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, "802.3ad aggregation selection logic; "
 156                             "0 for stable (default), 1 for bandwidth, "
 157                             "2 for count");
 158 module_param(min_links, int, 0);
 159 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 160
 161 module_param(xmit_hash_policy, charp, 0);
 162 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 163                                    "0 for layer 2 (default), 1 for layer 3+4, "
 164                                    "2 for layer 2+3, 3 for encap layer 2+3, "
 165                                    "4 for encap layer 3+4");
 166 module_param(arp_interval, int, 0);
 167 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 168 module_param_array(arp_ip_target, charp, NULL, 0);
 169 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 170 module_param(arp_validate, charp, 0);
 171 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 172                                "0 for none (default), 1 for active, "
 173                                "2 for backup, 3 for all");
 174 module_param(arp_all_targets, charp, 0);
 175 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 176 module_param(fail_over_mac, charp, 0);
 177 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 178                                 "the same MAC; 0 for none (default), "
 179                                 "1 for active, 2 for follow");
 180 module_param(all_slaves_active, int, 0);
 181 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
 182                                      "by setting active flag for all slaves; "
 183                                      "0 for never (default), 1 for always.");
 184 module_param(resend_igmp, int, 0);
 185 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 186                               "link failure");
 187 module_param(packets_per_slave, int, 0);
 188 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
 189                                     "mode; 0 for a random slave, 1 packet per "
 190                                     "slave (default), >1 packets per slave.");
 191 module_param(lp_interval, uint, 0);
 192 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
 193                               "the bonding driver sends learning packets to "
 194                               "each slaves peer switch. The default is 1.");
 195
 196 /*----------------------------- Global variables ----------------------------*/
 197
 198 #ifdef CONFIG_NET_POLL_CONTROLLER
 199 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 200 #endif
 201
 202 int bond_net_id __read_mostly;
 203
 204 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 205 static int arp_ip_count;
 206 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 207 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 208 static int lacp_fast;
 209
 210 /*-------------------------- Forward declarations ---------------------------*/
 211
 212 static int bond_init(struct net_device *bond_dev);
 213 static void bond_uninit(struct net_device *bond_dev);
 214 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
 215                                                 struct rtnl_link_stats64 *stats);
 216 static void bond_slave_arr_handler(struct work_struct *work);
 217 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
 218                                   int mod);
 219
 220 /*---------------------------- General routines -----------------------------*/
 221
 222 const char *bond_mode_name(int mode)
 223 {
 224         static const char *names[] = {
 225                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 226                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 227                 [BOND_MODE_XOR] = "load balancing (xor)",
 228                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 229                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 230                 [BOND_MODE_TLB] = "transmit load balancing",
 231                 [BOND_MODE_ALB] = "adaptive load balancing",
 232         };
 233
 234         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 235                 return "unknown";
 236
 237         return names[mode];
 238 }
 239
 240 /*---------------------------------- VLAN -----------------------------------*/
 241
 242 /**
 243  * bond_dev_queue_xmit - Prepare skb for xmit.
 244  *
 245  * @bond: bond device that got this skb for tx.
 246  * @skb: hw accel VLAN tagged skb to transmit
 247  * @slave_dev: slave that is supposed to xmit this skbuff
 248  */
 249 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 250                         struct net_device *slave_dev)
 251 {
 252         skb->dev = slave_dev;
 253
 254         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 255                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 256         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 257
 258         if (unlikely(netpoll_tx_running(bond->dev)))
 259                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 260         else
 261                 dev_queue_xmit(skb);
 262 }
 263
 264 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 265  * We don't protect the slave list iteration with a lock because:
 266  * a. This operation is performed in IOCTL context,
 267  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 268  * c. Holding a lock with BH disabled while directly calling a base driver
 269  *    entry point is generally a BAD idea.
 270  *
 271  * The design of synchronization/protection for this operation in the 8021q
 272  * module is good for one or more VLAN devices over a single physical device
 273  * and cannot be extended for a teaming solution like bonding, so there is a
 274  * potential race condition here where a net device from the vlan group might
 275  * be referenced (either by a base driver or the 8021q code) while it is being
 276  * removed from the system. However, it turns out we're not making matters
 277  * worse, and if it works for regular VLAN usage it will work here too.
 278 */
 279
 280 /**
 281  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 282  * @bond_dev: bonding net device that got called
 283  * @vid: vlan id being added
 284  */
 285 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 286                                 __be16 proto, u16 vid)
 287 {
 288         struct bonding *bond = netdev_priv(bond_dev);
 289         struct slave *slave, *rollback_slave;
 290         struct list_head *iter;
 291         int res;
 292
 293         bond_for_each_slave(bond, slave, iter) {
 294                 res = vlan_vid_add(slave->dev, proto, vid);
 295                 if (res)
 296                         goto unwind;
 297         }
 298
 299         return 0;
 300
 301 unwind:
 302         /* unwind to the slave that failed */
 303         bond_for_each_slave(bond, rollback_slave, iter) {
 304                 if (rollback_slave == slave)
 305                         break;
 306
 307                 vlan_vid_del(rollback_slave->dev, proto, vid);
 308         }
 309
 310         return res;
 311 }
 312
 313 /**
 314  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 315  * @bond_dev: bonding net device that got called
 316  * @vid: vlan id being removed
 317  */
 318 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 319                                  __be16 proto, u16 vid)
 320 {
 321         struct bonding *bond = netdev_priv(bond_dev);
 322         struct list_head *iter;
 323         struct slave *slave;
 324
 325         bond_for_each_slave(bond, slave, iter)
 326                 vlan_vid_del(slave->dev, proto, vid);
 327
 328         if (bond_is_lb(bond))
 329                 bond_alb_clear_vlan(bond, vid);
 330
 331         return 0;
 332 }
 333
 334 /*------------------------------- Link status -------------------------------*/
 335
 336 /* Set the carrier state for the master according to the state of its
 337  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 338  * do special 802.3ad magic.
 339  *
 340  * Returns zero if carrier state does not change, nonzero if it does.
 341  */
 342 int bond_set_carrier(struct bonding *bond)
 343 {
 344         struct list_head *iter;
 345         struct slave *slave;
 346
 347         if (!bond_has_slaves(bond))
 348                 goto down;
 349
 350         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 351                 return bond_3ad_set_carrier(bond);
 352
 353         bond_for_each_slave(bond, slave, iter) {
 354                 if (slave->link == BOND_LINK_UP) {
 355                         if (!netif_carrier_ok(bond->dev)) {
 356                                 netif_carrier_on(bond->dev);
 357                                 return 1;
 358                         }
 359                         return 0;
 360                 }
 361         }
 362
 363 down:
 364         if (netif_carrier_ok(bond->dev)) {
 365                 netif_carrier_off(bond->dev);
 366                 return 1;
 367         }
 368         return 0;
 369 }
 370
 371 /* Get link speed and duplex from the slave's base driver
 372  * using ethtool. If for some reason the call fails or the
 373  * values are invalid, set speed and duplex to -1,
 374  * and return.
 375  */
 376 static void bond_update_speed_duplex(struct slave *slave)
 377 {
 378         struct net_device *slave_dev = slave->dev;
 379         struct ethtool_link_ksettings ecmd;
 380         int res;
 381
 382         slave->speed = SPEED_UNKNOWN;
 383         slave->duplex = DUPLEX_UNKNOWN;
 384
 385         res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
 386         if (res < 0)
 387                 return;
 388
 389         if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
 390                 return;
 391
 392         switch (ecmd.base.duplex) {
 393         case DUPLEX_FULL:
 394         case DUPLEX_HALF:
 395                 break;
 396         default:
 397                 return;
 398         }
 399
 400         slave->speed = ecmd.base.speed;
 401         slave->duplex = ecmd.base.duplex;
 402
 403         return;
 404 }
 405
 406 const char *bond_slave_link_status(s8 link)
 407 {
 408         switch (link) {
 409         case BOND_LINK_UP:
 410                 return "up";
 411         case BOND_LINK_FAIL:
 412                 return "going down";
 413         case BOND_LINK_DOWN:
 414                 return "down";
 415         case BOND_LINK_BACK:
 416                 return "going back";
 417         default:
 418                 return "unknown";
 419         }
 420 }
 421
 422 /* if <dev> supports MII link status reporting, check its link status.
 423  *
 424  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 425  * depending upon the setting of the use_carrier parameter.
 426  *
 427  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 428  * can't tell and just pretend it is), or 0, meaning that the link is
 429  * down.
 430  *
 431  * If reporting is non-zero, instead of faking link up, return -1 if
 432  * both ETHTOOL and MII ioctls fail (meaning the device does not
 433  * support them).  If use_carrier is set, return whatever it says.
 434  * It'd be nice if there was a good way to tell if a driver supports
 435  * netif_carrier, but there really isn't.
 436  */
 437 static int bond_check_dev_link(struct bonding *bond,
 438                                struct net_device *slave_dev, int reporting)
 439 {
 440         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 441         int (*ioctl)(struct net_device *, struct ifreq *, int);
 442         struct ifreq ifr;
 443         struct mii_ioctl_data *mii;
 444
 445         if (!reporting && !netif_running(slave_dev))
 446                 return 0;
 447
 448         if (bond->params.use_carrier)
 449                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 450
 451         /* Try to get link status using Ethtool first. */
 452         if (slave_dev->ethtool_ops->get_link)
 453                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 454                         BMSR_LSTATUS : 0;
 455
 456         /* Ethtool can't be used, fallback to MII ioctls. */
 457         ioctl = slave_ops->ndo_do_ioctl;
 458         if (ioctl) {
 459                 /* TODO: set pointer to correct ioctl on a per team member
 460                  *       bases to make this more efficient. that is, once
 461                  *       we determine the correct ioctl, we will always
 462                  *       call it and not the others for that team
 463                  *       member.
 464                  */
 465
 466                 /* We cannot assume that SIOCGMIIPHY will also read a
 467                  * register; not all network drivers (e.g., e100)
 468                  * support that.
 469                  */
 470
 471                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 472                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 473                 mii = if_mii(&ifr);
 474                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 475                         mii->reg_num = MII_BMSR;
 476                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 477                                 return mii->val_out & BMSR_LSTATUS;
 478                 }
 479         }
 480
 481         /* If reporting, report that either there's no dev->do_ioctl,
 482          * or both SIOCGMIIREG and get_link failed (meaning that we
 483          * cannot report link status).  If not reporting, pretend
 484          * we're ok.
 485          */
 486         return reporting ? -1 : BMSR_LSTATUS;
 487 }
 488
 489 /*----------------------------- Multicast list ------------------------------*/
 490
 491 /* Push the promiscuity flag down to appropriate slaves */
 492 static int bond_set_promiscuity(struct bonding *bond, int inc)
 493 {
 494         struct list_head *iter;
 495         int err = 0;
 496
 497         if (bond_uses_primary(bond)) {
 498                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 499
 500                 if (curr_active)
 501                         err = dev_set_promiscuity(curr_active->dev, inc);
 502         } else {
 503                 struct slave *slave;
 504
 505                 bond_for_each_slave(bond, slave, iter) {
 506                         err = dev_set_promiscuity(slave->dev, inc);
 507                         if (err)
 508                                 return err;
 509                 }
 510         }
 511         return err;
 512 }
 513
 514 /* Push the allmulti flag down to all slaves */
 515 static int bond_set_allmulti(struct bonding *bond, int inc)
 516 {
 517         struct list_head *iter;
 518         int err = 0;
 519
 520         if (bond_uses_primary(bond)) {
 521                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 522
 523                 if (curr_active)
 524                         err = dev_set_allmulti(curr_active->dev, inc);
 525         } else {
 526                 struct slave *slave;
 527
 528                 bond_for_each_slave(bond, slave, iter) {
 529                         err = dev_set_allmulti(slave->dev, inc);
 530                         if (err)
 531                                 return err;
 532                 }
 533         }
 534         return err;
 535 }
 536
 537 /* Retrieve the list of registered multicast addresses for the bonding
 538  * device and retransmit an IGMP JOIN request to the current active
 539  * slave.
 540  */
 541 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 542 {
 543         struct bonding *bond = container_of(work, struct bonding,
 544                                             mcast_work.work);
 545
 546         if (!rtnl_trylock()) {
 547                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 548                 return;
 549         }
 550         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 551
 552         if (bond->igmp_retrans > 1) {
 553                 bond->igmp_retrans--;
 554                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 555         }
 556         rtnl_unlock();
 557 }
 558
 559 /* Flush bond's hardware addresses from slave */
 560 static void bond_hw_addr_flush(struct net_device *bond_dev,
 561                                struct net_device *slave_dev)
 562 {
 563         struct bonding *bond = netdev_priv(bond_dev);
 564
 565         dev_uc_unsync(slave_dev, bond_dev);
 566         dev_mc_unsync(slave_dev, bond_dev);
 567
 568         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 569                 /* del lacpdu mc addr from mc list */
 570                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 571
 572                 dev_mc_del(slave_dev, lacpdu_multicast);
 573         }
 574 }
 575
 576 /*--------------------------- Active slave change ---------------------------*/
 577
 578 /* Update the hardware address list and promisc/allmulti for the new and
 579  * old active slaves (if any).  Modes that are not using primary keep all
 580  * slaves up date at all times; only the modes that use primary need to call
 581  * this function to swap these settings during a failover.
 582  */
 583 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 584                               struct slave *old_active)
 585 {
 586         if (old_active) {
 587                 if (bond->dev->flags & IFF_PROMISC)
 588                         dev_set_promiscuity(old_active->dev, -1);
 589
 590                 if (bond->dev->flags & IFF_ALLMULTI)
 591                         dev_set_allmulti(old_active->dev, -1);
 592
 593                 bond_hw_addr_flush(bond->dev, old_active->dev);
 594         }
 595
 596         if (new_active) {
 597                 /* FIXME: Signal errors upstream. */
 598                 if (bond->dev->flags & IFF_PROMISC)
 599                         dev_set_promiscuity(new_active->dev, 1);
 600
 601                 if (bond->dev->flags & IFF_ALLMULTI)
 602                         dev_set_allmulti(new_active->dev, 1);
 603
 604                 netif_addr_lock_bh(bond->dev);
 605                 dev_uc_sync(new_active->dev, bond->dev);
 606                 dev_mc_sync(new_active->dev, bond->dev);
 607                 netif_addr_unlock_bh(bond->dev);
 608         }
 609 }
 610
 611 /**
 612  * bond_set_dev_addr - clone slave's address to bond
 613  * @bond_dev: bond net device
 614  * @slave_dev: slave net device
 615  *
 616  * Should be called with RTNL held.
