exit */
3077                         break;
3078                 }
3079
3080                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3081                             primary ? slave_dev->name : "none");
3082
3083                 block_netpoll_tx();
3084                 bond_select_active_slave(bond);
3085                 unblock_netpoll_tx();
3086                 break;
3087         case NETDEV_FEAT_CHANGE:
3088                 bond_compute_features(bond);
3089                 break;
3090         case NETDEV_RESEND_IGMP:
3091                 /* Propagate to master device */
3092                 call_netdevice_notifiers(event, slave->bond->dev);
3093                 break;
3094         default:
3095                 break;
3096         }
3097
3098         return NOTIFY_DONE;
3099 }
3100
3101 /* bond_netdev_event: handle netdev notifier chain events.
3102  *
3103  * This function receives events for the netdev chain.  The caller (an
3104  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3105  * locks for us to safely manipulate the slave devices (RTNL lock,
3106  * dev_probe_lock).
3107  */
3108 static int bond_netdev_event(struct notifier_block *this,
3109                              unsigned long event, void *ptr)
3110 {
3111         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3112
3113         netdev_dbg(event_dev, "event: %lx\n", event);
3114
3115         if (!(event_dev->priv_flags & IFF_BONDING))
3116                 return NOTIFY_DONE;
3117
3118         if (event_dev->flags & IFF_MASTER) {
3119                 netdev_dbg(event_dev, "IFF_MASTER\n");
3120                 return bond_master_netdev_event(event, event_dev);
3121         }
3122
3123         if (event_dev->flags & IFF_SLAVE) {
3124                 netdev_dbg(event_dev, "IFF_SLAVE\n");
3125                 return bond_slave_netdev_event(event, event_dev);
3126         }
3127
3128         return NOTIFY_DONE;
3129 }
3130
3131 static struct notifier_block bond_netdev_notifier = {
3132         .notifier_call = bond_netdev_event,
3133 };
3134
3135 /*---------------------------- Hashing Policies -----------------------------*/
3136
3137 /* L2 hash helper */
3138 static inline u32 bond_eth_hash(struct sk_buff *skb)
3139 {
3140         struct ethhdr *ep, hdr_tmp;
3141
3142         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3143         if (ep)
3144                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3145         return 0;
3146 }
3147
3148 /* Extract the appropriate headers based on bond's xmit policy */
3149 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3150                               struct flow_keys *fk)
3151 {
3152         const struct ipv6hdr *iph6;
3153         const struct iphdr *iph;
3154         int noff, proto = -1;
3155
3156         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3157                 return skb_flow_dissect_flow_keys(skb, fk, 0);
3158
3159         fk->ports.ports = 0;
3160         noff = skb_network_offset(skb);
3161         if (skb->protocol == htons(ETH_P_IP)) {
3162                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3163                         return false;
3164                 iph = ip_hdr(skb);
3165                 iph_to_flow_copy_v4addrs(fk, iph);
3166                 noff += iph->ihl << 2;
3167                 if (!ip_is_fragment(iph))
3168                         proto = iph->protocol;
3169         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3170                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3171                         return false;
3172                 iph6 = ipv6_hdr(skb);
3173                 iph_to_flow_copy_v6addrs(fk, iph6);
3174                 noff += sizeof(*iph6);
3175                 proto = iph6->nexthdr;
3176         } else {
3177                 return false;
3178         }
3179         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3180                 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3181
3182         return true;
3183 }
3184
3185 /**
3186  * bond_xmit_hash - generate a hash value based on the xmit policy
3187  * @bond: bonding device
3188  * @skb: buffer to use for headers
3189  *
3190  * This function will extract the necessary headers from the skb buffer and use
3191  * them to generate a hash based on the xmit_policy set in the bonding device
3192  */
3193 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3194 {
3195         struct flow_keys flow;
3196         u32 hash;
3197
3198         if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3199             skb->l4_hash)
3200                 return skb->hash;
3201
3202         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3203             !bond_flow_dissect(bond, skb, &flow))
3204                 return bond_eth_hash(skb);
3205
3206         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3207             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3208                 hash = bond_eth_hash(skb);
3209         else
3210                 hash = (__force u32)flow.ports.ports;
3211         hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3212                 (__force u32)flow_get_u32_src(&flow);
3213         hash ^= (hash >> 16);
3214         hash ^= (hash >> 8);
3215
3216         return hash;
3217 }
3218
3219 /*-------------------------- Device entry points ----------------------------*/
3220
3221 static void bond_work_init_all(struct bonding *bond)
3222 {
3223         INIT_DELAYED_WORK(&bond->mcast_work,
3224                           bond_resend_igmp_join_requests_delayed);
3225         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3226         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3227         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3228                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3229         else
3230                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3231         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3232         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3233 }
3234
3235 static void bond_work_cancel_all(struct bonding *bond)
3236 {
3237         cancel_delayed_work_sync(&bond->mii_work);
3238         cancel_delayed_work_sync(&bond->arp_work);
3239         cancel_delayed_work_sync(&bond->alb_work);
3240         cancel_delayed_work_sync(&bond->ad_work);
3241         cancel_delayed_work_sync(&bond->mcast_work);
3242         cancel_delayed_work_sync(&bond->slave_arr_work);
3243 }
3244
3245 static int bond_open(struct net_device *bond_dev)
3246 {
3247         struct bonding *bond = netdev_priv(bond_dev);
3248         struct list_head *iter;
3249         struct slave *slave;
3250
3251         /* reset slave->backup and slave->inactive */
3252         if (bond_has_slaves(bond)) {
3253                 bond_for_each_slave(bond, slave, iter) {
3254                         if (bond_uses_primary(bond) &&
3255                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3256                                 bond_set_slave_inactive_flags(slave,
3257                                                               BOND_SLAVE_NOTIFY_NOW);
3258                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3259                                 bond_set_slave_active_flags(slave,
3260                                                             BOND_SLAVE_NOTIFY_NOW);
3261                         }
3262                 }
3263         }
3264
3265         bond_work_init_all(bond);
3266
3267         if (bond_is_lb(bond)) {
3268                 /* bond_alb_initialize must be called before the timer
3269                  * is started.