 617  */
 618 static void bond_set_dev_addr(struct net_device *bond_dev,
 619                               struct net_device *slave_dev)
 620 {
 621         netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n",
 622                    bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len);
 623         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 624         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 625         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 626 }
 627
 628 static struct slave *bond_get_old_active(struct bonding *bond,
 629                                          struct slave *new_active)
 630 {
 631         struct slave *slave;
 632         struct list_head *iter;
 633
 634         bond_for_each_slave(bond, slave, iter) {
 635                 if (slave == new_active)
 636                         continue;
 637
 638                 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
 639                         return slave;
 640         }
 641
 642         return NULL;
 643 }
 644
 645 /* bond_do_fail_over_mac
 646  *
 647  * Perform special MAC address swapping for fail_over_mac settings
 648  *
 649  * Called with RTNL
 650  */
 651 static void bond_do_fail_over_mac(struct bonding *bond,
 652                                   struct slave *new_active,
 653                                   struct slave *old_active)
 654 {
 655         u8 tmp_mac[ETH_ALEN];
 656         struct sockaddr saddr;
 657         int rv;
 658
 659         switch (bond->params.fail_over_mac) {
 660         case BOND_FOM_ACTIVE:
 661                 if (new_active)
 662                         bond_set_dev_addr(bond->dev, new_active->dev);
 663                 break;
 664         case BOND_FOM_FOLLOW:
 665                 /* if new_active && old_active, swap them
 666                  * if just old_active, do nothing (going to no active slave)
 667                  * if just new_active, set new_active to bond's MAC
 668                  */
 669                 if (!new_active)
 670                         return;
 671
 672                 if (!old_active)
 673                         old_active = bond_get_old_active(bond, new_active);
 674
 675                 if (old_active) {
 676                         ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
 677                         ether_addr_copy(saddr.sa_data,
 678                                         old_active->dev->dev_addr);
 679                         saddr.sa_family = new_active->dev->type;
 680                 } else {
 681                         ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
 682                         saddr.sa_family = bond->dev->type;
 683                 }
 684
 685                 rv = dev_set_mac_address(new_active->dev, &saddr);
 686                 if (rv) {
 687                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 688                                    -rv, new_active->dev->name);
 689                         goto out;
 690                 }
 691
 692                 if (!old_active)
 693                         goto out;
 694
 695                 ether_addr_copy(saddr.sa_data, tmp_mac);
 696                 saddr.sa_family = old_active->dev->type;
 697
 698                 rv = dev_set_mac_address(old_active->dev, &saddr);
 699                 if (rv)
 700                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 701                                    -rv, new_active->dev->name);
 702 out:
 703                 break;
 704         default:
 705                 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
 706                            bond->params.fail_over_mac);
 707                 break;
 708         }
 709
 710 }
 711
 712 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
 713 {
 714         struct slave *prim = rtnl_dereference(bond->primary_slave);
 715         struct slave *curr = rtnl_dereference(bond->curr_active_slave);
 716
 717         if (!prim || prim->link != BOND_LINK_UP) {
 718                 if (!curr || curr->link != BOND_LINK_UP)
 719                         return NULL;
 720                 return curr;
 721         }
 722
 723         if (bond->force_primary) {
 724                 bond->force_primary = false;
 725                 return prim;
 726         }
 727
 728         if (!curr || curr->link != BOND_LINK_UP)
 729                 return prim;
 730
 731         /* At this point, prim and curr are both up */
 732         switch (bond->params.primary_reselect) {
 733         case BOND_PRI_RESELECT_ALWAYS:
 734                 return prim;
 735         case BOND_PRI_RESELECT_BETTER:
 736                 if (prim->speed < curr->speed)
 737                         return curr;
 738                 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
 739                         return curr;
 740                 return prim;
 741         case BOND_PRI_RESELECT_FAILURE:
 742                 return curr;
 743         default:
 744                 netdev_err(bond->dev, "impossible primary_reselect %d\n",
 745                            bond->params.primary_reselect);
 746                 return curr;
 747         }
 748 }
 749
 750 /**
 751  * bond_find_best_slave - select the best available slave to be the active one
 752  * @bond: our bonding struct
 753  */
 754 static struct slave *bond_find_best_slave(struct bonding *bond)
 755 {
 756         struct slave *slave, *bestslave = NULL;
 757         struct list_head *iter;
 758         int mintime = bond->params.updelay;
 759
 760         slave = bond_choose_primary_or_current(bond);
 761         if (slave)
 762                 return slave;
 763
 764         bond_for_each_slave(bond, slave, iter) {
 765                 if (slave->link == BOND_LINK_UP)
 766                         return slave;
 767                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
 768                     slave->delay < mintime) {
 769                         mintime = slave->delay;
 770                         bestslave = slave;
 771                 }
 772         }
 773
 774         return bestslave;
 775 }
 776
 777 static bool bond_should_notify_peers(struct bonding *bond)
 778 {
 779         struct slave *slave;
 780
 781         rcu_read_lock();
 782         slave = rcu_dereference(bond->curr_active_slave);
 783         rcu_read_unlock();
 784
 785         netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
 786                    slave ? slave->dev->name : "NULL");
 787
 788         if (!slave || !bond->send_peer_notif ||
 789             !netif_carrier_ok(bond->dev) ||
 790             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 791                 return false;
 792
 793         return true;
 794 }
 795
 796 /**
 797  * change_active_interface - change the active slave into the specified one
 798  * @bond: our bonding struct
 799  * @new: the new slave to make the active one
 800  *
 801  * Set the new slave to the bond's settings and unset them on the old
 802  * curr_active_slave.
 803  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 804  *
 805  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 806  * because it is apparently the best available slave we have, even though its
 807  * updelay hasn't timed out yet.
 808  *
 809  * Caller must hold RTNL.
 810  */
 811 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 812 {
 813         struct slave *old_active;
 814
 815         ASSERT_RTNL();
 816
 817         old_active = rtnl_dereference(bond->curr_active_slave);
 818
 819         if (old_active == new_active)
 820                 return;
 821
 822         if (new_active) {
 823                 new_active->last_link_up = jiffies;
 824
 825                 if (new_active->link == BOND_LINK_BACK) {
 826                         if (bond_uses_primary(bond)) {
 827                                 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
 828                                             new_active->dev->name,
 829                                             (bond->params.updelay - new_active->delay) * bond->params.miimon);
 830                         }
 831
 832                         new_active->delay = 0;
 833                         bond_set_slave_link_state(new_active, BOND_LINK_UP,
 834                                                   BOND_SLAVE_NOTIFY_NOW);
 835
 836                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 837                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 838
 839                         if (bond_is_lb(bond))
 840                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 841                 } else {
 842                         if (bond_uses_primary(bond)) {
 843                                 netdev_info(bond->dev, "making interface %s the new active one\n",
 844                                             new_active->dev->name);
 845                         }
 846                 }
 847         }
 848
 849         if (bond_uses_primary(bond))
 850                 bond_hw_addr_swap(bond, new_active, old_active);
 851
 852         if (bond_is_lb(bond)) {
 853                 bond_alb_handle_active_change(bond, new_active);
 854                 if (old_active)
 855                         bond_set_slave_inactive_flags(old_active,
 856                                                       BOND_SLAVE_NOTIFY_NOW);
 857                 if (new_active)
 858                         bond_set_slave_active_flags(new_active,
 859                                                     BOND_SLAVE_NOTIFY_NOW);
 860         } else {
 861                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 862         }
 863
 864         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
 865                 if (old_active)
 866                         bond_set_slave_inactive_flags(old_active,
 867                                                       BOND_SLAVE_NOTIFY_NOW);
 868
 869                 if (new_active) {
 870                         bool should_notify_peers = false;
 871
 872                         bond_set_slave_active_flags(new_active,
 873                                                     BOND_SLAVE_NOTIFY_NOW);
 874
 875                         if (bond->params.fail_over_mac)
 876                                 bond_do_fail_over_mac(bond, new_active,
 877                                                       old_active);
 878
 879                         if (netif_running(bond->dev)) {
 880                                 bond->send_peer_notif =
 881                                         bond->params.num_peer_notif;
 882                                 should_notify_peers =
 883                                         bond_should_notify_peers(bond);
 884                         }
 885
 886                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 887                         if (should_notify_peers)
 888                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 889                                                          bond->dev);
 890                 }
 891         }
 892
 893         /* resend IGMP joins since active slave has changed or
 894          * all were sent on curr_active_slave.
 895          * resend only if bond is brought up with the affected
 896          * bonding modes and the retransmission is enabled
 897          */
 898         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 899             ((bond_uses_primary(bond) && new_active) ||
 900              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
 901                 bond->igmp_retrans = bond->params.resend_igmp;
 902                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 903         }
 904 }
 905
 906 /**
 907  * bond_select_active_slave - select a new active slave, if needed
 908  * @bond: our bonding struct
 909  *
 910  * This functions should be called when one of the following occurs:
 911  * - The old curr_active_slave has been released or lost its link.
 912  * - The primary_slave has got its link back.
 913  * - A slave has got its link back and there's no old curr_active_slave.
 914  *
 915  * Caller must hold RTNL.
 916  */
 917 void bond_select_active_slave(struct bonding *bond)
 918 {
 919         struct slave *best_slave;
 920         int rv;
 921
 922         ASSERT_RTNL();
 923
 924         best_slave = bond_find_best_slave(bond);
 925         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
 926                 bond_change_active_slave(bond, best_slave);
 927                 rv = bond_set_carrier(bond);
 928                 if (!rv)
 929                         return;
 930
 931                 if (netif_carrier_ok(bond->dev))
 932                         netdev_info(bond->dev, "first active interface up!\n");
 933                 else
 934                         netdev_info(bond->dev, "now running without any active interface!\n");
 935         }
 936 }
 937
 938 #ifdef CONFIG_NET_POLL_CONTROLLER
 939 static inline int slave_enable_netpoll(struct slave *slave)
 940 {
 941         struct netpoll *np;
 942         int err = 0;
 943
 944         np = kzalloc(sizeof(*np), GFP_KERNEL);
 945         err = -ENOMEM;
 946         if (!np)
 947                 goto out;
 948
 949         err = __netpoll_setup(np, slave->dev);
 950         if (err) {
 951                 kfree(np);
 952                 goto out;
 953         }
 954         slave->np = np;
 955 out:
 956         return err;
 957 }
 958 static inline void slave_disable_netpoll(struct slave *slave)
 959 {
 960         struct netpoll *np = slave->np;
 961
 962         if (!np)
 963                 return;
 964
 965         slave->np = NULL;
 966         __netpoll_free_async(np);
 967 }
 968
 969 static void bond_poll_controller(struct net_device *bond_dev)
 970 {
 971         struct bonding *bond = netdev_priv(bond_dev);
 972         struct slave *slave = NULL;
 973         struct list_head *iter;
 974         struct ad_info ad_info;
 975         struct netpoll_info *ni;
 976         const struct net_device_ops *ops;
 977
 978         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 979                 if (bond_3ad_get_active_agg_info(bond, &ad_info))
 980                         return;
 981
 982         bond_for_each_slave_rcu(bond, slave, iter) {
 983                 ops = slave->dev->netdev_ops;
 984                 if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller)
 985                         continue;
 986
 987                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 988                         struct aggregator *agg =
 989                             SLAVE_AD_INFO(slave)->port.aggregator;
 990
 991                         if (agg &&
 992                             agg->aggregator_identifier != ad_info.aggregator_id)
 993                                 continue;
 994                 }
 995
 996                 ni = rcu_dereference_bh(slave->dev->npinfo);
 997                 if (down_trylock(&ni->dev_lock))
 998                         continue;
 999                 ops->ndo_poll_controller(slave->dev);
1000                 up(&ni->dev_lock);
1001         }
1002 }
1003
1004 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1005 {
1006         struct bonding *bond = netdev_priv(bond_dev);
1007         struct list_head *iter;
1008         struct slave *slave;
1009
1010         bond_for_each_slave(bond, slave, iter)
1011                 if (bond_slave_is_up(slave))
1012                         slave_disable_netpoll(slave);
1013 }
1014
1015 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1016 {
1017         struct bonding *bond = netdev_priv(dev);
1018         struct list_head *iter;
1019         struct slave *slave;
1020         int err = 0;
1021
1022         bond_for_each_slave(bond, slave, iter) {
1023                 err = slave_enable_netpoll(slave);
1024                 if (err) {
1025                         bond_netpoll_cleanup(dev);
1026                         break;
1027                 }
1028         }
1029         return err;
1030 }
1031 #else
1032 static inline int slave_enable_netpoll(struct slave *slave)
1033 {
1034         return 0;
1035 }
1036 static inline void slave_disable_netpoll(struct slave *slave)
1037 {
1038 }
1039 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1040 {
1041 }
1042 #endif
1043
1044 /*---------------------------------- IOCTL ----------------------------------*/
1045
1046 static netdev_features_t bond_fix_features(struct net_device *dev,
1047                                            netdev_features_t features)
1048 {
1049         struct bonding *bond = netdev_priv(dev);
1050         struct list_head *iter;
1051         netdev_features_t mask;
1052         struct slave *slave;
1053
1054         mask = features;
1055
1056         features &= ~NETIF_F_ONE_FOR_ALL;
1057         features |= NETIF_F_ALL_FOR_ALL;
1058
1059         bond_for_each_slave(bond, slave, iter) {
1060                 features = netdev_increment_features(features,
1061                                                      slave->dev->features,
1062                                                      mask);
1063         }
1064         features = netdev_add_tso_features(features, mask);
1065
1066         return features;
1067 }
1068
1069 #define BOND_VLAN_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1070                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1071                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1072
1073 #define BOND_ENC_FEATURES       (NETIF_F_HW_CSUM | NETIF_F_SG | \
1074                                  NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1075
1076 static void bond_compute_features(struct bonding *bond)
1077 {
1078         unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1079                                         IFF_XMIT_DST_RELEASE_PERM;
1080         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1081         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1082         struct net_device *bond_dev = bond->dev;
1083         struct list_head *iter;
1084         struct slave *slave;
1085         unsigned short max_hard_header_len = ETH_HLEN;
1086         unsigned int gso_max_size = GSO_MAX_SIZE;
1087         u16 gso_max_segs = GSO_MAX_SEGS;
1088
1089         if (!bond_has_slaves(bond))
1090                 goto done;
1091         vlan_features &= NETIF_F_ALL_FOR_ALL;
1092
1093         bond_for_each_slave(bond, slave, iter) {
1094                 vlan_features = netdev_increment_features(vlan_features,
1095                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1096
1097                 enc_features = netdev_increment_features(enc_features,
1098                                                          slave->dev->hw_enc_features,
1099                                                          BOND_ENC_FEATURES);
1100                 dst_release_flag &= slave->dev->priv_flags;
1101                 if (slave->dev->hard_header_len > max_hard_header_len)
1102                         max_hard_header_len = slave->dev->hard_header_len;
1103
1104                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1105                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1106         }
1107
1108 done:
1109         bond_dev->vlan_features = vlan_features;
1110         bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
1111         bond_dev->hard_header_len = max_hard_header_len;
1112         bond_dev->gso_max_segs = gso_max_segs;
1113         netif_set_gso_max_size(bond_dev, gso_max_size);
1114
1115         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1116         if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1117             dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1118                 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1119
1120         netdev_change_features(bond_dev);
1121 }
1122
1123 static void bond_setup_by_slave(struct net_device *bond_dev,
1124                                 struct net_device *slave_dev)
1125 {
1126         bond_dev->header_ops        = slave_dev->header_ops;
1127
1128         bond_dev->type              = slave_dev->type;
1129         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1130         bond_dev->addr_len          = slave_dev->addr_len;
1131
1132         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1133                 slave_dev->addr_len);
1134 }
1135
1136 /* On bonding slaves other than the currently active slave, suppress
1137  * duplicates except for alb non-mcast/bcast.
1138  */
1139 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1140                                             struct slave *slave,
1141                                             struct bonding *bond)
1142 {
1143         if (bond_is_slave_inactive(slave)) {
1144                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1145                     skb->pkt_type != PACKET_BROADCAST &&
1146                     skb->pkt_type != PACKET_MULTICAST)
1147                         return false;
1148                 return true;
1149         }
1150         return false;
1151 }
1152
1153 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1154 {
1155         struct sk_buff *skb = *pskb;
1156         struct slave *slave;
1157         struct bonding *bond;
1158         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1159                           struct slave *);
1160         int ret = RX_HANDLER_ANOTHER;
1161
1162         skb = skb_share_check(skb, GFP_ATOMIC);
1163         if (unlikely(!skb))
1164                 return RX_HANDLER_CONSUMED;
1165
1166         *pskb = skb;
1167
1168         slave = bond_slave_get_rcu(skb->dev);
1169         bond = slave->bond;
1170
1171         recv_probe = ACCESS_ONCE(bond->recv_probe);
1172         if (recv_probe) {
1173                 ret = recv_probe(skb, bond, slave);
1174                 if (ret == RX_HANDLER_CONSUMED) {
1175                         consume_skb(skb);
1176                         return ret;
1177                 }
1178         }
1179
1180         if (bond_should_deliver_exact_match(skb, slave, bond))
1181                 return RX_HANDLER_EXACT;
1182
1183         skb->dev = bond->dev;
1184
1185         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1186             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1187             skb->pkt_type == PACKET_HOST) {
1188
1189                 if (unlikely(skb_cow_head(skb,
1190                                           skb->data - skb_mac_header(skb)))) {
1191                         kfree_skb(skb);
1192                         return RX_HANDLER_CONSUMED;
1193                 }
1194                 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1195         }
1196
1197         return ret;
1198 }
1199
1200 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1201 {
1202         switch (BOND_MODE(bond)) {
1203         case BOND_MODE_ROUNDROBIN:
1204                 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1205         case BOND_MODE_ACTIVEBACKUP:
1206                 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1207         case BOND_MODE_BROADCAST:
1208                 return NETDEV_LAG_TX_TYPE_BROADCAST;
1209         case BOND_MODE_XOR:
1210         case BOND_MODE_8023AD:
1211                 return NETDEV_LAG_TX_TYPE_HASH;
1212         default:
1213                 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1214         }
1215 }
1216
1217 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave)
1218 {
1219         struct netdev_lag_upper_info lag_upper_info;
1220         int err;
1221
1222         lag_upper_info.tx_type = bond_lag_tx_type(bond);
1223         err = netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1224                                            &lag_upper_info);
1225         if (err)
1226                 return err;
1227         rtmsg_ifinfo(RTM_NEWLINK, slave->dev, IFF_SLAVE, GFP_KERNEL);
1228         return 0;
1229 }
1230
1231 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1232 {
1233         netdev_upper_dev_unlink(slave->dev, bond->dev);
1234         slave->dev->flags &= ~IFF_SLAVE;
1235         rtmsg_ifinfo(RTM_NEWLINK, slave->dev, IFF_SLAVE, GFP_KERNEL);
1236 }
1237
1238 static struct slave *bond_alloc_slave(struct bonding *bond)
1239 {
1240         struct slave *slave = NULL;
1241
1242         slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1243         if (!slave)
1244                 return NULL;
1245
1246         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1247                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1248                                                GFP_KERNEL);
1249                 if (!SLAVE_AD_INFO(slave)) {
1250                         kfree(slave);
1251                         return NULL;
1252                 }
1253         }
1254         return slave;
1255 }
1256
1257 static void bond_free_slave(struct slave *slave)
1258 {
1259         struct bonding *bond = bond_get_bond_by_slave(slave);
1260
1261         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1262                 kfree(SLAVE_AD_INFO(slave));
1263
1264         kfree(slave);
1265 }
1266
1267 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1268 {
1269         info->bond_mode = BOND_MODE(bond);
1270         info->miimon = bond->params.miimon;
1271         info->num_slaves = bond->slave_cnt;
1272 }
1273
1274 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1275 {
1276         strcpy(info->slave_name, slave->dev->name);
1277         info->link = slave->link;
1278         info->state = bond_slave_state(slave);
1279         info->link_failure_count = slave->link_failure_count;
1280 }
1281
1282 static void bond_netdev_notify(struct net_device *dev,
1283                                struct netdev_bonding_info *info)
1284 {
1285         rtnl_lock();
1286         netdev_bonding_info_change(dev, info);
1287         rtnl_unlock();
1288 }
1289
1290 static void bond_netdev_notify_work(struct work_struct *_work)
1291 {
1292         struct netdev_notify_work *w =
1293                 container_of(_work, struct netdev_notify_work, work.work);
1294
1295         bond_netdev_notify(w->dev, &w->bonding_info);
1296         dev_put(w->dev);
1297         kfree(w);
1298 }
1299
1300 void bond_queue_slave_event(struct slave *slave)
1301 {
1302         struct bonding *bond = slave->bond;
1303         struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1304
1305         if (!nnw)
1306                 return;
1307
1308         dev_hold(slave->dev);
1309         nnw->dev = slave->dev;
1310         bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1311         bond_fill_ifbond(bond, &nnw->bonding_info.master);
1312         INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1313
1314         queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1315 }
1316
1317 void bond_lower_state_changed(struct slave *slave)
1318 {
1319         struct netdev_lag_lower_state_info info;
1320
1321         info.link_up = slave->link == BOND_LINK_UP ||
1322                        slave->link == BOND_LINK_FAIL;
1323         info.tx_enabled = bond_is_active_slave(slave);
1324         netdev_lower_state_changed(slave->dev, &info);
1325 }
1326
1327 /* enslave device <slave> to bond device <master> */
1328 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1329 {
1330         struct bonding *bond = netdev_priv(bond_dev);
1331         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1332         struct slave *new_slave = NULL, *prev_slave;
1333         struct sockaddr addr;
1334         int link_reporting;
1335         int res = 0, i;
1336
1337         if (!bond->params.use_carrier &&
1338             slave_dev->ethtool_ops->get_link == NULL &&
1339             slave_ops->ndo_do_ioctl == NULL) {
1340                 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1341                             slave_dev->name);
1342         }
1343
1344         /* already enslaved */
1345         if (slave_dev->flags & IFF_SLAVE) {
1346                 netdev_dbg(bond_dev, "Error: Device was already enslaved\n");
1347                 return -EBUSY;
1348         }
1349
1350         if (bond_dev == slave_dev) {
1351                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1352                 return -EPERM;
1353         }
1354
1355         /* vlan challenged mutual exclusion */
1356         /* no need to lock since we're protected by rtnl_lock */
1357         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1358                 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1359                            slave_dev->name);
1360                 if (vlan_uses_dev(bond_dev)) {
1361                         netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1362                                    slave_dev->name, bond_dev->name);
1363                         return -EPERM;
1364                 } else {
1365                         netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1366                                     slave_dev->name, slave_dev->name,
1367                                     bond_dev->name);
1368                 }
1369         } else {
1370                 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1371                            slave_dev->name);
1372         }
1373
1374         /* Old ifenslave binaries are no longer supported.  These can
1375          * be identified with moderate accuracy by the state of the slave:
1376          * the current ifenslave will set the interface down prior to
1377          * enslaving it; the old ifenslave will not.
1378          */
1379         if (slave_dev->flags & IFF_UP) {
1380                 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1381                            slave_dev->name);
1382                 return -EPERM;
1383         }
1384
1385         /* set bonding device ether type by slave - bonding netdevices are
1386          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1387          * there is a need to override some of the type dependent attribs/funcs.
1388          *
1389          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1390          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1391          */
1392         if (!bond_has_slaves(bond)) {
1393                 if (bond_dev->type != slave_dev->type) {
1394                         netdev_dbg(bond_dev, "change device type from %d to %d\n",
1395                                    bond_dev->type, slave_dev->type);
1396
1397                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1398                                                        bond_dev);
1399                         res = notifier_to_errno(res);
1400                         if (res) {
1401                                 netdev_err(bond_dev, "refused to change device type\n");
1402                                 return -EBUSY;
1403                         }
1404
1405                         /* Flush unicast and multicast addresses */
1406                         dev_uc_flush(bond_dev);
1407                         dev_mc_flush(bond_dev);
1408
1409                         if (slave_dev->type != ARPHRD_ETHER)
1410                                 bond_setup_by_slave(bond_dev, slave_dev);
1411                         else {
1412                                 ether_setup(bond_dev);
1413                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1414                         }
1415
1416                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1417                                                  bond_dev);
1418                 }
1419         } else if (bond_dev->type != slave_dev->type) {
1420                 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1421                            slave_dev->name, slave_dev->type, bond_dev->type);
1422                 return -EINVAL;
1423         }
1424
1425         if (slave_dev->type == ARPHRD_INFINIBAND &&
1426             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1427                 netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n",
1428                             slave_dev->type);
1429                 res = -EOPNOTSUPP;
1430                 goto err_undo_flags;
1431         }
1432
1433         if (!slave_ops->ndo_set_mac_address ||
1434             slave_dev->type == ARPHRD_INFINIBAND) {
1435                 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1436                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1437                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1438                         if (!bond_has_slaves(bond)) {
1439                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1440                                 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1441                         } else {
1442                                 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");
1443                                 res = -EOPNOTSUPP;
1444                                 goto err_undo_flags;
1445                         }
1446                 }
1447         }
1448
1449         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1450
1451         /* If this is the first slave, then we need to set the master's hardware
1452          * address to be the same as the slave's.
1453          */
1454         if (!bond_has_slaves(bond) &&
1455             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1456                 bond_set_dev_addr(bond->dev, slave_dev);
1457
1458         new_slave = bond_alloc_slave(bond);
1459         if (!new_slave) {
1460                 res = -ENOMEM;
1461                 goto err_undo_flags;
1462         }
1463
1464         new_slave->bond = bond;
1465         new_slave->dev = slave_dev;
1466         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1467          * is set via sysfs or module option if desired.
1468          */
1469         new_slave->queue_id = 0;
1470
1471         /* Save slave's original mtu and then set it to match the bond */
1472         new_slave->original_mtu = slave_dev->mtu;
1473         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1474         if (res) {
1475                 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1476                 goto err_free;
1477         }
1478
1479         /* Save slave's original ("permanent") mac address for modes
1480          * that need it, and for restoring it upon release, and then
1481          * set it to the master's address
1482          */
1483         ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1484
1485         if (!bond->params.fail_over_mac ||
1486             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1487                 /* Set slave to master's mac address.  The application already
1488                  * set the master's mac address to that of the first slave
1489                  */
1490                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1491                 addr.sa_family = slave_dev->type;
1492                 res = dev_set_mac_address(slave_dev, &addr);
1493                 if (res) {
1494                         netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1495                         goto err_restore_mtu;
1496                 }
1497         }
1498
1499         /* set slave flag before open to prevent IPv6 addrconf */
1500         slave_dev->flags |= IFF_SLAVE;
1501
1502         /* open the slave since the application closed it */
1503         res = dev_open(slave_dev);
1504         if (res) {
1505                 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1506                 goto err_restore_mac;
1507         }
1508
1509         slave_dev->priv_flags |= IFF_BONDING;
1510         /* initialize slave stats */
1511         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1512
1513         if (bond_is_lb(bond)) {
1514                 /* bond_alb_init_slave() must be called before all other stages since
1515                  * it might fail and we do not want to have to undo everything
1516                  */
1517                 res = bond_alb_init_slave(bond, new_slave);
1518                 if (res)
1519                         goto err_close;
1520         }
1521
1522         /* If the mode uses primary, then the following is handled by
1523          * bond_change_active_slave().
1524          */
1525         if (!bond_uses_primary(bond)) {
1526                 /* set promiscuity level to new slave */
1527                 if (bond_dev->flags & IFF_PROMISC) {
1528                         res = dev_set_promiscuity(slave_dev, 1);
1529                         if (res)
1530                                 goto err_close;
1531                 }
1532
1533                 /* set allmulti level to new slave */
1534                 if (bond_dev->flags & IFF_ALLMULTI) {
1535                         res = dev_set_allmulti(slave_dev, 1);
1536                         if (res)
1537                                 goto err_close;
1538                 }
1539
1540                 netif_addr_lock_bh(bond_dev);
1541
1542                 dev_mc_sync_multiple(slave_dev, bond_dev);
1543                 dev_uc_sync_multiple(slave_dev, bond_dev);
1544
1545                 netif_addr_unlock_bh(bond_dev);
1546         }
1547
1548         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1549                 /* add lacpdu mc addr to mc list */
1550                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1551
1552                 dev_mc_add(slave_dev, lacpdu_multicast);
1553         }
1554
1555         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1556         if (res) {
1557                 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1558                            slave_dev->name);
1559                 goto err_close;
1560         }
1561
1562         prev_slave = bond_last_slave(bond);
1563
1564         new_slave->delay = 0;
1565         new_slave->link_failure_count = 0;
1566
1567         bond_update_speed_duplex(new_slave);
1568
1569         new_slave->last_rx = jiffies -
1570                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1571         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1572                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1573
1574         if (bond->params.miimon && !bond->params.use_carrier) {
1575                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1576
1577                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1578                         /* miimon is set but a bonded network driver
1579                          * does not support ETHTOOL/MII and
1580                          * arp_interval is not set.  Note: if
1581                          * use_carrier is enabled, we will never go
1582                          * here (because netif_carrier is always
1583                          * supported); thus, we don't need to change
1584                          * the messages for netif_carrier.
1585                          */
1586                         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",
1587                                     slave_dev->name);
1588                 } else if (link_reporting == -1) {
1589                         /* unable get link status using mii/ethtool */
1590                         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",
1591                                     slave_dev->name);
1592                 }
1593         }
1594
1595         /* check for initial state */
1596         new_slave->link = BOND_LINK_NOCHANGE;
1597         if (bond->params.miimon) {
1598                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1599                         if (bond->params.updelay) {
1600                                 bond_set_slave_link_state(new_slave,
1601                                                           BOND_LINK_BACK,
1602                                                           BOND_SLAVE_NOTIFY_NOW);
1603                                 new_slave->delay = bond->params.updelay;
1604                         } else {
1605                                 bond_set_slave_link_state(new_slave,
1606                                                           BOND_LINK_UP,
1607                                                           BOND_SLAVE_NOTIFY_NOW);
1608                         }
1609                 } else {
1610                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1611                                                   BOND_SLAVE_NOTIFY_NOW);
1612                 }
1613         } else if (bond->params.arp_interval) {
1614                 bond_set_slave_link_state(new_slave,
1615                                           (netif_carrier_ok(slave_dev) ?
1616                                           BOND_LINK_UP : BOND_LINK_DOWN),
1617                                           BOND_SLAVE_NOTIFY_NOW);
1618         } else {
1619                 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1620                                           BOND_SLAVE_NOTIFY_NOW);
1621         }
1622
1623         if (new_slave->link != BOND_LINK_DOWN)
1624                 new_slave->last_link_up = jiffies;
1625         netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1626                    new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1627                    (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1628
1629         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1630                 /* if there is a primary slave, remember it */
1631                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1632                         rcu_assign_pointer(bond->primary_slave, new_slave);
1633                         bond->force_primary = true;
1634                 }
1635         }
1636
1637         switch (BOND_MODE(bond)) {
1638         case BOND_MODE_ACTIVEBACKUP:
1639                 bond_set_slave_inactive_flags(new_slave,
1640                                               BOND_SLAVE_NOTIFY_NOW);
1641                 break;
1642         case BOND_MODE_8023AD:
1643                 /* in 802.3ad mode, the internal mechanism
1644                  * will activate the slaves in the selected
1645                  * aggregator
1646                  */
1647                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1648                 /* if this is the first slave */
1649                 if (!prev_slave) {
1650                         SLAVE_AD_INFO(new_slave)->id = 1;
1651                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1652                          * can be called only after the mac address of the bond is set
1653                          */
1654                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1655                 } else {
1656                         SLAVE_AD_INFO(new_slave)->id =
1657                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1658                 }
1659
1660                 bond_3ad_bind_slave(new_slave);
1661                 break;
1662         case BOND_MODE_TLB:
1663         case BOND_MODE_ALB:
1664                 bond_set_active_slave(new_slave);
1665                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1666                 break;
1667         default:
1668                 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1669
1670                 /* always active in trunk mode */
1671                 bond_set_active_slave(new_slave);
1672
1673                 /* In trunking mode there is little meaning to curr_active_slave
1674                  * anyway (it holds no special properties of the bond device),
1675                  * so we can change it without calling change_active_interface()
1676                  */
1677                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1678                     new_slave->link == BOND_LINK_UP)
1679                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1680
1681                 break;
1682         } /* switch(bond_mode) */
1683
1684 #ifdef CONFIG_NET_POLL_CONTROLLER
1685         slave_dev->npinfo = bond->dev->npinfo;
1686         if (slave_dev->npinfo) {
1687                 if (slave_enable_netpoll(new_slave)) {
1688                         netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1689                         res = -EBUSY;
1690                         goto err_detach;
1691                 }
1692         }
1693 #endif
1694
1695         if (!(bond_dev->features & NETIF_F_LRO))
1696                 dev_disable_lro(slave_dev);
1697
1698         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1699                                          new_slave);
1700         if (res) {
1701                 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1702                 goto err_detach;
1703         }
1704
1705         res = bond_master_upper_dev_link(bond, new_slave);
1706         if (res) {
1707                 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1708                 goto err_unregister;
1709         }
1710
1711         res = bond_sysfs_slave_add(new_slave);
1712         if (res) {
1713                 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1714                 goto err_upper_unlink;
1715         }
1716
1717         bond->slave_cnt++;
1718         bond_compute_features(bond);
1719         bond_set_carrier(bond);
1720
1721         if (bond_uses_primary(bond)) {
1722                 block_netpoll_tx();
1723                 bond_select_active_slave(bond);
1724                 unblock_netpoll_tx();
1725         }
1726
1727         if (bond_mode_uses_xmit_hash(bond))
1728                 bond_update_slave_arr(bond, NULL);
1729
1730         netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1731                     slave_dev->name,
1732                     bond_is_active_slave(new_slave) ? "an active" : "a backup",
1733                     new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1734
1735         /* enslave is successful */
1736         bond_queue_slave_event(new_slave);
1737         return 0;
1738
1739 /* Undo stages on error */
1740 err_upper_unlink:
1741         bond_upper_dev_unlink(bond, new_slave);
1742
1743 err_unregister:
1744         netdev_rx_handler_unregister(slave_dev);
1745
1746 err_detach:
1747         if (!bond_uses_primary(bond))
1748                 bond_hw_addr_flush(bond_dev, slave_dev);
1749
1750         vlan_vids_del_by_dev(slave_dev, bond_dev);
1751         if (rcu_access_pointer(bond->primary_slave) == new_slave)
1752                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1753         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1754                 block_netpoll_tx();
1755                 bond_change_active_slave(bond, NULL);
1756                 bond_select_active_slave(bond);
1757                 unblock_netpoll_tx();
1758         }
1759         /* either primary_slave or curr_active_slave might've changed */
1760         synchronize_rcu();
1761         slave_disable_netpoll(new_slave);
1762
1763 err_close:
1764         slave_dev->priv_flags &= ~IFF_BONDING;
1765         dev_close(slave_dev);
1766
1767 err_restore_mac:
1768         slave_dev->flags &= ~IFF_SLAVE;
1769         if (!bond->params.fail_over_mac ||
1770             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1771                 /* XXX TODO - fom follow mode needs to change master's
1772                  * MAC if this slave's MAC is in use by the bond, or at
1773                  * least print a warning.