3270                  */
3271                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3272                         return -ENOMEM;
3273                 if (bond->params.tlb_dynamic_lb)
3274                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3275         }
3276
3277         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3278                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3279
3280         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3281                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3282                 bond->recv_probe = bond_arp_rcv;
3283         }
3284
3285         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3286                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3287                 /* register to receive LACPDUs */
3288                 bond->recv_probe = bond_3ad_lacpdu_recv;
3289                 bond_3ad_initiate_agg_selection(bond, 1);
3290         }
3291
3292         if (bond_mode_uses_xmit_hash(bond))
3293                 bond_update_slave_arr(bond, NULL);
3294
3295         return 0;
3296 }
3297
3298 static int bond_close(struct net_device *bond_dev)
3299 {
3300         struct bonding *bond = netdev_priv(bond_dev);
3301
3302         bond_work_cancel_all(bond);
3303         bond->send_peer_notif = 0;
3304         if (bond_is_lb(bond))
3305                 bond_alb_deinitialize(bond);
3306         bond->recv_probe = NULL;
3307
3308         return 0;
3309 }
3310
3311 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3312                                                 struct rtnl_link_stats64 *stats)
3313 {
3314         struct bonding *bond = netdev_priv(bond_dev);
3315         struct rtnl_link_stats64 temp;
3316         struct list_head *iter;
3317         struct slave *slave;
3318
3319         memcpy(stats, &bond->bond_stats, sizeof(*stats));
3320
3321         bond_for_each_slave(bond, slave, iter) {
3322                 const struct rtnl_link_stats64 *sstats =
3323                         dev_get_stats(slave->dev, &temp);
3324                 struct rtnl_link_stats64 *pstats = &slave->slave_stats;
3325
3326                 stats->rx_packets +=  sstats->rx_packets - pstats->rx_packets;
3327                 stats->rx_bytes += sstats->rx_bytes - pstats->rx_bytes;
3328                 stats->rx_errors += sstats->rx_errors - pstats->rx_errors;
3329                 stats->rx_dropped += sstats->rx_dropped - pstats->rx_dropped;
3330
3331                 stats->tx_packets += sstats->tx_packets - pstats->tx_packets;;
3332                 stats->tx_bytes += sstats->tx_bytes - pstats->tx_bytes;
3333                 stats->tx_errors += sstats->tx_errors - pstats->tx_errors;
3334                 stats->tx_dropped += sstats->tx_dropped - pstats->tx_dropped;
3335
3336                 stats->multicast += sstats->multicast - pstats->multicast;
3337                 stats->collisions += sstats->collisions - pstats->collisions;
3338
3339                 stats->rx_length_errors += sstats->rx_length_errors - pstats->rx_length_errors;
3340                 stats->rx_over_errors += sstats->rx_over_errors - pstats->rx_over_errors;
3341                 stats->rx_crc_errors += sstats->rx_crc_errors - pstats->rx_crc_errors;
3342                 stats->rx_frame_errors += sstats->rx_frame_errors - pstats->rx_frame_errors;
3343                 stats->rx_fifo_errors += sstats->rx_fifo_errors - pstats->rx_fifo_errors;
3344                 stats->rx_missed_errors += sstats->rx_missed_errors - pstats->rx_missed_errors;
3345
3346                 stats->tx_aborted_errors += sstats->tx_aborted_errors - pstats->tx_aborted_errors;
3347                 stats->tx_carrier_errors += sstats->tx_carrier_errors - pstats->tx_carrier_errors;
3348                 stats->tx_fifo_errors += sstats->tx_fifo_errors - pstats->tx_fifo_errors;
3349                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors - pstats->tx_heartbeat_errors;
3350                 stats->tx_window_errors += sstats->tx_window_errors - pstats->tx_window_errors;
3351
3352                 /* save off the slave stats for the next run */
3353                 memcpy(pstats, sstats, sizeof(*sstats));
3354         }
3355         memcpy(&bond->bond_stats, stats, sizeof(*stats));
3356
3357         return stats;
3358 }
3359
3360 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3361 {
3362         struct bonding *bond = netdev_priv(bond_dev);
3363         struct net_device *slave_dev = NULL;
3364         struct ifbond k_binfo;
3365         struct ifbond __user *u_binfo = NULL;
3366         struct ifslave k_sinfo;
3367         struct ifslave __user *u_sinfo = NULL;
3368         struct mii_ioctl_data *mii = NULL;
3369         struct bond_opt_value newval;
3370         struct net *net;
3371         int res = 0;
3372
3373         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3374
3375         switch (cmd) {
3376         case SIOCGMIIPHY:
3377                 mii = if_mii(ifr);
3378                 if (!mii)
3379                         return -EINVAL;
3380
3381                 mii->phy_id = 0;
3382                 /* Fall Through */
3383         case SIOCGMIIREG:
3384                 /* We do this again just in case we were called by SIOCGMIIREG
3385                  * instead of SIOCGMIIPHY.
3386                  */
3387                 mii = if_mii(ifr);
3388                 if (!mii)
3389                         return -EINVAL;
3390
3391                 if (mii->reg_num == 1) {
3392                         mii->val_out = 0;
3393                         if (netif_carrier_ok(bond->dev))
3394                                 mii->val_out = BMSR_LSTATUS;
3395                 }
3396
3397                 return 0;
3398         case BOND_INFO_QUERY_OLD:
3399         case SIOCBONDINFOQUERY:
3400                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3401
3402                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3403                         return -EFAULT;
3404
3405                 res = bond_info_query(bond_dev, &k_binfo);
3406                 if (res == 0 &&
3407                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3408                         return -EFAULT;
3409
3410                 return res;
3411         case BOND_SLAVE_INFO_QUERY_OLD:
3412         case SIOCBONDSLAVEINFOQUERY:
3413                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3414
3415                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3416                         return -EFAULT;
3417
3418                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3419                 if (res == 0 &&
3420                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3421                         return -EFAULT;
3422
3423                 return res;
3424         default:
3425                 break;
3426         }
3427
3428         net = dev_net(bond_dev);
3429
3430         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3431                 return -EPERM;
3432
3433         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3434
3435         netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3436
3437         if (!slave_dev)
3438                 return -ENODEV;
3439
3440         netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3441         switch (cmd) {
3442         case BOND_ENSLAVE_OLD:
3443         case SIOCBONDENSLAVE:
3444                 res = bond_enslave(bond_dev, slave_dev);
3445                 break;
3446         case BOND_RELEASE_OLD:
3447         case SIOCBONDRELEASE:
3448                 res = bond_release(bond_dev, slave_dev);
3449                 break;
3450         case BOND_SETHWADDR_OLD:
3451         case SIOCBONDSETHWADDR:
3452                 bond_set_dev_addr(bond_dev, slave_dev);
3453                 res = 0;
3454                 break;
3455         case BOND_CHANGE_ACTIVE_OLD:
3456         case SIOCBONDCHANGEACTIVE:
3457                 bond_opt_initstr(&newval, slave_dev->name);
3458                 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3459                 break;
3460         default:
3461                 res = -EOPNOTSUPP;
3462         }
3463
3464         return res;
3465 }
3466
3467 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3468 {
3469         struct bonding *bond = netdev_priv(bond_dev);
3470
3471         if (change & IFF_PROMISC)
3472                 bond_set_promiscuity(bond,
3473                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3474
3475         if (change & IFF_ALLMULTI)
3476                 bond_set_allmulti(bond,
3477                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3478 }
3479
3480 static void bond_set_rx_mode(struct net_device *bond_dev)
3481 {
3482         struct bonding *bond = netdev_priv(bond_dev);
3483         struct list_head *iter;
3484         struct slave *slave;
3485
3486         rcu_read_lock();
3487         if (bond_uses_primary(bond)) {
3488                 slave = rcu_dereference(bond->curr_active_slave);
3489                 if (slave) {
3490                         dev_uc_sync(slave->dev, bond_dev);
3491                         dev_mc_sync(slave->dev, bond_dev);
3492                 }
3493         } else {
3494                 bond_for_each_slave_rcu(bond, slave, iter) {
3495                         dev_uc_sync_multiple(slave->dev, bond_dev);
3496                         dev_mc_sync_multiple(slave->dev, bond_dev);
3497                 }
3498         }
3499         rcu_read_unlock();
3500 }
3501
3502 static int bond_neigh_init(struct neighbour *n)
3503 {
3504         struct bonding *bond = netdev_priv(n->dev);
3505         const struct net_device_ops *slave_ops;
3506         struct neigh_parms parms;
3507         struct slave *slave;
3508         int ret;
3509
3510         slave = bond_first_slave(bond);
3511         if (!slave)
3512                 return 0;
3513         slave_ops = slave->dev->netdev_ops;
3514         if (!slave_ops->ndo_neigh_setup)
3515                 return 0;
3516
3517         parms.neigh_setup = NULL;
3518         parms.neigh_cleanup = NULL;
3519         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3520         if (ret)
3521                 return ret;
3522
3523         /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3524          * after the last slave has been detached.  Assumes that all slaves
3525          * utilize the same neigh_cleanup (true at this writing as only user
3526          * is ipoib).
3527          */
3528         n->parms->neigh_cleanup = parms.neigh_cleanup;
3529
3530         if (!parms.neigh_setup)
3531                 return 0;
3532
3533         return parms.neigh_setup(n);
3534 }
3535
3536 /* The bonding ndo_neigh_setup is called at init time beofre any
3537  * slave exists. So we must declare proxy setup function which will
3538  * be used at run time to resolve the actual slave neigh param setup.
3539  *
3540  * It's also called by master devices (such as vlans) to setup their
3541  * underlying devices. In that case - do nothing, we're already set up from
3542  * our init.
3543  */
3544 static int bond_neigh_setup(struct net_device *dev,
3545                             struct neigh_parms *parms)
3546 {
3547         /* modify only our neigh_parms */
3548         if (parms->dev == dev)
3549                 parms->neigh_setup = bond_neigh_init;
3550
3551         return 0;
3552 }
3553
3554 /* Change the MTU of all of a master's slaves to match the master */
3555 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3556 {
3557         struct bonding *bond = netdev_priv(bond_dev);
3558         struct slave *slave, *rollback_slave;
3559         struct list_head *iter;
3560         int res = 0;
3561
3562         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3563
3564         bond_for_each_slave(bond, slave, iter) {
3565                 netdev_dbg(bond_dev, "s %p c_m %p\n",
3566                            slave, slave->dev->netdev_ops->ndo_change_mtu);
3567
3568                 res = dev_set_mtu(slave->dev, new_mtu);
3569
3570                 if (res) {
3571                         /* If we failed to set the slave's mtu to the new value
3572                          * we must abort the operation even in ACTIVE_BACKUP
3573                          * mode, because if we allow the backup slaves to have
3574                          * different mtu values than the active slave we'll
3575                          * need to change their mtu when doing a failover. That
3576                          * means changing their mtu from timer context, which
3577                          * is probably not a good idea.
3578                          */
3579                         netdev_dbg(bond_dev, "err %d %s\n", res,
3580                                    slave->dev->name);
3581                         goto unwind;
3582                 }
3583         }
3584
3585         bond_dev->mtu = new_mtu;
3586
3587         return 0;
3588
3589 unwind:
3590         /* unwind from head to the slave that failed */
3591         bond_for_each_slave(bond, rollback_slave, iter) {
3592                 int tmp_res;
3593
3594                 if (rollback_slave == slave)
3595                         break;
3596
3597                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3598                 if (tmp_res) {
3599                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3600                                    tmp_res, rollback_slave->dev->name);
3601                 }
3602         }
3603
3604         return res;
3605 }
3606
3607 /* Change HW address
3608  *
3609  * Note that many devices must be down to change the HW address, and
3610  * downing the master releases all slaves.  We can make bonds full of
3611  * bonding devices to test this, however.
3612  */
3613 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3614 {
3615         struct bonding *bond = netdev_priv(bond_dev);
3616         struct slave *slave, *rollback_slave;
3617         struct sockaddr *sa = addr, tmp_sa;
3618         struct list_head *iter;
3619         int res = 0;
3620
3621         if (BOND_MODE(bond) == BOND_MODE_ALB)
3622                 return bond_alb_set_mac_address(bond_dev, addr);
3623
3624
3625         netdev_dbg(bond_dev, "bond=%p\n", bond);
3626
3627         /* If fail_over_mac is enabled, do nothing and return success.
3628          * Returning an error causes ifenslave to fail.
3629          */
3630         if (bond->params.fail_over_mac &&
3631             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3632                 return 0;
3633
3634         if (!is_valid_ether_addr(sa->sa_data))
3635                 return -EADDRNOTAVAIL;
3636
3637         bond_for_each_slave(bond, slave, iter) {
3638                 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3639                 res = dev_set_mac_address(slave->dev, addr);
3640                 if (res) {
3641                         /* TODO: consider downing the slave
3642                          * and retry ?
3643                          * User should expect communications
3644                          * breakage anyway until ARP finish
3645                          * updating, so...
3646                          */
3647                         netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3648                         goto unwind;
3649                 }
3650         }
3651
3652         /* success */
3653         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3654         return 0;
3655
3656 unwind:
3657         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3658         tmp_sa.sa_family = bond_dev->type;
3659
3660         /* unwind from head to the slave that failed */
3661         bond_for_each_slave(bond, rollback_slave, iter) {
3662                 int tmp_res;
3663
3664                 if (rollback_slave == slave)
3665                         break;
3666
3667                 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3668                 if (tmp_res) {
3669                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3670                                    tmp_res, rollback_slave->dev->name);
3671                 }
3672         }
3673
3674         return res;
3675 }
3676
3677 /**
3678  * bond_xmit_slave_id - transmit skb through slave with slave_id
3679  * @bond: bonding device that is transmitting
3680  * @skb: buffer to transmit
3681  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3682  *
3683  * This function tries to transmit through slave with slave_id but in case
3684  * it fails, it tries to find the first available slave for transmission.
3685  * The skb is consumed in all cases, thus the function is void.
3686  */
3687 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3688 {
3689         struct list_head *iter;
3690         struct slave *slave;
3691         int i = slave_id;
3692
3693         /* Here we start from the slave with slave_id */
3694         bond_for_each_slave_rcu(bond, slave, iter) {
3695                 if (--i < 0) {
3696                         if (bond_slave_can_tx(slave)) {
3697                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3698                                 return;
3699                         }
3700                 }
3701         }
3702
3703         /* Here we start from the first slave up to slave_id */
3704         i = slave_id;
3705         bond_for_each_slave_rcu(bond, slave, iter) {
3706                 if (--i < 0)
3707                         break;
3708                 if (bond_slave_can_tx(slave)) {
3709                         bond_dev_queue_xmit(bond, skb, slave->dev);
3710                         return;
3711                 }
3712         }
3713         /* no slave that can tx has been found */
3714         bond_tx_drop(bond->dev, skb);
3715 }
3716
3717 /**
3718  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3719  * @bond: bonding device to use
3720  *
3721  * Based on the value of the bonding device's packets_per_slave parameter
3722  * this function generates a slave id, which is usually used as the next
3723  * slave to transmit through.