1774                  */
1775                 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1776                 addr.sa_family = slave_dev->type;
1777                 dev_set_mac_address(slave_dev, &addr);
1778         }
1779
1780 err_restore_mtu:
1781         dev_set_mtu(slave_dev, new_slave->original_mtu);
1782
1783 err_free:
1784         bond_free_slave(new_slave);
1785
1786 err_undo_flags:
1787         /* Enslave of first slave has failed and we need to fix master's mac */
1788         if (!bond_has_slaves(bond)) {
1789                 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1790                                             slave_dev->dev_addr))
1791                         eth_hw_addr_random(bond_dev);
1792                 if (bond_dev->type != ARPHRD_ETHER) {
1793                         dev_close(bond_dev);
1794                         ether_setup(bond_dev);
1795                         bond_dev->flags |= IFF_MASTER;
1796                         bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1797                 }
1798         }
1799
1800         return res;
1801 }
1802
1803 /* Try to release the slave device <slave> from the bond device <master>
1804  * It is legal to access curr_active_slave without a lock because all the function
1805  * is RTNL-locked. If "all" is true it means that the function is being called
1806  * while destroying a bond interface and all slaves are being released.
1807  *
1808  * The rules for slave state should be:
1809  *   for Active/Backup:
1810  *     Active stays on all backups go down
1811  *   for Bonded connections:
1812  *     The first up interface should be left on and all others downed.
1813  */
1814 static int __bond_release_one(struct net_device *bond_dev,
1815                               struct net_device *slave_dev,
1816                               bool all)
1817 {
1818         struct bonding *bond = netdev_priv(bond_dev);
1819         struct slave *slave, *oldcurrent;
1820         struct sockaddr addr;
1821         int old_flags = bond_dev->flags;
1822         netdev_features_t old_features = bond_dev->features;
1823
1824         /* slave is not a slave or master is not master of this slave */
1825         if (!(slave_dev->flags & IFF_SLAVE) ||
1826             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1827                 netdev_dbg(bond_dev, "cannot release %s\n",
1828                            slave_dev->name);
1829                 return -EINVAL;
1830         }
1831
1832         block_netpoll_tx();
1833
1834         slave = bond_get_slave_by_dev(bond, slave_dev);
1835         if (!slave) {
1836                 /* not a slave of this bond */
1837                 netdev_info(bond_dev, "%s not enslaved\n",
1838                             slave_dev->name);
1839                 unblock_netpoll_tx();
1840                 return -EINVAL;
1841         }
1842
1843         bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1844
1845         bond_sysfs_slave_del(slave);
1846
1847         /* recompute stats just before removing the slave */
1848         bond_get_stats(bond->dev, &bond->bond_stats);
1849
1850         bond_upper_dev_unlink(bond, slave);
1851         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1852          * for this slave anymore.
1853          */
1854         netdev_rx_handler_unregister(slave_dev);
1855
1856         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1857                 bond_3ad_unbind_slave(slave);
1858
1859         if (bond_mode_uses_xmit_hash(bond))
1860                 bond_update_slave_arr(bond, slave);
1861
1862         netdev_info(bond_dev, "Releasing %s interface %s\n",
1863                     bond_is_active_slave(slave) ? "active" : "backup",
1864                     slave_dev->name);
1865
1866         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1867
1868         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1869
1870         if (!all && (!bond->params.fail_over_mac ||
1871                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1872                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1873                     bond_has_slaves(bond))
1874                         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",
1875                                     slave_dev->name, slave->perm_hwaddr,
1876                                     bond_dev->name, slave_dev->name);
1877         }
1878
1879         if (rtnl_dereference(bond->primary_slave) == slave)
1880                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1881
1882         if (oldcurrent == slave)
1883                 bond_change_active_slave(bond, NULL);
1884
1885         if (bond_is_lb(bond)) {
1886                 /* Must be called only after the slave has been
1887                  * detached from the list and the curr_active_slave
1888                  * has been cleared (if our_slave == old_current),
1889                  * but before a new active slave is selected.
1890                  */
1891                 bond_alb_deinit_slave(bond, slave);
1892         }
1893
1894         if (all) {
1895                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1896         } else if (oldcurrent == slave) {
1897                 /* Note that we hold RTNL over this sequence, so there
1898                  * is no concern that another slave add/remove event
1899                  * will interfere.
1900                  */
1901                 bond_select_active_slave(bond);
1902         }
1903
1904         if (!bond_has_slaves(bond)) {
1905                 bond_set_carrier(bond);
1906                 eth_hw_addr_random(bond_dev);
1907         }
1908
1909         unblock_netpoll_tx();
1910         synchronize_rcu();
1911         bond->slave_cnt--;
1912
1913         if (!bond_has_slaves(bond)) {
1914                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1915                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1916         }
1917
1918         bond_compute_features(bond);
1919         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1920             (old_features & NETIF_F_VLAN_CHALLENGED))
1921                 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1922                             slave_dev->name, bond_dev->name);
1923
1924         vlan_vids_del_by_dev(slave_dev, bond_dev);
1925
1926         /* If the mode uses primary, then this case was handled above by
1927          * bond_change_active_slave(..., NULL)
1928          */
1929         if (!bond_uses_primary(bond)) {
1930                 /* unset promiscuity level from slave
1931                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1932                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1933                  * value of that flag before that change, as that was the value
1934                  * when this slave was attached, so we cache at the start of the
1935                  * function and use it here. Same goes for ALLMULTI below
1936                  */
1937                 if (old_flags & IFF_PROMISC)
1938                         dev_set_promiscuity(slave_dev, -1);
1939
1940                 /* unset allmulti level from slave */
1941                 if (old_flags & IFF_ALLMULTI)
1942                         dev_set_allmulti(slave_dev, -1);
1943
1944                 bond_hw_addr_flush(bond_dev, slave_dev);
1945         }
1946
1947         slave_disable_netpoll(slave);
1948
1949         /* close slave before restoring its mac address */
1950         dev_close(slave_dev);
1951
1952         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1953             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1954                 /* restore original ("permanent") mac address */
1955                 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1956                 addr.sa_family = slave_dev->type;
1957                 dev_set_mac_address(slave_dev, &addr);
1958         }
1959
1960         dev_set_mtu(slave_dev, slave->original_mtu);
1961
1962         slave_dev->priv_flags &= ~IFF_BONDING;
1963
1964         bond_free_slave(slave);
1965
1966         return 0;
1967 }
1968
1969 /* A wrapper used because of ndo_del_link */
1970 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1971 {
1972         return __bond_release_one(bond_dev, slave_dev, false);
1973 }
1974
1975 /* First release a slave and then destroy the bond if no more slaves are left.
1976  * Must be under rtnl_lock when this function is called.
1977  */
1978 static int  bond_release_and_destroy(struct net_device *bond_dev,
1979                                      struct net_device *slave_dev)
1980 {
1981         struct bonding *bond = netdev_priv(bond_dev);
1982         int ret;
1983
1984         ret = bond_release(bond_dev, slave_dev);
1985         if (ret == 0 && !bond_has_slaves(bond)) {
1986                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1987                 netdev_info(bond_dev, "Destroying bond %s\n",
1988                             bond_dev->name);
1989                 bond_remove_proc_entry(bond);
1990                 unregister_netdevice(bond_dev);
1991         }
1992         return ret;
1993 }
1994
1995 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1996 {
1997         struct bonding *bond = netdev_priv(bond_dev);
1998         bond_fill_ifbond(bond, info);
1999         return 0;
2000 }
2001
2002 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2003 {
2004         struct bonding *bond = netdev_priv(bond_dev);
2005         struct list_head *iter;
2006         int i = 0, res = -ENODEV;
2007         struct slave *slave;
2008
2009         bond_for_each_slave(bond, slave, iter) {
2010                 if (i++ == (int)info->slave_id) {
2011                         res = 0;
2012                         bond_fill_ifslave(slave, info);
2013                         break;
2014                 }
2015         }
2016
2017         return res;
2018 }
2019
2020 /*-------------------------------- Monitoring -------------------------------*/
2021
2022 /* called with rcu_read_lock() */
2023 static int bond_miimon_inspect(struct bonding *bond)
2024 {
2025         int link_state, commit = 0;
2026         struct list_head *iter;
2027         struct slave *slave;
2028         bool ignore_updelay;
2029
2030         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2031
2032         bond_for_each_slave_rcu(bond, slave, iter) {
2033                 slave->new_link = BOND_LINK_NOCHANGE;
2034
2035                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2036
2037                 switch (slave->link) {
2038                 case BOND_LINK_UP:
2039                         if (link_state)
2040                                 continue;
2041
2042                         bond_set_slave_link_state(slave, BOND_LINK_FAIL,
2043                                                   BOND_SLAVE_NOTIFY_LATER);
2044                         slave->delay = bond->params.downdelay;
2045                         if (slave->delay) {
2046                                 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2047                                             (BOND_MODE(bond) ==
2048                                              BOND_MODE_ACTIVEBACKUP) ?
2049                                              (bond_is_active_slave(slave) ?
2050                                               "active " : "backup ") : "",
2051                                             slave->dev->name,
2052                                             bond->params.downdelay * bond->params.miimon);
2053                         }
2054                         /*FALLTHRU*/
2055                 case BOND_LINK_FAIL:
2056                         if (link_state) {
2057                                 /* recovered before downdelay expired */
2058                                 bond_set_slave_link_state(slave, BOND_LINK_UP,
2059                                                           BOND_SLAVE_NOTIFY_LATER);
2060                                 slave->last_link_up = jiffies;
2061                                 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2062                                             (bond->params.downdelay - slave->delay) *
2063                                             bond->params.miimon,
2064                                             slave->dev->name);
2065                                 continue;
2066                         }
2067
2068                         if (slave->delay <= 0) {
2069                                 slave->new_link = BOND_LINK_DOWN;
2070                                 commit++;
2071                                 continue;
2072                         }
2073
2074                         slave->delay--;
2075                         break;
2076
2077                 case BOND_LINK_DOWN:
2078                         if (!link_state)
2079                                 continue;
2080
2081                         bond_set_slave_link_state(slave, BOND_LINK_BACK,
2082                                                   BOND_SLAVE_NOTIFY_LATER);
2083                         slave->delay = bond->params.updelay;
2084
2085                         if (slave->delay) {
2086                                 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2087                                             slave->dev->name,
2088                                             ignore_updelay ? 0 :
2089                                             bond->params.updelay *
2090                                             bond->params.miimon);
2091                         }
2092                         /*FALLTHRU*/
2093                 case BOND_LINK_BACK:
2094                         if (!link_state) {
2095                                 bond_set_slave_link_state(slave,
2096                                                           BOND_LINK_DOWN,
2097                                                           BOND_SLAVE_NOTIFY_LATER);
2098                                 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2099                                             (bond->params.updelay - slave->delay) *
2100                                             bond->params.miimon,
2101                                             slave->dev->name);
2102
2103                                 continue;
2104                         }
2105
2106                         if (ignore_updelay)
2107                                 slave->delay = 0;
2108
2109                         if (slave->delay <= 0) {
2110                                 slave->new_link = BOND_LINK_UP;
2111                                 commit++;
2112                                 ignore_updelay = false;
2113                                 continue;
2114                         }
2115
2116                         slave->delay--;
2117                         break;
2118                 }
2119         }
2120
2121         return commit;
2122 }
2123
2124 static void bond_miimon_commit(struct bonding *bond)
2125 {
2126         struct list_head *iter;
2127         struct slave *slave, *primary;
2128
2129         bond_for_each_slave(bond, slave, iter) {
2130                 switch (slave->new_link) {
2131                 case BOND_LINK_NOCHANGE:
2132                         continue;
2133
2134                 case BOND_LINK_UP:
2135                         bond_update_speed_duplex(slave);
2136                         bond_set_slave_link_state(slave, BOND_LINK_UP,
2137                                                   BOND_SLAVE_NOTIFY_NOW);
2138                         slave->last_link_up = jiffies;
2139
2140                         primary = rtnl_dereference(bond->primary_slave);
2141                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2142                                 /* prevent it from being the active one */
2143                                 bond_set_backup_slave(slave);
2144                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2145                                 /* make it immediately active */
2146                                 bond_set_active_slave(slave);
2147                         } else if (slave != primary) {
2148                                 /* prevent it from being the active one */
2149                                 bond_set_backup_slave(slave);
2150                         }
2151
2152                         netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2153                                     slave->dev->name,
2154                                     slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2155                                     slave->duplex ? "full" : "half");
2156
2157                         /* notify ad that the link status has changed */
2158                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2159                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2160
2161                         if (bond_is_lb(bond))
2162                                 bond_alb_handle_link_change(bond, slave,
2163                                                             BOND_LINK_UP);
2164
2165                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2166                                 bond_update_slave_arr(bond, NULL);
2167
2168                         if (!bond->curr_active_slave || slave == primary)
2169                                 goto do_failover;
2170
2171                         continue;
2172
2173                 case BOND_LINK_DOWN:
2174                         if (slave->link_failure_count < UINT_MAX)
2175                                 slave->link_failure_count++;
2176
2177                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2178                                                   BOND_SLAVE_NOTIFY_NOW);
2179
2180                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2181                             BOND_MODE(bond) == BOND_MODE_8023AD)
2182                                 bond_set_slave_inactive_flags(slave,
2183                                                               BOND_SLAVE_NOTIFY_NOW);
2184
2185                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2186                                     slave->dev->name);
2187
2188                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2189                                 bond_3ad_handle_link_change(slave,
2190                                                             BOND_LINK_DOWN);
2191
2192                         if (bond_is_lb(bond))
2193                                 bond_alb_handle_link_change(bond, slave,
2194                                                             BOND_LINK_DOWN);
2195
2196                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2197                                 bond_update_slave_arr(bond, NULL);
2198
2199                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2200                                 goto do_failover;
2201
2202                         continue;
2203
2204                 default:
2205                         netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2206                                    slave->new_link, slave->dev->name);
2207                         slave->new_link = BOND_LINK_NOCHANGE;
2208
2209                         continue;
2210                 }
2211
2212 do_failover:
2213                 block_netpoll_tx();
2214                 bond_select_active_slave(bond);
2215                 unblock_netpoll_tx();
2216         }
2217
2218         bond_set_carrier(bond);
2219 }
2220
2221 /* bond_mii_monitor
2222  *
2223  * Really a wrapper that splits the mii monitor into two phases: an
2224  * inspection, then (if inspection indicates something needs to be done)
2225  * an acquisition of appropriate locks followed by a commit phase to
2226  * implement whatever link state changes are indicated.