3724  */
3725 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3726 {
3727         u32 slave_id;
3728         struct reciprocal_value reciprocal_packets_per_slave;
3729         int packets_per_slave = bond->params.packets_per_slave;
3730
3731         switch (packets_per_slave) {
3732         case 0:
3733                 slave_id = prandom_u32();
3734                 break;
3735         case 1:
3736                 slave_id = bond->rr_tx_counter;
3737                 break;
3738         default:
3739                 reciprocal_packets_per_slave =
3740                         bond->params.reciprocal_packets_per_slave;
3741                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3742                                              reciprocal_packets_per_slave);
3743                 break;
3744         }
3745         bond->rr_tx_counter++;
3746
3747         return slave_id;
3748 }
3749
3750 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3751 {
3752         struct bonding *bond = netdev_priv(bond_dev);
3753         struct iphdr *iph = ip_hdr(skb);
3754         struct slave *slave;
3755         u32 slave_id;
3756
3757         /* Start with the curr_active_slave that joined the bond as the
3758          * default for sending IGMP traffic.  For failover purposes one
3759          * needs to maintain some consistency for the interface that will
3760          * send the join/membership reports.  The curr_active_slave found
3761          * will send all of this type of traffic.
3762          */
3763         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3764                 slave = rcu_dereference(bond->curr_active_slave);
3765                 if (slave)
3766                         bond_dev_queue_xmit(bond, skb, slave->dev);
3767                 else
3768                         bond_xmit_slave_id(bond, skb, 0);
3769         } else {
3770                 int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3771
3772                 if (likely(slave_cnt)) {
3773                         slave_id = bond_rr_gen_slave_id(bond);
3774                         bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3775                 } else {
3776                         bond_tx_drop(bond_dev, skb);
3777                 }
3778         }
3779
3780         return NETDEV_TX_OK;
3781 }
3782
3783 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3784  * the bond has a usable interface.
3785  */
3786 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3787 {
3788         struct bonding *bond = netdev_priv(bond_dev);
3789         struct slave *slave;
3790
3791         slave = rcu_dereference(bond->curr_active_slave);
3792         if (slave)
3793                 bond_dev_queue_xmit(bond, skb, slave->dev);
3794         else
3795                 bond_tx_drop(bond_dev, skb);
3796
3797         return NETDEV_TX_OK;
3798 }
3799
3800 /* Use this to update slave_array when (a) it's not appropriate to update
3801  * slave_array right away (note that update_slave_array() may sleep)
3802  * and / or (b) RTNL is not held.
3803  */
3804 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3805 {
3806         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3807 }
3808
3809 /* Slave array work handler. Holds only RTNL */
3810 static void bond_slave_arr_handler(struct work_struct *work)
3811 {
3812         struct bonding *bond = container_of(work, struct bonding,
3813                                             slave_arr_work.work);
3814         int ret;
3815
3816         if (!rtnl_trylock())
3817                 goto err;
3818
3819         ret = bond_update_slave_arr(bond, NULL);
3820         rtnl_unlock();
3821         if (ret) {
3822                 pr_warn_ratelimited("Failed to update slave array from WT\n");
3823                 goto err;
3824         }
3825         return;
3826
3827 err:
3828         bond_slave_arr_work_rearm(bond, 1);
3829 }
3830
3831 /* Build the usable slaves array in control path for modes that use xmit-hash
3832  * to determine the slave interface -
3833  * (a) BOND_MODE_8023AD
3834  * (b) BOND_MODE_XOR
3835  * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3836  *
3837  * The caller is expected to hold RTNL only and NO other lock!
3838  */
3839 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3840 {
3841         struct slave *slave;
3842         struct list_head *iter;
3843         struct bond_up_slave *new_arr, *old_arr;
3844         int agg_id = 0;
3845         int ret = 0;
3846
3847 #ifdef CONFIG_LOCKDEP
3848         WARN_ON(lockdep_is_held(&bond->mode_lock));
3849 #endif
3850
3851         new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3852                           GFP_KERNEL);
3853         if (!new_arr) {
3854                 ret = -ENOMEM;
3855                 pr_err("Failed to build slave-array.\n");
3856                 goto out;
3857         }
3858         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3859                 struct ad_info ad_info;
3860
3861                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3862                         pr_debug("bond_3ad_get_active_agg_info failed\n");
3863                         kfree_rcu(new_arr, rcu);
3864                         /* No active aggragator means it's not safe to use
3865                          * the previous array.
3866                          */
3867                         old_arr = rtnl_dereference(bond->slave_arr);
3868                         if (old_arr) {
3869                                 RCU_INIT_POINTER(bond->slave_arr, NULL);
3870                                 kfree_rcu(old_arr, rcu);
3871                         }
3872                         goto out;
3873                 }
3874                 agg_id = ad_info.aggregator_id;
3875         }
3876         bond_for_each_slave(bond, slave, iter) {
3877                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3878                         struct aggregator *agg;
3879
3880                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
3881                         if (!agg || agg->aggregator_identifier != agg_id)
3882                                 continue;
3883                 }
3884                 if (!bond_slave_can_tx(slave))
3885                         continue;
3886                 if (skipslave == slave)
3887                         continue;
3888                 new_arr->arr[new_arr->count++] = slave;
3889         }
3890
3891         old_arr = rtnl_dereference(bond->slave_arr);
3892         rcu_assign_pointer(bond->slave_arr, new_arr);
3893         if (old_arr)
3894                 kfree_rcu(old_arr, rcu);
3895 out:
3896         if (ret != 0 && skipslave) {
3897                 int idx;
3898
3899                 /* Rare situation where caller has asked to skip a specific
3900                  * slave but allocation failed (most likely!). BTW this is
3901                  * only possible when the call is initiated from
3902                  * __bond_release_one(). In this situation; overwrite the
3903                  * skipslave entry in the array with the last entry from the
3904                  * array to avoid a situation where the xmit path may choose
3905                  * this to-be-skipped slave to send a packet out.
3906                  */
3907                 old_arr = rtnl_dereference(bond->slave_arr);
3908                 for (idx = 0; idx < old_arr->count; idx++) {
3909                         if (skipslave == old_arr->arr[idx]) {
3910                                 old_arr->arr[idx] =
3911                                     old_arr->arr[old_arr->count-1];
3912                                 old_arr->count--;
3913                                 break;
3914                         }
3915                 }
3916         }
3917         return ret;
3918 }
3919
3920 /* Use this Xmit function for 3AD as well as XOR modes. The current
3921  * usable slave array is formed in the control path. The xmit function
3922  * just calculates hash and sends the packet out.