2227  */
2228 static void bond_mii_monitor(struct work_struct *work)
2229 {
2230         struct bonding *bond = container_of(work, struct bonding,
2231                                             mii_work.work);
2232         bool should_notify_peers = false;
2233         unsigned long delay;
2234
2235         delay = msecs_to_jiffies(bond->params.miimon);
2236
2237         if (!bond_has_slaves(bond))
2238                 goto re_arm;
2239
2240         rcu_read_lock();
2241
2242         should_notify_peers = bond_should_notify_peers(bond);
2243
2244         if (bond_miimon_inspect(bond)) {
2245                 rcu_read_unlock();
2246
2247                 /* Race avoidance with bond_close cancel of workqueue */
2248                 if (!rtnl_trylock()) {
2249                         delay = 1;
2250                         should_notify_peers = false;
2251                         goto re_arm;
2252                 }
2253
2254                 bond_miimon_commit(bond);
2255
2256                 rtnl_unlock();  /* might sleep, hold no other locks */
2257         } else
2258                 rcu_read_unlock();
2259
2260 re_arm:
2261         if (bond->params.miimon)
2262                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2263
2264         if (should_notify_peers) {
2265                 if (!rtnl_trylock())
2266                         return;
2267                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2268                 rtnl_unlock();
2269         }
2270 }
2271
2272 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2273 {
2274         struct net_device *upper;
2275         struct list_head *iter;
2276         bool ret = false;
2277
2278         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2279                 return true;
2280
2281         rcu_read_lock();
2282         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2283                 if (ip == bond_confirm_addr(upper, 0, ip)) {
2284                         ret = true;
2285                         break;
2286                 }
2287         }
2288         rcu_read_unlock();
2289
2290         return ret;
2291 }
2292
2293 /* We go to the (large) trouble of VLAN tagging ARP frames because
2294  * switches in VLAN mode (especially if ports are configured as
2295  * "native" to a VLAN) might not pass non-tagged frames.
2296  */
2297 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2298                           __be32 dest_ip, __be32 src_ip,
2299                           struct bond_vlan_tag *tags)
2300 {
2301         struct sk_buff *skb;
2302         struct bond_vlan_tag *outer_tag = tags;
2303
2304         netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2305                    arp_op, slave_dev->name, &dest_ip, &src_ip);
2306
2307         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2308                          NULL, slave_dev->dev_addr, NULL);
2309
2310         if (!skb) {
2311                 net_err_ratelimited("ARP packet allocation failed\n");
2312                 return;
2313         }
2314
2315         if (!tags || tags->vlan_proto == VLAN_N_VID)
2316                 goto xmit;
2317
2318         tags++;
2319
2320         /* Go through all the tags backwards and add them to the packet */
2321         while (tags->vlan_proto != VLAN_N_VID) {
2322                 if (!tags->vlan_id) {
2323                         tags++;
2324                         continue;
2325                 }
2326
2327                 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2328                            ntohs(outer_tag->vlan_proto), tags->vlan_id);
2329                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2330                                                 tags->vlan_id);
2331                 if (!skb) {
2332                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2333                         return;
2334                 }
2335
2336                 tags++;
2337         }
2338         /* Set the outer tag */
2339         if (outer_tag->vlan_id) {
2340                 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2341                            ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2342                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2343                                        outer_tag->vlan_id);
2344         }
2345
2346 xmit:
2347         arp_xmit(skb);
2348 }
2349
2350 /* Validate the device path between the @start_dev and the @end_dev.
2351  * The path is valid if the @end_dev is reachable through device
2352  * stacking.
2353  * When the path is validated, collect any vlan information in the
2354  * path.
2355  */
2356 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2357                                               struct net_device *end_dev,
2358                                               int level)
2359 {
2360         struct bond_vlan_tag *tags;
2361         struct net_device *upper;
2362         struct list_head  *iter;
2363
2364         if (start_dev == end_dev) {
2365                 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2366                 if (!tags)
2367                         return ERR_PTR(-ENOMEM);
2368                 tags[level].vlan_proto = VLAN_N_VID;
2369                 return tags;
2370         }
2371
2372         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2373                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2374                 if (IS_ERR_OR_NULL(tags)) {
2375                         if (IS_ERR(tags))
2376                                 return tags;
2377                         continue;
2378                 }
2379                 if (is_vlan_dev(upper)) {
2380                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2381                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2382                 }
2383
2384                 return tags;
2385         }
2386
2387         return NULL;
2388 }
2389
2390 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2391 {
2392         struct rtable *rt;
2393         struct bond_vlan_tag *tags;
2394         __be32 *targets = bond->params.arp_targets, addr;
2395         int i;
2396
2397         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2398                 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2399                 tags = NULL;
2400
2401                 /* Find out through which dev should the packet go */
2402                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2403                                      RTO_ONLINK, 0);
2404                 if (IS_ERR(rt)) {
2405                         /* there's no route to target - try to send arp
2406                          * probe to generate any traffic (arp_validate=0)
2407                          */
2408                         if (bond->params.arp_validate)
2409                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2410                                                      bond->dev->name,
2411                                                      &targets[i]);
2412                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2413                                       0, tags);
2414                         continue;
2415                 }
2416
2417                 /* bond device itself */
2418                 if (rt->dst.dev == bond->dev)
2419                         goto found;
2420
2421                 rcu_read_lock();
2422                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2423                 rcu_read_unlock();
2424
2425                 if (!IS_ERR_OR_NULL(tags))
2426                         goto found;
2427
2428                 /* Not our device - skip */
2429                 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2430                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2431
2432                 ip_rt_put(rt);
2433                 continue;
2434
2435 found:
2436                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2437                 ip_rt_put(rt);
2438                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2439                               addr, tags);
2440                 kfree(tags);
2441         }
2442 }
2443
2444 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2445 {
2446         int i;
2447
2448         if (!sip || !bond_has_this_ip(bond, tip)) {
2449                 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2450                            &sip, &tip);
2451                 return;
2452         }
2453
2454         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2455         if (i == -1) {
2456                 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2457                            &sip);
2458                 return;
2459         }
2460         slave->last_rx = jiffies;
2461         slave->target_last_arp_rx[i] = jiffies;
2462 }
2463
2464 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2465                  struct slave *slave)
2466 {
2467         struct arphdr *arp = (struct arphdr *)skb->data;
2468         struct slave *curr_active_slave, *curr_arp_slave;
2469         unsigned char *arp_ptr;
2470         __be32 sip, tip;
2471         int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2472
2473         if (!slave_do_arp_validate(bond, slave)) {
2474                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2475                     !slave_do_arp_validate_only(bond))
2476                         slave->last_rx = jiffies;
2477                 return RX_HANDLER_ANOTHER;
2478         } else if (!is_arp) {
2479                 return RX_HANDLER_ANOTHER;
2480         }
2481
2482         alen = arp_hdr_len(bond->dev);
2483
2484         netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2485                    skb->dev->name);
2486
2487         if (alen > skb_headlen(skb)) {
2488                 arp = kmalloc(alen, GFP_ATOMIC);
2489                 if (!arp)
2490                         goto out_unlock;
2491                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2492                         goto out_unlock;
2493         }
2494
2495         if (arp->ar_hln != bond->dev->addr_len ||
2496             skb->pkt_type == PACKET_OTHERHOST ||
2497             skb->pkt_type == PACKET_LOOPBACK ||
2498             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2499             arp->ar_pro != htons(ETH_P_IP) ||
2500             arp->ar_pln != 4)
2501                 goto out_unlock;
2502
2503         arp_ptr = (unsigned char *)(arp + 1);
2504         arp_ptr += bond->dev->addr_len;
2505         memcpy(&sip, arp_ptr, 4);
2506         arp_ptr += 4 + bond->dev->addr_len;
2507         memcpy(&tip, arp_ptr, 4);
2508
2509         netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2510                    slave->dev->name, bond_slave_state(slave),
2511                      bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2512                      &sip, &tip);
2513
2514         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2515         curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2516
2517         /* We 'trust' the received ARP enough to validate it if:
2518          *
2519          * (a) the slave receiving the ARP is active (which includes the
2520          * current ARP slave, if any), or
2521          *
2522          * (b) the receiving slave isn't active, but there is a currently
2523          * active slave and it received valid arp reply(s) after it became
2524          * the currently active slave, or
2525          *
2526          * (c) there is an ARP slave that sent an ARP during the prior ARP
2527          * interval, and we receive an ARP reply on any slave.  We accept
2528          * these because switch FDB update delays may deliver the ARP
2529          * reply to a slave other than the sender of the ARP request.
2530          *
2531          * Note: for (b), backup slaves are receiving the broadcast ARP
2532          * request, not a reply.  This request passes from the sending
2533          * slave through the L2 switch(es) to the receiving slave.  Since
2534          * this is checking the request, sip/tip are swapped for
2535          * validation.
2536          *
2537          * This is done to avoid endless looping when we can't reach the
2538          * arp_ip_target and fool ourselves with our own arp requests.
2539          */
2540         if (bond_is_active_slave(slave))
2541                 bond_validate_arp(bond, slave, sip, tip);
2542         else if (curr_active_slave &&
2543                  time_after(slave_last_rx(bond, curr_active_slave),
2544                             curr_active_slave->last_link_up))
2545                 bond_validate_arp(bond, slave, tip, sip);
2546         else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2547                  bond_time_in_interval(bond,
2548                                        dev_trans_start(curr_arp_slave->dev), 1))
2549                 bond_validate_arp(bond, slave, sip, tip);
2550
2551 out_unlock:
2552         if (arp != (struct arphdr *)skb->data)
2553                 kfree(arp);
2554         return RX_HANDLER_ANOTHER;
2555 }
2556
2557 /* function to verify if we're in the arp_interval timeslice, returns true if
2558  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2559  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2560  */
2561 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2562                                   int mod)
2563 {
2564         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2565
2566         return time_in_range(jiffies,
2567                              last_act - delta_in_ticks,
2568                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2569 }
2570
2571 /* This function is called regularly to monitor each slave's link
2572  * ensuring that traffic is being sent and received when arp monitoring
2573  * is used in load-balancing mode. if the adapter has been dormant, then an
2574  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2575  * arp monitoring in active backup mode.
2576  */
2577 static void bond_loadbalance_arp_mon(struct work_struct *work)
2578 {
2579         struct bonding *bond = container_of(work, struct bonding,
2580                                             arp_work.work);
2581         struct slave *slave, *oldcurrent;
2582         struct list_head *iter;
2583         int do_failover = 0, slave_state_changed = 0;
2584
2585         if (!bond_has_slaves(bond))
2586                 goto re_arm;
2587
2588         rcu_read_lock();
2589
2590         oldcurrent = rcu_dereference(bond->curr_active_slave);
2591         /* see if any of the previous devices are up now (i.e. they have
2592          * xmt and rcv traffic). the curr_active_slave does not come into
2593          * the picture unless it is null. also, slave->last_link_up is not
2594          * needed here because we send an arp on each slave and give a slave
2595          * as long as it needs to get the tx/rx within the delta.
2596          * TODO: what about up/down delay in arp mode? it wasn't here before
2597          *       so it can wait
2598          */
2599         bond_for_each_slave_rcu(bond, slave, iter) {
2600                 unsigned long trans_start = dev_trans_start(slave->dev);
2601
2602                 if (slave->link != BOND_LINK_UP) {
2603                         if (bond_time_in_interval(bond, trans_start, 1) &&
2604                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2605
2606                                 slave->link  = BOND_LINK_UP;
2607                                 slave_state_changed = 1;
2608
2609                                 /* primary_slave has no meaning in round-robin
2610                                  * mode. the window of a slave being up and
2611                                  * curr_active_slave being null after enslaving
2612                                  * is closed.
2613                                  */
2614                                 if (!oldcurrent) {
2615                                         netdev_info(bond->dev, "link status definitely up for interface %s\n",
2616                                                     slave->dev->name);
2617                                         do_failover = 1;
2618                                 } else {
2619                                         netdev_info(bond->dev, "interface %s is now up\n",
2620                                                     slave->dev->name);
2621                                 }
2622                         }
2623                 } else {
2624                         /* slave->link == BOND_LINK_UP */
2625
2626                         /* not all switches will respond to an arp request
2627                          * when the source ip is 0, so don't take the link down
2628                          * if we don't know our ip yet
2629                          */
2630                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2631                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2632
2633                                 slave->link  = BOND_LINK_DOWN;
2634                                 slave_state_changed = 1;
2635
2636                                 if (slave->link_failure_count < UINT_MAX)
2637                                         slave->link_failure_count++;
2638
2639                                 netdev_info(bond->dev, "interface %s is now down\n",
2640                                             slave->dev->name);
2641
2642                                 if (slave == oldcurrent)
2643                                         do_failover = 1;
2644                         }
2645                 }
2646
2647                 /* note: if switch is in round-robin mode, all links
2648                  * must tx arp to ensure all links rx an arp - otherwise
2649                  * links may oscillate or not come up at all; if switch is
2650                  * in something like xor mode, there is nothing we can
2651                  * do - all replies will be rx'ed on same link causing slaves
2652                  * to be unstable during low/no traffic periods
2653                  */
2654                 if (bond_slave_is_up(slave))
2655                         bond_arp_send_all(bond, slave);
2656         }
2657
2658         rcu_read_unlock();
2659
2660         if (do_failover || slave_state_changed) {
2661                 if (!rtnl_trylock())
2662                         goto re_arm;
2663
2664                 if (slave_state_changed) {
2665                         bond_slave_state_change(bond);
2666                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2667                                 bond_update_slave_arr(bond, NULL);
2668                 }
2669                 if (do_failover) {
2670                         block_netpoll_tx();
2671                         bond_select_active_slave(bond);
2672                         unblock_netpoll_tx();
2673                 }
2674                 rtnl_unlock();
2675         }
2676
2677 re_arm:
2678         if (bond->params.arp_interval)
2679                 queue_delayed_work(bond->wq, &bond->arp_work,
2680                                    msecs_to_jiffies(bond->params.arp_interval));
2681 }
2682
2683 /* Called to inspect slaves for active-backup mode ARP monitor link state
2684  * changes.  Sets new_link in slaves to specify what action should take
2685  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2686  * to link states must be committed.
2687  *
2688  * Called with rcu_read_lock held.
2689  */
2690 static int bond_ab_arp_inspect(struct bonding *bond)
2691 {
2692         unsigned long trans_start, last_rx;
2693         struct list_head *iter;
2694         struct slave *slave;
2695         int commit = 0;
2696
2697         bond_for_each_slave_rcu(bond, slave, iter) {
2698                 slave->new_link = BOND_LINK_NOCHANGE;
2699                 last_rx = slave_last_rx(bond, slave);
2700
2701                 if (slave->link != BOND_LINK_UP) {
2702                         if (bond_time_in_interval(bond, last_rx, 1)) {
2703                                 slave->new_link = BOND_LINK_UP;
2704                                 commit++;
2705                         }
2706                         continue;
2707                 }
2708
2709                 /* Give slaves 2*delta after being enslaved or made
2710                  * active.  This avoids bouncing, as the last receive
2711                  * times need a full ARP monitor cycle to be updated.