3923  */
3924 static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3925 {
3926         struct bonding *bond = netdev_priv(dev);
3927         struct slave *slave;
3928         struct bond_up_slave *slaves;
3929         unsigned int count;
3930
3931         slaves = rcu_dereference(bond->slave_arr);
3932         count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3933         if (likely(count)) {
3934                 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3935                 bond_dev_queue_xmit(bond, skb, slave->dev);
3936         } else {
3937                 bond_tx_drop(dev, skb);
3938         }
3939
3940         return NETDEV_TX_OK;
3941 }
3942
3943 /* in broadcast mode, we send everything to all usable interfaces. */
3944 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3945 {
3946         struct bonding *bond = netdev_priv(bond_dev);
3947         struct slave *slave = NULL;
3948         struct list_head *iter;
3949
3950         bond_for_each_slave_rcu(bond, slave, iter) {
3951                 if (bond_is_last_slave(bond, slave))
3952                         break;
3953                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3954                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3955
3956                         if (!skb2) {
3957                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3958                                                     bond_dev->name, __func__);
3959                                 continue;
3960                         }
3961                         bond_dev_queue_xmit(bond, skb2, slave->dev);
3962                 }
3963         }
3964         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3965                 bond_dev_queue_xmit(bond, skb, slave->dev);
3966         else
3967                 bond_tx_drop(bond_dev, skb);
3968
3969         return NETDEV_TX_OK;
3970 }
3971
3972 /*------------------------- Device initialization ---------------------------*/
3973
3974 /* Lookup the slave that corresponds to a qid */
3975 static inline int bond_slave_override(struct bonding *bond,
3976                                       struct sk_buff *skb)
3977 {
3978         struct slave *slave = NULL;
3979         struct list_head *iter;
3980
3981         if (!skb->queue_mapping)
3982                 return 1;
3983
3984         /* Find out if any slaves have the same mapping as this skb. */
3985         bond_for_each_slave_rcu(bond, slave, iter) {
3986                 if (slave->queue_id == skb->queue_mapping) {
3987                         if (bond_slave_is_up(slave) &&
3988                             slave->link == BOND_LINK_UP) {
3989                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3990                                 return 0;
3991                         }
3992                         /* If the slave isn't UP, use default transmit policy. */
3993                         break;
3994                 }
3995         }
3996
3997         return 1;
3998 }
3999
4000
4001 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4002                              void *accel_priv, select_queue_fallback_t fallback)
4003 {
4004         /* This helper function exists to help dev_pick_tx get the correct
4005          * destination queue.  Using a helper function skips a call to
4006          * skb_tx_hash and will put the skbs in the queue we expect on their
4007          * way down to the bonding driver.
4008          */
4009         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4010
4011         /* Save the original txq to restore before passing to the driver */
4012         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4013
4014         if (unlikely(txq >= dev->real_num_tx_queues)) {
4015                 do {
4016                         txq -= dev->real_num_tx_queues;
4017                 } while (txq >= dev->real_num_tx_queues);
4018         }
4019         return txq;
4020 }
4021
4022 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4023 {
4024         struct bonding *bond = netdev_priv(dev);
4025
4026         if (bond_should_override_tx_queue(bond) &&
4027             !bond_slave_override(bond, skb))
4028                 return NETDEV_TX_OK;
4029
4030         switch (BOND_MODE(bond)) {
4031         case BOND_MODE_ROUNDROBIN:
4032                 return bond_xmit_roundrobin(skb, dev);
4033         case BOND_MODE_ACTIVEBACKUP:
4034                 return bond_xmit_activebackup(skb, dev);
4035         case BOND_MODE_8023AD:
4036         case BOND_MODE_XOR:
4037                 return bond_3ad_xor_xmit(skb, dev);
4038         case BOND_MODE_BROADCAST:
4039                 return bond_xmit_broadcast(skb, dev);
4040         case BOND_MODE_ALB:
4041                 return bond_alb_xmit(skb, dev);
4042         case BOND_MODE_TLB:
4043                 return bond_tlb_xmit(skb, dev);
4044         default:
4045                 /* Should never happen, mode already checked */
4046                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4047                 WARN_ON_ONCE(1);
4048                 bond_tx_drop(dev, skb);
4049                 return NETDEV_TX_OK;
4050         }
4051 }
4052
4053 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4054 {
4055         struct bonding *bond = netdev_priv(dev);
4056         netdev_tx_t ret = NETDEV_TX_OK;
4057
4058         /* If we risk deadlock from transmitting this in the
4059          * netpoll path, tell netpoll to queue the frame for later tx
4060          */
4061         if (unlikely(is_netpoll_tx_blocked(dev)))
4062                 return NETDEV_TX_BUSY;
4063
4064         rcu_read_lock();
4065         if (bond_has_slaves(bond))
4066                 ret = __bond_start_xmit(skb, dev);
4067         else
4068                 bond_tx_drop(dev, skb);
4069         rcu_read_unlock();
4070
4071         return ret;
4072 }
4073
4074 static int bond_ethtool_get_settings(struct net_device *bond_dev,
4075                                      struct ethtool_cmd *ecmd)
4076 {
4077         struct bonding *bond = netdev_priv(bond_dev);
4078         unsigned long speed = 0;
4079         struct list_head *iter;
4080         struct slave *slave;
4081
4082         ecmd->duplex = DUPLEX_UNKNOWN;
4083         ecmd->port = PORT_OTHER;
4084
4085         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4086          * do not need to check mode.  Though link speed might not represent
4087          * the true receive or transmit bandwidth (not all modes are symmetric)
4088          * this is an accurate maximum.
4089          */
4090         bond_for_each_slave(bond, slave, iter) {
4091                 if (bond_slave_can_tx(slave)) {
4092                         if (slave->speed != SPEED_UNKNOWN)
4093                                 speed += slave->speed;
4094                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
4095                             slave->duplex != DUPLEX_UNKNOWN)
4096                                 ecmd->duplex = slave->duplex;
4097                 }
4098         }
4099         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
4100
4101         return 0;
4102 }
4103
4104 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4105                                      struct ethtool_drvinfo *drvinfo)
4106 {
4107         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4108         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4109         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4110                  BOND_ABI_VERSION);
4111 }
4112
4113 static const struct ethtool_ops bond_ethtool_ops = {
4114         .get_drvinfo            = bond_ethtool_get_drvinfo,
4115         .get_settings           = bond_ethtool_get_settings,
4116         .get_link               = ethtool_op_get_link,
4117 };
4118
4119 static const struct net_device_ops bond_netdev_ops = {
4120         .ndo_init               = bond_init,
4121         .ndo_uninit             = bond_uninit,
4122         .ndo_open               = bond_open,
4123         .ndo_stop               = bond_close,
4124         .ndo_start_xmit         = bond_start_xmit,
4125         .ndo_select_queue       = bond_select_queue,
4126         .ndo_get_stats64        = bond_get_stats,
4127         .ndo_do_ioctl           = bond_do_ioctl,
4128         .ndo_change_rx_flags    = bond_change_rx_flags,
4129         .ndo_set_rx_mode        = bond_set_rx_mode,
4130         .ndo_change_mtu         = bond_change_mtu,
4131         .ndo_set_mac_address    = bond_set_mac_address,
4132         .ndo_neigh_setup        = bond_neigh_setup,
4133         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4134         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4135 #ifdef CONFIG_NET_POLL_CONTROLLER
4136         .ndo_netpoll_setup      = bond_netpoll_setup,
4137         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4138         .ndo_poll_controller    = bond_poll_controller,
4139 #endif
4140         .ndo_add_slave          = bond_enslave,
4141         .ndo_del_slave          = bond_release,
4142         .ndo_fix_features       = bond_fix_features,
4143         .ndo_bridge_setlink     = switchdev_port_bridge_setlink,
4144         .ndo_bridge_getlink     = switchdev_port_bridge_getlink,
4145         .ndo_bridge_dellink     = switchdev_port_bridge_dellink,
4146         .ndo_fdb_add            = switchdev_port_fdb_add,
4147         .ndo_fdb_del            = switchdev_port_fdb_del,
4148         .ndo_fdb_dump           = switchdev_port_fdb_dump,
4149         .ndo_features_check     = passthru_features_check,
4150 };
4151
4152 static const struct device_type bond_type = {
4153         .name = "bond",
4154 };
4155
4156 static void bond_destructor(struct net_device *bond_dev)
4157 {
4158         struct bonding *bond = netdev_priv(bond_dev);
4159         if (bond->wq)
4160                 destroy_workqueue(bond->wq);
4161         free_netdev(bond_dev);
4162 }
4163
4164 void bond_setup(struct net_device *bond_dev)
4165 {
4166         struct bonding *bond = netdev_priv(bond_dev);
4167
4168         spin_lock_init(&bond->mode_lock);
4169         bond->params = bonding_defaults;
4170
4171         /* Initialize pointers */
4172         bond->dev = bond_dev;
4173
4174         /* Initialize the device entry points */
4175         ether_setup(bond_dev);
4176         bond_dev->netdev_ops = &bond_netdev_ops;
4177         bond_dev->ethtool_ops = &bond_ethtool_ops;
4178
4179         bond_dev->destructor = bond_destructor;
4180
4181         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4182
4183         /* Initialize the device options */
4184         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4185         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4186         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4187
4188         /* don't acquire bond device's netif_tx_lock when transmitting */
4189         bond_dev->features |= NETIF_F_LLTX;
4190
4191         /* By default, we declare the bond to be fully
4192          * VLAN hardware accelerated capable. Special
4193          * care is taken in the various xmit functions
4194          * when there are slaves that are not hw accel
4195          * capable
4196          */
4197
4198         /* Don't allow bond devices to change network namespaces. */
4199         bond_dev->features |= NETIF_F_NETNS_LOCAL;
4200
4201         bond_dev->hw_features = BOND_VLAN_FEATURES |
4202                                 NETIF_F_HW_VLAN_CTAG_TX |
4203                                 NETIF_F_HW_VLAN_CTAG_RX |
4204                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4205
4206         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4207         bond_dev->features |= bond_dev->hw_features;
4208 }
4209
4210 /* Destroy a bonding device.
4211  * Must be under rtnl_lock when this function is called.
4212  */
4213 static void bond_uninit(struct net_device *bond_dev)
4214 {
4215         struct bonding *bond = netdev_priv(bond_dev);
4216         struct list_head *iter;
4217         struct slave *slave;
4218         struct bond_up_slave *arr;
4219
4220         bond_netpoll_cleanup(bond_dev);
4221
4222         /* Release the bonded slaves */
4223         bond_for_each_slave(bond, slave, iter)
4224                 __bond_release_one(bond_dev, slave->dev, true);
4225         netdev_info(bond_dev, "Released all slaves\n");
4226
4227         arr = rtnl_dereference(bond->slave_arr);
4228         if (arr) {
4229                 RCU_INIT_POINTER(bond->slave_arr, NULL);
4230                 kfree_rcu(arr, rcu);
4231         }
4232
4233         list_del(&bond->bond_list);
4234
4235         bond_debug_unregister(bond);
4236 }
4237
4238 /*------------------------- Module initialization ---------------------------*/
4239
4240 static int bond_check_params(struct bond_params *params)
4241 {
4242         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4243         struct bond_opt_value newval;
4244         const struct bond_opt_value *valptr;
4245         int arp_all_targets_value;
4246         u16 ad_actor_sys_prio = 0;
4247         u16 ad_user_port_key = 0;
4248
4249         /* Convert string parameters. */
4250         if (mode) {
4251                 bond_opt_initstr(&newval, mode);
4252                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4253                 if (!valptr) {
4254                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4255                         return -EINVAL;
4256                 }
4257                 bond_mode = valptr->value;
4258         }
4259
4260         if (xmit_hash_policy) {
4261                 if ((bond_mode != BOND_MODE_XOR) &&
4262                     (bond_mode != BOND_MODE_8023AD) &&
4263                     (bond_mode != BOND_MODE_TLB)) {
4264                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4265                                 bond_mode_name(bond_mode));
4266                 } else {
4267                         bond_opt_initstr(&newval, xmit_hash_policy);
4268                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4269                                                 &newval);
4270                         if (!valptr) {
4271                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4272                                        xmit_hash_policy);
4273                                 return -EINVAL;
4274                         }
4275                         xmit_hashtype = valptr->value;
4276                 }
4277         }
4278
4279         if (lacp_rate) {
4280                 if (bond_mode != BOND_MODE_8023AD) {
4281                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4282                                 bond_mode_name(bond_mode));
4283                 } else {
4284                         bond_opt_initstr(&newval, lacp_rate);
4285                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4286                                                 &newval);
4287                         if (!valptr) {
4288                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4289                                        lacp_rate);
4290                                 return -EINVAL;
4291                         }
4292                         lacp_fast = valptr->value;
4293                 }
4294         }
4295
4296         if (ad_select) {
4297                 bond_opt_initstr(&newval, ad_select);
4298                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4299                                         &newval);
4300                 if (!valptr) {
4301                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4302                         return -EINVAL;
4303                 }
4304                 params->ad_select = valptr->value;
4305                 if (bond_mode != BOND_MODE_8023AD)
4306                         pr_warn("ad_select param only affects 802.3ad mode\n");
4307         } else {
4308                 params->ad_select = BOND_AD_STABLE;
4309         }
4310
4311         if (max_bonds < 0) {
4312                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4313                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4314                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4315         }
4316
4317         if (miimon < 0) {
4318                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4319                         miimon, INT_MAX);
4320                 miimon = 0;
4321         }
4322
4323         if (updelay < 0) {
4324                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4325                         updelay, INT_MAX);
4326                 updelay = 0;
4327         }
4328
4329         if (downdelay < 0) {
4330                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4331                         downdelay, INT_MAX);
4332                 downdelay = 0;
4333         }
4334
4335         if ((use_carrier != 0) && (use_carrier != 1)) {
4336                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4337                         use_carrier);
4338                 use_carrier = 1;
4339         }
4340
4341         if (num_peer_notif < 0 || num_peer_notif > 255) {
4342                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4343                         num_peer_notif);
4344                 num_peer_notif = 1;
4345         }
4346
4347         /* reset values for 802.3ad/TLB/ALB */
4348         if (!bond_mode_uses_arp(bond_mode)) {
4349                 if (!miimon) {
4350                         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");
4351                         pr_warn("Forcing miimon to 100msec\n");
4352                         miimon = BOND_DEFAULT_MIIMON;
4353                 }
4354         }
4355
4356         if (tx_queues < 1 || tx_queues > 255) {
4357                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4358                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4359                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4360         }
4361
4362         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4363                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4364                         all_slaves_active);
4365                 all_slaves_active = 0;
4366         }
4367
4368         if (resend_igmp < 0 || resend_igmp > 255) {
4369                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4370                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4371                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4372         }
4373
4374         bond_opt_initval(&newval, packets_per_slave);
4375         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4376                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4377                         packets_per_slave, USHRT_MAX);
4378                 packets_per_slave = 1;
4379         }
4380
4381         if (bond_mode == BOND_MODE_ALB) {
4382                 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",
4383                           updelay);
4384         }
4385
4386         if (!miimon) {
4387                 if (updelay || downdelay) {
4388                         /* just warn the user the up/down delay will have
4389                          * no effect since miimon is zero...