2712                  */
2713                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2714                         continue;
2715
2716                 /* Backup slave is down if:
2717                  * - No current_arp_slave AND
2718                  * - more than 3*delta since last receive AND
2719                  * - the bond has an IP address
2720                  *
2721                  * Note: a non-null current_arp_slave indicates
2722                  * the curr_active_slave went down and we are
2723                  * searching for a new one; under this condition
2724                  * we only take the curr_active_slave down - this
2725                  * gives each slave a chance to tx/rx traffic
2726                  * before being taken out
2727                  */
2728                 if (!bond_is_active_slave(slave) &&
2729                     !rcu_access_pointer(bond->current_arp_slave) &&
2730                     !bond_time_in_interval(bond, last_rx, 3)) {
2731                         slave->new_link = BOND_LINK_DOWN;
2732                         commit++;
2733                 }
2734
2735                 /* Active slave is down if:
2736                  * - more than 2*delta since transmitting OR
2737                  * - (more than 2*delta since receive AND
2738                  *    the bond has an IP address)
2739                  */
2740                 trans_start = dev_trans_start(slave->dev);
2741                 if (bond_is_active_slave(slave) &&
2742                     (!bond_time_in_interval(bond, trans_start, 2) ||
2743                      !bond_time_in_interval(bond, last_rx, 2))) {
2744                         slave->new_link = BOND_LINK_DOWN;
2745                         commit++;
2746                 }
2747         }
2748
2749         return commit;
2750 }
2751
2752 /* Called to commit link state changes noted by inspection step of
2753  * active-backup mode ARP monitor.
2754  *
2755  * Called with RTNL hold.
2756  */
2757 static void bond_ab_arp_commit(struct bonding *bond)
2758 {
2759         unsigned long trans_start;
2760         struct list_head *iter;
2761         struct slave *slave;
2762
2763         bond_for_each_slave(bond, slave, iter) {
2764                 switch (slave->new_link) {
2765                 case BOND_LINK_NOCHANGE:
2766                         continue;
2767
2768                 case BOND_LINK_UP:
2769                         trans_start = dev_trans_start(slave->dev);
2770                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
2771                             (!rtnl_dereference(bond->curr_active_slave) &&
2772                              bond_time_in_interval(bond, trans_start, 1))) {
2773                                 struct slave *current_arp_slave;
2774
2775                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2776                                 bond_set_slave_link_state(slave, BOND_LINK_UP,
2777                                                           BOND_SLAVE_NOTIFY_NOW);
2778                                 if (current_arp_slave) {
2779                                         bond_set_slave_inactive_flags(
2780                                                 current_arp_slave,
2781                                                 BOND_SLAVE_NOTIFY_NOW);
2782                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2783                                 }
2784
2785                                 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2786                                             slave->dev->name);
2787
2788                                 if (!rtnl_dereference(bond->curr_active_slave) ||
2789                                     slave == rtnl_dereference(bond->primary_slave))
2790                                         goto do_failover;
2791
2792                         }
2793
2794                         continue;
2795
2796                 case BOND_LINK_DOWN:
2797                         if (slave->link_failure_count < UINT_MAX)
2798                                 slave->link_failure_count++;
2799
2800                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2801                                                   BOND_SLAVE_NOTIFY_NOW);
2802                         bond_set_slave_inactive_flags(slave,
2803                                                       BOND_SLAVE_NOTIFY_NOW);
2804
2805                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2806                                     slave->dev->name);
2807
2808                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
2809                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2810                                 goto do_failover;
2811                         }
2812
2813                         continue;
2814
2815                 default:
2816                         netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2817                                    slave->new_link, slave->dev->name);
2818                         continue;
2819                 }
2820
2821 do_failover:
2822                 block_netpoll_tx();
2823                 bond_select_active_slave(bond);
2824                 unblock_netpoll_tx();
2825         }
2826
2827         bond_set_carrier(bond);
2828 }
2829
2830 /* Send ARP probes for active-backup mode ARP monitor.
2831  *
2832  * Called with rcu_read_lock held.
2833  */
2834 static bool bond_ab_arp_probe(struct bonding *bond)
2835 {
2836         struct slave *slave, *before = NULL, *new_slave = NULL,
2837                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2838                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2839         struct list_head *iter;
2840         bool found = false;
2841         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2842
2843         if (curr_arp_slave && curr_active_slave)
2844                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2845                             curr_arp_slave->dev->name,
2846                             curr_active_slave->dev->name);
2847
2848         if (curr_active_slave) {
2849                 bond_arp_send_all(bond, curr_active_slave);
2850                 return should_notify_rtnl;
2851         }
2852
2853         /* if we don't have a curr_active_slave, search for the next available
2854          * backup slave from the current_arp_slave and make it the candidate
2855          * for becoming the curr_active_slave
2856          */
2857
2858         if (!curr_arp_slave) {
2859                 curr_arp_slave = bond_first_slave_rcu(bond);
2860                 if (!curr_arp_slave)
2861                         return should_notify_rtnl;
2862         }
2863
2864         bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2865
2866         bond_for_each_slave_rcu(bond, slave, iter) {
2867                 if (!found && !before && bond_slave_is_up(slave))
2868                         before = slave;
2869
2870                 if (found && !new_slave && bond_slave_is_up(slave))
2871                         new_slave = slave;
2872                 /* if the link state is up at this point, we
2873                  * mark it down - this can happen if we have
2874                  * simultaneous link failures and
2875                  * reselect_active_interface doesn't make this
2876                  * one the current slave so it is still marked
2877                  * up when it is actually down
2878                  */
2879                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2880                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2881                                                   BOND_SLAVE_NOTIFY_LATER);
2882                         if (slave->link_failure_count < UINT_MAX)
2883                                 slave->link_failure_count++;
2884
2885                         bond_set_slave_inactive_flags(slave,
2886                                                       BOND_SLAVE_NOTIFY_LATER);
2887
2888                         netdev_info(bond->dev, "backup interface %s is now down\n",
2889                                     slave->dev->name);
2890                 }
2891                 if (slave == curr_arp_slave)
2892                         found = true;
2893         }
2894
2895         if (!new_slave && before)
2896                 new_slave = before;
2897
2898         if (!new_slave)
2899                 goto check_state;
2900
2901         bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2902                                   BOND_SLAVE_NOTIFY_LATER);
2903         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2904         bond_arp_send_all(bond, new_slave);
2905         new_slave->last_link_up = jiffies;
2906         rcu_assign_pointer(bond->current_arp_slave, new_slave);
2907
2908 check_state:
2909         bond_for_each_slave_rcu(bond, slave, iter) {
2910                 if (slave->should_notify || slave->should_notify_link) {
2911                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2912                         break;
2913                 }
2914         }
2915         return should_notify_rtnl;
2916 }
2917
2918 static void bond_activebackup_arp_mon(struct work_struct *work)
2919 {
2920         struct bonding *bond = container_of(work, struct bonding,
2921                                             arp_work.work);
2922         bool should_notify_peers = false;
2923         bool should_notify_rtnl = false;
2924         int delta_in_ticks;
2925
2926         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2927
2928         if (!bond_has_slaves(bond))
2929                 goto re_arm;
2930
2931         rcu_read_lock();
2932
2933         should_notify_peers = bond_should_notify_peers(bond);
2934
2935         if (bond_ab_arp_inspect(bond)) {
2936                 rcu_read_unlock();
2937
2938                 /* Race avoidance with bond_close flush of workqueue */
2939                 if (!rtnl_trylock()) {
2940                         delta_in_ticks = 1;
2941                         should_notify_peers = false;
2942                         goto re_arm;
2943                 }
2944
2945                 bond_ab_arp_commit(bond);
2946
2947                 rtnl_unlock();
2948                 rcu_read_lock();
2949         }
2950
2951         should_notify_rtnl = bond_ab_arp_probe(bond);
2952         rcu_read_unlock();
2953
2954 re_arm:
2955         if (bond->params.arp_interval)
2956                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2957
2958         if (should_notify_peers || should_notify_rtnl) {
2959                 if (!rtnl_trylock())
2960                         return;
2961
2962                 if (should_notify_peers)
2963                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2964                                                  bond->dev);
2965                 if (should_notify_rtnl) {
2966                         bond_slave_state_notify(bond);
2967                         bond_slave_link_notify(bond);
2968                 }
2969
2970                 rtnl_unlock();
2971         }
2972 }
2973
2974 /*-------------------------- netdev event handling --------------------------*/
2975
2976 /* Change device name */
2977 static int bond_event_changename(struct bonding *bond)
2978 {
2979         bond_remove_proc_entry(bond);
2980         bond_create_proc_entry(bond);
2981
2982         bond_debug_reregister(bond);
2983
2984         return NOTIFY_DONE;
2985 }
2986
2987 static int bond_master_netdev_event(unsigned long event,
2988                                     struct net_device *bond_dev)
2989 {
2990         struct bonding *event_bond = netdev_priv(bond_dev);
2991
2992         switch (event) {
2993         case NETDEV_CHANGENAME:
2994                 return bond_event_changename(event_bond);
2995         case NETDEV_UNREGISTER:
2996                 bond_remove_proc_entry(event_bond);
2997                 break;
2998         case NETDEV_REGISTER:
2999                 bond_create_proc_entry(event_bond);
3000                 break;
3001         case NETDEV_NOTIFY_PEERS:
3002                 if (event_bond->send_peer_notif)
3003                         event_bond->send_peer_notif--;
3004                 break;
3005         default:
3006                 break;
3007         }
3008
3009         return NOTIFY_DONE;
3010 }
3011
3012 static int bond_slave_netdev_event(unsigned long event,
3013                                    struct net_device *slave_dev)
3014 {
3015         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3016         struct bonding *bond;
3017         struct net_device *bond_dev;
3018
3019         /* A netdev event can be generated while enslaving a device
3020          * before netdev_rx_handler_register is called in which case
3021          * slave will be NULL
3022          */
3023         if (!slave)
3024                 return NOTIFY_DONE;
3025         bond_dev = slave->bond->dev;
3026         bond = slave->bond;
3027         primary = rtnl_dereference(bond->primary_slave);
3028
3029         switch (event) {
3030         case NETDEV_UNREGISTER:
3031                 if (bond_dev->type != ARPHRD_ETHER)
3032                         bond_release_and_destroy(bond_dev, slave_dev);
3033                 else
3034                         bond_release(bond_dev, slave_dev);
3035                 break;
3036         case NETDEV_UP:
3037         case NETDEV_CHANGE:
3038                 bond_update_speed_duplex(slave);
3039                 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3040                         bond_3ad_adapter_speed_duplex_changed(slave);
3041                 /* Fallthrough */
3042         case NETDEV_DOWN:
3043                 /* Refresh slave-array if applicable!
3044                  * If the setup does not use miimon or arpmon (mode-specific!),
3045                  * then these events will not cause the slave-array to be
3046                  * refreshed. This will cause xmit to use a slave that is not
3047                  * usable. Avoid such situation by refeshing the array at these
3048                  * events. If these (miimon/arpmon) parameters are configured
3049                  * then array gets refreshed twice and that should be fine!
3050                  */
3051                 if (bond_mode_uses_xmit_hash(bond))
3052                         bond_update_slave_arr(bond, NULL);
3053                 break;
3054         case NETDEV_CHANGEMTU:
3055                 /* TODO: Should slaves be allowed to
3056                  * independently alter their MTU?  For
3057                  * an active-backup bond, slaves need
3058                  * not be the same type of device, so
3059                  * MTUs may vary.  For other modes,
3060                  * slaves arguably should have the
3061                  * same MTUs. To do this, we'd need to
3062                  * take over the slave's change_mtu
3063                  * function for the duration of their
3064                  * servitude.
3065                  */
3066                 break;
3067         case NETDEV_CHANGENAME:
3068                 /* we don't care if we don't have primary set */
3069                 if (!bond_uses_primary(bond) ||
3070                     !bond->params.primary[0])
3071                         break;
3072
3073                 if (slave == primary) {
3074                         /* slave's name changed - he's no longer primary */
3075                         RCU_INIT_POINTER(bond->primary_slave, NULL);
3076                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3077                         /* we have a new primary slave */
3078                         rcu_assign_pointer(bond->primary_slave, slave);
3079                 } else { /* we didn't change primary - exit */
3080                         break;
3081                 }
3082
3083                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3084                             primary ? slave_dev->name : "none");
3085
3086                 block_netpoll_tx();
3087                 bond_select_active_slave(bond);
3088                 unblock_netpoll_tx();
3089                 break;
3090         case NETDEV_FEAT_CHANGE:
3091                 bond_compute_features(bond);
3092                 break;
3093         case NETDEV_RESEND_IGMP:
3094                 /* Propagate to master device */
3095                 call_netdevice_notifiers(event, slave->bond->dev);
3096                 break;
3097         default:
3098                 break;
3099         }
3100
3101         return NOTIFY_DONE;
3102 }
3103
3104 /* bond_netdev_event: handle netdev notifier chain events.
3105  *
3106  * This function receives events for the netdev chain.  The caller (an
3107  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3108  * locks for us to safely manipulate the slave devices (RTNL lock,
3109  * dev_probe_lock).
3110  */
3111 static int bond_netdev_event(struct notifier_block *this,
3112                              unsigned long event, void *ptr)
3113 {
3114         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3115
3116         netdev_dbg(event_dev, "event: %lx\n", event);
3117
3118         if (!(event_dev->priv_flags & IFF_BONDING))
3119                 return NOTIFY_DONE;
3120
3121         if (event_dev->flags & IFF_MASTER) {
3122                 netdev_dbg(event_dev, "IFF_MASTER\n");
3123                 return bond_master_netdev_event(event, event_dev);
3124         }
3125
3126         if (event_dev->flags & IFF_SLAVE) {
3127                 netdev_dbg(event_dev, "IFF_SLAVE\n");
3128                 return bond_slave_netdev_event(event, event_dev);
3129         }
3130
3131         return NOTIFY_DONE;
3132 }
3133
3134 static struct notifier_block bond_netdev_notifier = {
3135         .notifier_call = bond_netdev_event,
3136 };
3137
3138 /*---------------------------- Hashing Policies -----------------------------*/
3139
3140 /* L2 hash helper */
3141 static inline u32 bond_eth_hash(struct sk_buff *skb)
3142 {
3143         struct ethhdr *ep, hdr_tmp;
3144
3145         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3146         if (ep)
3147                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3148         return 0;
3149 }
3150
3151 /* Extract the appropriate headers based on bond's xmit policy */
3152 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3153                               struct flow_keys *fk)
3154 {
3155         const struct ipv6hdr *iph6;
3156         const struct iphdr *iph;
3157         int noff, proto = -1;
3158
3159         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3160                 return skb_flow_dissect_flow_keys(skb, fk, 0);
3161
3162         fk->ports.ports = 0;
3163         noff = skb_network_offset(skb);
3164         if (skb->protocol == htons(ETH_P_IP)) {
3165                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3166                         return false;
3167                 iph = ip_hdr(skb);
3168                 iph_to_flow_copy_v4addrs(fk, iph);
3169                 noff += iph->ihl << 2;
3170                 if (!ip_is_fragment(iph))
3171                         proto = iph->protocol;
3172         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3173                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3174                         return false;
3175                 iph6 = ipv6_hdr(skb);
3176                 iph_to_flow_copy_v6addrs(fk, iph6);
3177                 noff += sizeof(*iph6);
3178                 proto = iph6->nexthdr;
3179         } else {
3180                 return false;
3181         }
3182         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3183                 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3184
3185         return true;
3186 }
3187
3188 /**
3189  * bond_xmit_hash - generate a hash value based on the xmit policy
3190  * @bond: bonding device
3191  * @skb: buffer to use for headers
3192  *
3193  * This function will extract the necessary headers from the skb buffer and use
3194  * them to generate a hash based on the xmit_policy set in the bonding device
3195  */
3196 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3197 {
3198         struct flow_keys flow;
3199         u32 hash;
3200
3201         if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3202             skb->l4_hash)
3203                 return skb->hash;
3204
3205         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3206             !bond_flow_dissect(bond, skb, &flow))
3207                 return bond_eth_hash(skb);
3208
3209         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3210             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3211                 hash = bond_eth_hash(skb);
3212         else
3213                 hash = (__force u32)flow.ports.ports;
3214         hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3215                 (__force u32)flow_get_u32_src(&flow);
3216         hash ^= (hash >> 16);
3217         hash ^= (hash >> 8);
3218
3219         return hash;
3220 }
3221
3222 /*-------------------------- Device entry points ----------------------------*/
3223
3224 static void bond_work_init_all(struct bonding *bond)
3225 {
3226         INIT_DELAYED_WORK(&bond->mcast_work,
3227                           bond_resend_igmp_join_requests_delayed);
3228         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3229         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3230         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3231                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3232         else
3233                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3234         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3235         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3236 }
3237
3238 static void bond_work_cancel_all(struct bonding *bond)
3239 {
3240         cancel_delayed_work_sync(&bond->mii_work);
3241         cancel_delayed_work_sync(&bond->arp_work);
3242         cancel_delayed_work_sync(&bond->alb_work);
3243         cancel_delayed_work_sync(&bond->ad_work);
3244         cancel_delayed_work_sync(&bond->mcast_work);
3245         cancel_delayed_work_sync(&bond->slave_arr_work);
3246 }
3247
3248 static int bond_open(struct net_device *bond_dev)
3249 {
3250         struct bonding *bond = netdev_priv(bond_dev);
3251         struct list_head *iter;
3252         struct slave *slave;
3253
3254         /* reset slave->backup and slave->inactive */
3255         if (bond_has_slaves(bond)) {
3256                 bond_for_each_slave(bond, slave, iter) {
3257                         if (bond_uses_primary(bond) &&
3258                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3259                                 bond_set_slave_inactive_flags(slave,
3260                                                               BOND_SLAVE_NOTIFY_NOW);
3261                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3262                                 bond_set_slave_active_flags(slave,
3263                                                             BOND_SLAVE_NOTIFY_NOW);
3264                         }
3265                 }
3266         }
3267
3268         bond_work_init_all(bond);
3269
3270         if (bond_is_lb(bond)) {
3271                 /* bond_alb_initialize must be called before the timer
3272                  * is started.
3273                  */
3274                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3275                         return -ENOMEM;
3276                 if (bond->params.tlb_dynamic_lb)
3277                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3278         }
3279
3280         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3281                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3282
3283         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3284                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3285                 bond->recv_probe = bond_arp_rcv;
3286         }
3287
3288         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3289                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3290                 /* register to receive LACPDUs */
3291                 bond->recv_probe = bond_3ad_lacpdu_recv;
3292                 bond_3ad_initiate_agg_selection(bond, 1);
3293         }
3294
3295         if (bond_mode_uses_xmit_hash(bond))
3296                 bond_update_slave_arr(bond, NULL);
3297
3298         return 0;
3299 }
3300
3301 static int bond_close(struct net_device *bond_dev)
3302 {
3303         struct bonding *bond = netdev_priv(bond_dev);
3304
3305         bond_work_cancel_all(bond);
3306         bond->send_peer_notif = 0;
3307         if (bond_is_lb(bond))
3308                 bond_alb_deinitialize(bond);
3309         bond->recv_probe = NULL;
3310
3311         return 0;
3312 }
3313
3314 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3315  * that some drivers can provide 32bit values only.
3316  */
3317 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3318                             const struct rtnl_link_stats64 *_new,
3319                             const struct rtnl_link_stats64 *_old)
3320 {
3321         const u64 *new = (const u64 *)_new;
3322         const u64 *old = (const u64 *)_old;
3323         u64 *res = (u64 *)_res;
3324         int i;
3325
3326         for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3327                 u64 nv = new[i];
3328                 u64 ov = old[i];
3329
3330                 /* detects if this particular field is 32bit only */
3331                 if (((nv | ov) >> 32) == 0)
3332                         res[i] += (u32)nv - (u32)ov;
3333                 else
3334                         res[i] += nv - ov;
3335         }
3336 }
3337
3338 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3339                                                 struct rtnl_link_stats64 *stats)
3340 {
3341         struct bonding *bond = netdev_priv(bond_dev);
3342         struct rtnl_link_stats64 temp;
3343         struct list_head *iter;
3344         struct slave *slave;
3345
3346         spin_lock(&bond->stats_lock);
3347         memcpy(stats, &bond->bond_stats, sizeof(*stats));
3348
3349         rcu_read_lock();
3350         bond_for_each_slave_rcu(bond, slave, iter) {
3351                 const struct rtnl_link_stats64 *new =
3352                         dev_get_stats(slave->dev, &temp);
3353
3354                 bond_fold_stats(stats, new, &slave->slave_stats);
3355
3356                 /* save off the slave stats for the next run */
3357                 memcpy(&slave->slave_stats, new, sizeof(*new));
3358         }
3359         rcu_read_unlock();
3360
3361         memcpy(&bond->bond_stats, stats, sizeof(*stats));
3362         spin_unlock(&bond->stats_lock);
3363
3364         return stats;
3365 }
3366
3367 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3368 {
3369         struct bonding *bond = netdev_priv(bond_dev);
3370         struct net_device *slave_dev = NULL;
3371         struct ifbond k_binfo;
3372         struct ifbond __user *u_binfo = NULL;
3373         struct ifslave k_sinfo;
3374         struct ifslave __user *u_sinfo = NULL;
3375         struct mii_ioctl_data *mii = NULL;
3376         struct bond_opt_value newval;
3377         struct net *net;
3378         int res = 0;
3379
3380         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3381
3382         switch (cmd) {
3383         case SIOCGMIIPHY:
3384                 mii = if_mii(ifr);
3385                 if (!mii)
3386                         return -EINVAL;
3387
3388                 mii->phy_id = 0;
3389                 /* Fall Through */
3390         case SIOCGMIIREG:
3391                 /* We do this again just in case we were called by SIOCGMIIREG
3392                  * instead of SIOCGMIIPHY.
3393                  */
3394                 mii = if_mii(ifr);
3395                 if (!mii)
3396                         return -EINVAL;
3397
3398                 if (mii->reg_num == 1) {
3399                         mii->val_out = 0;
3400                         if (netif_carrier_ok(bond->dev))
3401                                 mii->val_out = BMSR_LSTATUS;
3402                 }
3403
3404                 return 0;
3405         case BOND_INFO_QUERY_OLD:
3406         case SIOCBONDINFOQUERY:
3407                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3408
3409                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3410                         return -EFAULT;
3411
3412                 res = bond_info_query(bond_dev, &k_binfo);
3413                 if (res == 0 &&
3414                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3415                         return -EFAULT;
3416
3417                 return res;
3418         case BOND_SLAVE_INFO_QUERY_OLD:
3419         case SIOCBONDSLAVEINFOQUERY:
3420                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3421
3422                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3423                         return -EFAULT;
3424
3425                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3426                 if (res == 0 &&
3427                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3428                         return -EFAULT;
3429
3430                 return res;
3431         default:
3432                 break;
3433         }
3434
3435         net = dev_net(bond_dev);
3436
3437         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3438                 return -EPERM;
3439
3440         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3441
3442         netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3443
3444         if (!slave_dev)
3445                 return -ENODEV;
3446
3447         netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3448         switch (cmd) {
3449         case BOND_ENSLAVE_OLD:
3450         case SIOCBONDENSLAVE:
3451                 res = bond_enslave(bond_dev, slave_dev);
3452                 break;
3453         case BOND_RELEASE_OLD:
3454         case SIOCBONDRELEASE:
3455                 res = bond_release(bond_dev, slave_dev);
3456                 break;
3457         case BOND_SETHWADDR_OLD:
3458         case SIOCBONDSETHWADDR:
3459                 bond_set_dev_addr(bond_dev, slave_dev);
3460                 res = 0;
3461                 break;
3462         case BOND_CHANGE_ACTIVE_OLD:
3463         case SIOCBONDCHANGEACTIVE:
3464                 bond_opt_initstr(&newval, slave_dev->name);
3465                 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3466                 break;
3467         default:
3468                 res = -EOPNOTSUPP;
3469         }
3470
3471         return res;
3472 }
3473
3474 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3475 {
3476         struct bonding *bond = netdev_priv(bond_dev);
3477
3478         if (change & IFF_PROMISC)
3479                 bond_set_promiscuity(bond,
3480                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3481
3482         if (change & IFF_ALLMULTI)
3483                 bond_set_allmulti(bond,
3484                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3485 }
3486
3487 static void bond_set_rx_mode(struct net_device *bond_dev)
3488 {
3489         struct bonding *bond = netdev_priv(bond_dev);
3490         struct list_head *iter;
3491         struct slave *slave;
3492
3493         rcu_read_lock();
3494         if (bond_uses_primary(bond)) {
3495                 slave = rcu_dereference(bond->curr_active_slave);
3496                 if (slave) {
3497                         dev_uc_sync(slave->dev, bond_dev);
3498                         dev_mc_sync(slave->dev, bond_dev);
3499                 }
3500         } else {
3501                 bond_for_each_slave_rcu(bond, slave, iter) {
3502                         dev_uc_sync_multiple(slave->dev, bond_dev);
3503                         dev_mc_sync_multiple(slave->dev, bond_dev);
3504                 }
3505         }
3506         rcu_read_unlock();
3507 }
3508
3509 static int bond_neigh_init(struct neighbour *n)
3510 {
3511         struct bonding *bond = netdev_priv(n->dev);
3512         const struct net_device_ops *slave_ops;
3513         struct neigh_parms parms;
3514         struct slave *slave;
3515         int ret;
3516
3517         slave = bond_first_slave(bond);
3518         if (!slave)
3519                 return 0;
3520         slave_ops = slave->dev->netdev_ops;
3521         if (!slave_ops->ndo_neigh_setup)
3522                 return 0;
3523
3524         parms.neigh_setup = NULL;
3525         parms.neigh_cleanup = NULL;
3526         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3527         if (ret)
3528                 return ret;
3529
3530         /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3531          * after the last slave has been detached.  Assumes that all slaves
3532          * utilize the same neigh_cleanup (true at this writing as only user
3533          * is ipoib).
3534          */
3535         n->parms->neigh_cleanup = parms.neigh_cleanup;
3536
3537         if (!parms.neigh_setup)
3538                 return 0;
3539
3540         return parms.neigh_setup(n);
3541 }
3542
3543 /* The bonding ndo_neigh_setup is called at init time beofre any
3544  * slave exists. So we must declare proxy setup function which will
3545  * be used at run time to resolve the actual slave neigh param setup.
3546  *
3547  * It's also called by master devices (such as vlans) to setup their
3548  * underlying devices. In that case - do nothing, we're already set up from
3549  * our init.
3550  */
3551 static int bond_neigh_setup(struct net_device *dev,
3552                             struct neigh_parms *parms)
3553 {
3554         /* modify only our neigh_parms */
3555         if (parms->dev == dev)
3556                 parms->neigh_setup = bond_neigh_init;
3557
3558         return 0;
3559 }
3560
3561 /* Change the MTU of all of a master's slaves to match the master */
3562 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3563 {
3564         struct bonding *bond = netdev_priv(bond_dev);
3565         struct slave *slave, *rollback_slave;
3566         struct list_head *iter;
3567         int res = 0;
3568
3569         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3570
3571         bond_for_each_slave(bond, slave, iter) {
3572                 netdev_dbg(bond_dev, "s %p c_m %p\n",
3573                            slave, slave->dev->netdev_ops->ndo_change_mtu);
3574
3575                 res = dev_set_mtu(slave->dev, new_mtu);
3576
3577                 if (res) {
3578                         /* If we failed to set the slave's mtu to the new value
3579                          * we must abort the operation even in ACTIVE_BACKUP
3580                          * mode, because if we allow the backup slaves to have
3581                          * different mtu values than the active slave we'll
3582                          * need to change their mtu when doing a failover. That
3583                          * means changing their mtu from timer context, which
3584                          * is probably not a good idea.
3585                          */
3586                         netdev_dbg(bond_dev, "err %d %s\n", res,
3587                                    slave->dev->name);
3588                         goto unwind;
3589                 }
3590         }
3591
3592         bond_dev->mtu = new_mtu;
3593
3594         return 0;
3595
3596 unwind:
3597         /* unwind from head to the slave that failed */
3598         bond_for_each_slave(bond, rollback_slave, iter) {
3599                 int tmp_res;
3600
3601                 if (rollback_slave == slave)
3602                         break;
3603
3604                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3605                 if (tmp_res) {
3606                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3607                                    tmp_res, rollback_slave->dev->name);
3608                 }
3609         }
3610
3611         return res;
3612 }
3613
3614 /* Change HW address
3615  *
3616  * Note that many devices must be down to change the HW address, and
3617  * downing the master releases all slaves.  We can make bonds full of
3618  * bonding devices to test this, however.
3619  */
3620 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3621 {
3622         struct bonding *bond = netdev_priv(bond_dev);
3623         struct slave *slave, *rollback_slave;
3624         struct sockaddr *sa = addr, tmp_sa;
3625         struct list_head *iter;
3626         int res = 0;
3627
3628         if (BOND_MODE(bond) == BOND_MODE_ALB)
3629                 return bond_alb_set_mac_address(bond_dev, addr);
3630
3631
3632         netdev_dbg(bond_dev, "bond=%p\n", bond);
3633
3634         /* If fail_over_mac is enabled, do nothing and return success.
3635          * Returning an error causes ifenslave to fail.
3636          */
3637         if (bond->params.fail_over_mac &&
3638             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3639                 return 0;
3640
3641         if (!is_valid_ether_addr(sa->sa_data))
3642                 return -EADDRNOTAVAIL;
3643
3644         bond_for_each_slave(bond, slave, iter) {
3645                 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3646                 res = dev_set_mac_address(slave->dev, addr);
3647                 if (res) {
3648                         /* TODO: consider downing the slave
3649                          * and retry ?
3650                          * User should expect communications
3651                          * breakage anyway until ARP finish
3652                          * updating, so...
3653                          */
3654                         netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3655                         goto unwind;
3656                 }
3657         }
3658
3659         /* success */
3660         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3661         return 0;
3662
3663 unwind:
3664         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3665         tmp_sa.sa_family = bond_dev->type;
3666
3667         /* unwind from head to the slave that failed */
3668         bond_for_each_slave(bond, rollback_slave, iter) {
3669                 int tmp_res;
3670
3671                 if (rollback_slave == slave)
3672                         break;
3673
3674                 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3675                 if (tmp_res) {
3676                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3677                                    tmp_res, rollback_slave->dev->name);
3678                 }
3679         }
3680
3681         return res;
3682 }
3683
3684 /**
3685  * bond_xmit_slave_id - transmit skb through slave with slave_id
3686  * @bond: bonding device that is transmitting
3687  * @skb: buffer to transmit
3688  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3689  *
3690  * This function tries to transmit through slave with slave_id but in case
3691  * it fails, it tries to find the first available slave for transmission.
3692  * The skb is consumed in all cases, thus the function is void.