4390                          */
4391                         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",
4392                                 updelay, downdelay);
4393                 }
4394         } else {
4395                 /* don't allow arp monitoring */
4396                 if (arp_interval) {
4397                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4398                                 miimon, arp_interval);
4399                         arp_interval = 0;
4400                 }
4401
4402                 if ((updelay % miimon) != 0) {
4403                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4404                                 updelay, miimon, (updelay / miimon) * miimon);
4405                 }
4406
4407                 updelay /= miimon;
4408
4409                 if ((downdelay % miimon) != 0) {
4410                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4411                                 downdelay, miimon,
4412                                 (downdelay / miimon) * miimon);
4413                 }
4414
4415                 downdelay /= miimon;
4416         }
4417
4418         if (arp_interval < 0) {
4419                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4420                         arp_interval, INT_MAX);
4421                 arp_interval = 0;
4422         }
4423
4424         for (arp_ip_count = 0, i = 0;
4425              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4426                 __be32 ip;
4427
4428                 /* not a complete check, but good enough to catch mistakes */
4429                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4430                     !bond_is_ip_target_ok(ip)) {
4431                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4432                                 arp_ip_target[i]);
4433                         arp_interval = 0;
4434                 } else {
4435                         if (bond_get_targets_ip(arp_target, ip) == -1)
4436                                 arp_target[arp_ip_count++] = ip;
4437                         else
4438                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4439                                         &ip);
4440                 }
4441         }
4442
4443         if (arp_interval && !arp_ip_count) {
4444                 /* don't allow arping if no arp_ip_target given... */
4445                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4446                         arp_interval);
4447                 arp_interval = 0;
4448         }
4449
4450         if (arp_validate) {
4451                 if (!arp_interval) {
4452                         pr_err("arp_validate requires arp_interval\n");
4453                         return -EINVAL;
4454                 }
4455
4456                 bond_opt_initstr(&newval, arp_validate);
4457                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4458                                         &newval);
4459                 if (!valptr) {
4460                         pr_err("Error: invalid arp_validate \"%s\"\n",
4461                                arp_validate);
4462                         return -EINVAL;
4463                 }
4464                 arp_validate_value = valptr->value;
4465         } else {
4466                 arp_validate_value = 0;
4467         }
4468
4469         arp_all_targets_value = 0;
4470         if (arp_all_targets) {
4471                 bond_opt_initstr(&newval, arp_all_targets);
4472                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4473                                         &newval);
4474                 if (!valptr) {
4475                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4476                                arp_all_targets);
4477                         arp_all_targets_value = 0;
4478                 } else {
4479                         arp_all_targets_value = valptr->value;
4480                 }
4481         }
4482
4483         if (miimon) {
4484                 pr_info("MII link monitoring set to %d ms\n", miimon);
4485         } else if (arp_interval) {
4486                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4487                                           arp_validate_value);
4488                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4489                         arp_interval, valptr->string, arp_ip_count);
4490
4491                 for (i = 0; i < arp_ip_count; i++)
4492                         pr_cont(" %s", arp_ip_target[i]);
4493
4494                 pr_cont("\n");
4495
4496         } else if (max_bonds) {
4497                 /* miimon and arp_interval not set, we need one so things
4498                  * work as expected, see bonding.txt for details
4499                  */
4500                 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");
4501         }
4502
4503         if (primary && !bond_mode_uses_primary(bond_mode)) {
4504                 /* currently, using a primary only makes sense
4505                  * in active backup, TLB or ALB modes
4506                  */
4507                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4508                         primary, bond_mode_name(bond_mode));
4509                 primary = NULL;
4510         }
4511
4512         if (primary && primary_reselect) {
4513                 bond_opt_initstr(&newval, primary_reselect);
4514                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4515                                         &newval);
4516                 if (!valptr) {
4517                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4518                                primary_reselect);
4519                         return -EINVAL;
4520                 }
4521                 primary_reselect_value = valptr->value;
4522         } else {
4523                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4524         }
4525
4526         if (fail_over_mac) {
4527                 bond_opt_initstr(&newval, fail_over_mac);
4528                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4529                                         &newval);
4530                 if (!valptr) {
4531                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4532                                fail_over_mac);
4533                         return -EINVAL;
4534                 }
4535                 fail_over_mac_value = valptr->value;
4536                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4537                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4538         } else {
4539                 fail_over_mac_value = BOND_FOM_NONE;
4540         }
4541
4542         bond_opt_initstr(&newval, "default");
4543         valptr = bond_opt_parse(
4544                         bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4545                                      &newval);
4546         if (!valptr) {
4547                 pr_err("Error: No ad_actor_sys_prio default value");
4548                 return -EINVAL;
4549         }
4550         ad_actor_sys_prio = valptr->value;
4551
4552         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4553                                 &newval);
4554         if (!valptr) {
4555                 pr_err("Error: No ad_user_port_key default value");
4556                 return -EINVAL;
4557         }
4558         ad_user_port_key = valptr->value;
4559
4560         if (lp_interval == 0) {
4561                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4562                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4563                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4564         }
4565
4566         /* fill params struct with the proper values */
4567         params->mode = bond_mode;
4568         params->xmit_policy = xmit_hashtype;
4569         params->miimon = miimon;
4570         params->num_peer_notif = num_peer_notif;
4571         params->arp_interval = arp_interval;
4572         params->arp_validate = arp_validate_value;
4573         params->arp_all_targets = arp_all_targets_value;
4574         params->updelay = updelay;
4575         params->downdelay = downdelay;
4576         params->use_carrier = use_carrier;
4577         params->lacp_fast = lacp_fast;
4578         params->primary[0] = 0;
4579         params->primary_reselect = primary_reselect_value;
4580         params->fail_over_mac = fail_over_mac_value;
4581         params->tx_queues = tx_queues;
4582         params->all_slaves_active = all_slaves_active;
4583         params->resend_igmp = resend_igmp;
4584         params->min_links = min_links;
4585         params->lp_interval = lp_interval;
4586         params->packets_per_slave = packets_per_slave;
4587         params->tlb_dynamic_lb = 1; /* Default value */
4588         params->ad_actor_sys_prio = ad_actor_sys_prio;
4589         eth_zero_addr(params->ad_actor_system);
4590         params->ad_user_port_key = ad_user_port_key;
4591         if (packets_per_slave > 0) {
4592                 params->reciprocal_packets_per_slave =
4593                         reciprocal_value(packets_per_slave);
4594         } else {
4595                 /* reciprocal_packets_per_slave is unused if
4596                  * packets_per_slave is 0 or 1, just initialize it
4597                  */
4598                 params->reciprocal_packets_per_slave =
4599                         (struct reciprocal_value) { 0 };
4600         }
4601
4602         if (primary) {
4603                 strncpy(params->primary, primary, IFNAMSIZ);
4604                 params->primary[IFNAMSIZ - 1] = 0;
4605         }
4606
4607         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4608
4609         return 0;
4610 }
4611
4612 static struct lock_class_key bonding_netdev_xmit_lock_key;
4613 static struct lock_class_key bonding_netdev_addr_lock_key;
4614 static struct lock_class_key bonding_tx_busylock_key;
4615
4616 static void bond_set_lockdep_class_one(struct net_device *dev,
4617                                        struct netdev_queue *txq,
4618                                        void *_unused)
4619 {
4620         lockdep_set_class(&txq->_xmit_lock,
4621                           &bonding_netdev_xmit_lock_key);
4622 }
4623
4624 static void bond_set_lockdep_class(struct net_device *dev)
4625 {
4626         lockdep_set_class(&dev->addr_list_lock,
4627                           &bonding_netdev_addr_lock_key);
4628         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4629         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4630 }
4631
4632 /* Called from registration process */
4633 static int bond_init(struct net_device *bond_dev)
4634 {
4635         struct bonding *bond = netdev_priv(bond_dev);
4636         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4637
4638         netdev_dbg(bond_dev, "Begin bond_init\n");
4639
4640         bond->wq = create_singlethread_workqueue(bond_dev->name);
4641         if (!bond->wq)
4642                 return -ENOMEM;
4643
4644         bond_set_lockdep_class(bond_dev);
4645
4646         list_add_tail(&bond->bond_list, &bn->dev_list);
4647
4648         bond_prepare_sysfs_group(bond);
4649
4650         bond_debug_register(bond);
4651
4652         /* Ensure valid dev_addr */
4653         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4654             bond_dev->addr_assign_type == NET_ADDR_PERM)
4655                 eth_hw_addr_random(bond_dev);
4656
4657         return 0;
4658 }
4659
4660 unsigned int bond_get_num_tx_queues(void)
4661 {
4662         return tx_queues;
4663 }
4664
4665 /* Create a new bond based on the specified name and bonding parameters.
4666  * If name is NULL, obtain a suitable "bond%d" name for us.
4667  * Caller must NOT hold rtnl_lock; we need to release it here before we
4668  * set up our sysfs entries.
4669  */
4670 int bond_create(struct net *net, const char *name)
4671 {
4672         struct net_device *bond_dev;
4673         struct bonding *bond;
4674         struct alb_bond_info *bond_info;
4675         int res;
4676
4677         rtnl_lock();
4678
4679         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4680                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
4681                                    bond_setup, tx_queues);
4682         if (!bond_dev) {
4683                 pr_err("%s: eek! can't alloc netdev!\n", name);
4684                 rtnl_unlock();
4685                 return -ENOMEM;
4686         }
4687
4688         /*
4689          * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4690          * It is set to 0 by default which is wrong.
4691          */
4692         bond = netdev_priv(bond_dev);
4693         bond_info = &(BOND_ALB_INFO(bond));
4694         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4695
4696         dev_net_set(bond_dev, net);
4697         bond_dev->rtnl_link_ops = &bond_link_ops;
4698
4699         res = register_netdevice(bond_dev);
4700
4701         netif_carrier_off(bond_dev);
4702
4703         rtnl_unlock();
4704         if (res < 0)
4705                 bond_destructor(bond_dev);
4706         return res;
4707 }
4708
4709 static int __net_init bond_net_init(struct net *net)
4710 {
4711         struct bond_net *bn = net_generic(net, bond_net_id);
4712
4713         bn->net = net;
4714         INIT_LIST_HEAD(&bn->dev_list);
4715
4716         bond_create_proc_dir(bn);
4717         bond_create_sysfs(bn);
4718
4719         return 0;
4720 }
4721
4722 static void __net_exit bond_net_exit(struct net *net)
4723 {
4724         struct bond_net *bn = net_generic(net, bond_net_id);
4725         struct bonding *bond, *tmp_bond;
4726         LIST_HEAD(list);
4727
4728         bond_destroy_sysfs(bn);
4729
4730         /* Kill off any bonds created after unregistering bond rtnl ops */
4731         rtnl_lock();
4732         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4733                 unregister_netdevice_queue(bond->dev, &list);
4734         unregister_netdevice_many(&list);
4735         rtnl_unlock();
4736
4737         bond_destroy_proc_dir(bn);
4738 }
4739
4740 static struct pernet_operations bond_net_ops = {
4741         .init = bond_net_init,
4742         .exit = bond_net_exit,
4743         .id   = &bond_net_id,
4744         .size = sizeof(struct bond_net),
4745 };
4746
4747 static int __init bonding_init(void)
4748 {
4749         int i;
4750         int res;
4751
4752         pr_info("%s", bond_version);
4753
4754         res = bond_check_params(&bonding_defaults);
4755         if (res)
4756                 goto out;
4757
4758         res = register_pernet_subsys(&bond_net_ops);
4759         if (res)
4760                 goto out;
4761
4762         res = bond_netlink_init();
4763         if (res)
4764                 goto err_link;
4765
4766         bond_create_debugfs();
4767
4768         for (i = 0; i < max_bonds; i++) {
4769                 res = bond_create(&init_net, NULL);
4770                 if (res)
4771                         goto err;
4772         }
4773
4774         register_netdevice_notifier(&bond_netdev_notifier);
4775 out:
4776         return res;
4777 err:
4778         bond_destroy_debugfs();
4779         bond_netlink_fini();
4780 err_link:
4781         unregister_pernet_subsys(&bond_net_ops);
4782         goto out;
4783
4784 }
4785
4786 static void __exit bonding_exit(void)
4787 {
4788         unregister_netdevice_notifier(&bond_netdev_notifier);
4789
4790         bond_destroy_debugfs();
4791
4792         bond_netlink_fini();
4793         unregister_pernet_subsys(&bond_net_ops);
4794
4795 #ifdef CONFIG_NET_POLL_CONTROLLER
4796         /* Make sure we don't have an imbalance on our netpoll blocking */
4797         WARN_ON(atomic_read(&netpoll_block_tx));
4798 #endif
4799 }
4800
4801 module_init(bonding_init);
4802 module_exit(bonding_exit);
4803 MODULE_LICENSE("GPL");
4804 MODULE_VERSION(DRV_VERSION);
4805 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4806 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");