3693  */
3694 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3695 {
3696         struct list_head *iter;
3697         struct slave *slave;
3698         int i = slave_id;
3699
3700         /* Here we start from the slave with slave_id */
3701         bond_for_each_slave_rcu(bond, slave, iter) {
3702                 if (--i < 0) {
3703                         if (bond_slave_can_tx(slave)) {
3704                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3705                                 return;
3706                         }
3707                 }
3708         }
3709
3710         /* Here we start from the first slave up to slave_id */
3711         i = slave_id;
3712         bond_for_each_slave_rcu(bond, slave, iter) {
3713                 if (--i < 0)
3714                         break;
3715                 if (bond_slave_can_tx(slave)) {
3716                         bond_dev_queue_xmit(bond, skb, slave->dev);
3717                         return;
3718                 }
3719         }
3720         /* no slave that can tx has been found */
3721         bond_tx_drop(bond->dev, skb);
3722 }
3723
3724 /**
3725  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3726  * @bond: bonding device to use
3727  *
3728  * Based on the value of the bonding device's packets_per_slave parameter
3729  * this function generates a slave id, which is usually used as the next
3730  * slave to transmit through.
3731  */
3732 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3733 {
3734         u32 slave_id;
3735         struct reciprocal_value reciprocal_packets_per_slave;
3736         int packets_per_slave = bond->params.packets_per_slave;
3737
3738         switch (packets_per_slave) {
3739         case 0:
3740                 slave_id = prandom_u32();
3741                 break;
3742         case 1:
3743                 slave_id = bond->rr_tx_counter;
3744                 break;
3745         default:
3746                 reciprocal_packets_per_slave =
3747                         bond->params.reciprocal_packets_per_slave;
3748                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3749                                              reciprocal_packets_per_slave);
3750                 break;
3751         }
3752         bond->rr_tx_counter++;
3753
3754         return slave_id;
3755 }
3756
3757 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3758 {
3759         struct bonding *bond = netdev_priv(bond_dev);
3760         struct iphdr *iph = ip_hdr(skb);
3761         struct slave *slave;
3762         u32 slave_id;
3763
3764         /* Start with the curr_active_slave that joined the bond as the
3765          * default for sending IGMP traffic.  For failover purposes one
3766          * needs to maintain some consistency for the interface that will
3767          * send the join/membership reports.  The curr_active_slave found
3768          * will send all of this type of traffic.
3769          */
3770         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3771                 slave = rcu_dereference(bond->curr_active_slave);
3772                 if (slave)
3773                         bond_dev_queue_xmit(bond, skb, slave->dev);
3774                 else
3775                         bond_xmit_slave_id(bond, skb, 0);
3776         } else {
3777                 int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3778
3779                 if (likely(slave_cnt)) {
3780                         slave_id = bond_rr_gen_slave_id(bond);
3781                         bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3782                 } else {
3783                         bond_tx_drop(bond_dev, skb);
3784                 }
3785         }
3786
3787         return NETDEV_TX_OK;
3788 }
3789
3790 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3791  * the bond has a usable interface.
3792  */
3793 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3794 {
3795         struct bonding *bond = netdev_priv(bond_dev);
3796         struct slave *slave;
3797
3798         slave = rcu_dereference(bond->curr_active_slave);
3799         if (slave)
3800                 bond_dev_queue_xmit(bond, skb, slave->dev);
3801         else
3802                 bond_tx_drop(bond_dev, skb);
3803
3804         return NETDEV_TX_OK;
3805 }
3806
3807 /* Use this to update slave_array when (a) it's not appropriate to update
3808  * slave_array right away (note that update_slave_array() may sleep)
3809  * and / or (b) RTNL is not held.
3810  */
3811 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3812 {
3813         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3814 }
3815
3816 /* Slave array work handler. Holds only RTNL */
3817 static void bond_slave_arr_handler(struct work_struct *work)
3818 {
3819         struct bonding *bond = container_of(work, struct bonding,
3820                                             slave_arr_work.work);
3821         int ret;
3822
3823         if (!rtnl_trylock())
3824                 goto err;
3825
3826         ret = bond_update_slave_arr(bond, NULL);
3827         rtnl_unlock();
3828         if (ret) {
3829                 pr_warn_ratelimited("Failed to update slave array from WT\n");
3830                 goto err;
3831         }
3832         return;
3833
3834 err:
3835         bond_slave_arr_work_rearm(bond, 1);
3836 }
3837
3838 /* Build the usable slaves array in control path for modes that use xmit-hash
3839  * to determine the slave interface -
3840  * (a) BOND_MODE_8023AD
3841  * (b) BOND_MODE_XOR
3842  * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3843  *
3844  * The caller is expected to hold RTNL only and NO other lock!
3845  */
3846 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3847 {
3848         struct slave *slave;
3849         struct list_head *iter;
3850         struct bond_up_slave *new_arr, *old_arr;
3851         int agg_id = 0;
3852         int ret = 0;
3853
3854 #ifdef CONFIG_LOCKDEP
3855         WARN_ON(lockdep_is_held(&bond->mode_lock));
3856 #endif
3857
3858         new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3859                           GFP_KERNEL);
3860         if (!new_arr) {
3861                 ret = -ENOMEM;
3862                 pr_err("Failed to build slave-array.\n");
3863                 goto out;
3864         }
3865         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3866                 struct ad_info ad_info;
3867
3868                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3869                         pr_debug("bond_3ad_get_active_agg_info failed\n");
3870                         kfree_rcu(new_arr, rcu);
3871                         /* No active aggragator means it's not safe to use
3872                          * the previous array.
3873                          */
3874                         old_arr = rtnl_dereference(bond->slave_arr);
3875                         if (old_arr) {
3876                                 RCU_INIT_POINTER(bond->slave_arr, NULL);
3877                                 kfree_rcu(old_arr, rcu);
3878                         }
3879                         goto out;
3880                 }
3881                 agg_id = ad_info.aggregator_id;
3882         }
3883         bond_for_each_slave(bond, slave, iter) {
3884                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3885                         struct aggregator *agg;
3886
3887                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
3888                         if (!agg || agg->aggregator_identifier != agg_id)
3889                                 continue;
3890                 }
3891                 if (!bond_slave_can_tx(slave))
3892                         continue;
3893                 if (skipslave == slave)
3894                         continue;
3895                 new_arr->arr[new_arr->count++] = slave;
3896         }
3897
3898         old_arr = rtnl_dereference(bond->slave_arr);
3899         rcu_assign_pointer(bond->slave_arr, new_arr);
3900         if (old_arr)
3901                 kfree_rcu(old_arr, rcu);
3902 out:
3903         if (ret != 0 && skipslave) {
3904                 int idx;
3905
3906                 /* Rare situation where caller has asked to skip a specific
3907                  * slave but allocation failed (most likely!). BTW this is
3908                  * only possible when the call is initiated from
3909                  * __bond_release_one(). In this situation; overwrite the
3910                  * skipslave entry in the array with the last entry from the
3911                  * array to avoid a situation where the xmit path may choose
3912                  * this to-be-skipped slave to send a packet out.
3913                  */
3914                 old_arr = rtnl_dereference(bond->slave_arr);
3915                 for (idx = 0; idx < old_arr->count; idx++) {
3916                         if (skipslave == old_arr->arr[idx]) {
3917                                 old_arr->arr[idx] =
3918                                     old_arr->arr[old_arr->count-1];
3919                                 old_arr->count--;
3920                                 break;
3921                         }
3922                 }
3923         }
3924         return ret;
3925 }
3926
3927 /* Use this Xmit function for 3AD as well as XOR modes. The current
3928  * usable slave array is formed in the control path. The xmit function
3929  * just calculates hash and sends the packet out.
3930  */
3931 static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3932 {
3933         struct bonding *bond = netdev_priv(dev);
3934         struct slave *slave;
3935         struct bond_up_slave *slaves;
3936         unsigned int count;
3937
3938         slaves = rcu_dereference(bond->slave_arr);
3939         count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3940         if (likely(count)) {
3941                 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3942                 bond_dev_queue_xmit(bond, skb, slave->dev);
3943         } else {
3944                 bond_tx_drop(dev, skb);
3945         }
3946
3947         return NETDEV_TX_OK;
3948 }
3949
3950 /* in broadcast mode, we send everything to all usable interfaces. */
3951 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3952 {
3953         struct bonding *bond = netdev_priv(bond_dev);
3954         struct slave *slave = NULL;
3955         struct list_head *iter;
3956
3957         bond_for_each_slave_rcu(bond, slave, iter) {
3958                 if (bond_is_last_slave(bond, slave))
3959                         break;
3960                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3961                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3962
3963                         if (!skb2) {
3964                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3965                                                     bond_dev->name, __func__);
3966                                 continue;
3967                         }
3968                         bond_dev_queue_xmit(bond, skb2, slave->dev);
3969                 }
3970         }
3971         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3972                 bond_dev_queue_xmit(bond, skb, slave->dev);
3973         else
3974                 bond_tx_drop(bond_dev, skb);
3975
3976         return NETDEV_TX_OK;
3977 }
3978
3979 /*------------------------- Device initialization ---------------------------*/
3980
3981 /* Lookup the slave that corresponds to a qid */
3982 static inline int bond_slave_override(struct bonding *bond,
3983                                       struct sk_buff *skb)
3984 {
3985         struct slave *slave = NULL;
3986         struct list_head *iter;
3987
3988         if (!skb->queue_mapping)
3989                 return 1;
3990
3991         /* Find out if any slaves have the same mapping as this skb. */
3992         bond_for_each_slave_rcu(bond, slave, iter) {
3993                 if (slave->queue_id == skb->queue_mapping) {
3994                         if (bond_slave_is_up(slave) &&
3995                             slave->link == BOND_LINK_UP) {
3996                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3997                                 return 0;
3998                         }
3999                         /* If the slave isn't UP, use default transmit policy. */
4000                         break;
4001                 }
4002         }
4003
4004         return 1;
4005 }
4006
4007
4008 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4009                              void *accel_priv, select_queue_fallback_t fallback)
4010 {
4011         /* This helper function exists to help dev_pick_tx get the correct
4012          * destination queue.  Using a helper function skips a call to
4013          * skb_tx_hash and will put the skbs in the queue we expect on their
4014          * way down to the bonding driver.
4015          */
4016         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4017
4018         /* Save the original txq to restore before passing to the driver */
4019         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4020
4021         if (unlikely(txq >= dev->real_num_tx_queues)) {
4022                 do {
4023                         txq -= dev->real_num_tx_queues;
4024                 } while (txq >= dev->real_num_tx_queues);
4025         }
4026         return txq;
4027 }
4028
4029 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4030 {
4031         struct bonding *bond = netdev_priv(dev);
4032
4033         if (bond_should_override_tx_queue(bond) &&
4034             !bond_slave_override(bond, skb))
4035                 return NETDEV_TX_OK;
4036
4037         switch (BOND_MODE(bond)) {
4038         case BOND_MODE_ROUNDROBIN:
4039                 return bond_xmit_roundrobin(skb, dev);
4040         case BOND_MODE_ACTIVEBACKUP:
4041                 return bond_xmit_activebackup(skb, dev);
4042         case BOND_MODE_8023AD:
4043         case BOND_MODE_XOR:
4044                 return bond_3ad_xor_xmit(skb, dev);
4045         case BOND_MODE_BROADCAST:
4046                 return bond_xmit_broadcast(skb, dev);
4047         case BOND_MODE_ALB:
4048                 return bond_alb_xmit(skb, dev);
4049         case BOND_MODE_TLB:
4050                 return bond_tlb_xmit(skb, dev);
4051         default:
4052                 /* Should never happen, mode already checked */
4053                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4054                 WARN_ON_ONCE(1);
4055                 bond_tx_drop(dev, skb);
4056                 return NETDEV_TX_OK;
4057         }
4058 }
4059
4060 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4061 {
4062         struct bonding *bond = netdev_priv(dev);
4063         netdev_tx_t ret = NETDEV_TX_OK;
4064
4065         /* If we risk deadlock from transmitting this in the
4066          * netpoll path, tell netpoll to queue the frame for later tx
4067          */
4068         if (unlikely(is_netpoll_tx_blocked(dev)))
4069                 return NETDEV_TX_BUSY;
4070
4071         rcu_read_lock();
4072         if (bond_has_slaves(bond))
4073                 ret = __bond_start_xmit(skb, dev);
4074         else
4075                 bond_tx_drop(dev, skb);
4076         rcu_read_unlock();
4077
4078         return ret;
4079 }
4080
4081 static int bond_ethtool_get_settings(struct net_device *bond_dev,
4082                                      struct ethtool_cmd *ecmd)
4083 {
4084         struct bonding *bond = netdev_priv(bond_dev);
4085         unsigned long speed = 0;
4086         struct list_head *iter;
4087         struct slave *slave;
4088
4089         ecmd->duplex = DUPLEX_UNKNOWN;
4090         ecmd->port = PORT_OTHER;
4091
4092         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4093          * do not need to check mode.  Though link speed might not represent
4094          * the true receive or transmit bandwidth (not all modes are symmetric)
4095          * this is an accurate maximum.
4096          */
4097         bond_for_each_slave(bond, slave, iter) {
4098                 if (bond_slave_can_tx(slave)) {
4099                         if (slave->speed != SPEED_UNKNOWN)
4100                                 speed += slave->speed;
4101                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
4102                             slave->duplex != DUPLEX_UNKNOWN)
4103                                 ecmd->duplex = slave->duplex;
4104                 }
4105         }
4106         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
4107
4108         return 0;
4109 }
4110
4111 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4112                                      struct ethtool_drvinfo *drvinfo)
4113 {
4114         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4115         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4116         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4117                  BOND_ABI_VERSION);
4118 }
4119
4120 static const struct ethtool_ops bond_ethtool_ops = {
4121         .get_drvinfo            = bond_ethtool_get_drvinfo,
4122         .get_settings           = bond_ethtool_get_settings,
4123         .get_link               = ethtool_op_get_link,
4124 };
4125
4126 static const struct net_device_ops bond_netdev_ops = {
4127         .ndo_init               = bond_init,
4128         .ndo_uninit             = bond_uninit,
4129         .ndo_open               = bond_open,
4130         .ndo_stop               = bond_close,
4131         .ndo_start_xmit         = bond_start_xmit,
4132         .ndo_select_queue       = bond_select_queue,
4133         .ndo_get_stats64        = bond_get_stats,
4134         .ndo_do_ioctl           = bond_do_ioctl,
4135         .ndo_change_rx_flags    = bond_change_rx_flags,
4136         .ndo_set_rx_mode        = bond_set_rx_mode,
4137         .ndo_change_mtu         = bond_change_mtu,
4138         .ndo_set_mac_address    = bond_set_mac_address,
4139         .ndo_neigh_setup        = bond_neigh_setup,
4140         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4141         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4142 #ifdef CONFIG_NET_POLL_CONTROLLER
4143         .ndo_netpoll_setup      = bond_netpoll_setup,
4144         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4145         .ndo_poll_controller    = bond_poll_controller,
4146 #endif
4147         .ndo_add_slave          = bond_enslave,
4148         .ndo_del_slave          = bond_release,
4149         .ndo_fix_features       = bond_fix_features,
4150         .ndo_neigh_construct    = netdev_default_l2upper_neigh_construct,
4151         .ndo_neigh_destroy      = netdev_default_l2upper_neigh_destroy,
4152         .ndo_bridge_setlink     = switchdev_port_bridge_setlink,
4153         .ndo_bridge_getlink     = switchdev_port_bridge_getlink,
4154         .ndo_bridge_dellink     = switchdev_port_bridge_dellink,
4155         .ndo_fdb_add            = switchdev_port_fdb_add,
4156         .ndo_fdb_del            = switchdev_port_fdb_del,
4157         .ndo_fdb_dump           = switchdev_port_fdb_dump,
4158         .ndo_features_check     = passthru_features_check,
4159 };
4160
4161 static const struct device_type bond_type = {
4162         .name = "bond",
4163 };
4164
4165 static void bond_destructor(struct net_device *bond_dev)
4166 {
4167         struct bonding *bond = netdev_priv(bond_dev);
4168         if (bond->wq)
4169                 destroy_workqueue(bond->wq);
4170         free_netdev(bond_dev);
4171 }
4172
4173 void bond_setup(struct net_device *bond_dev)
4174 {
4175         struct bonding *bond = netdev_priv(bond_dev);
4176
4177         spin_lock_init(&bond->mode_lock);
4178         spin_lock_init(&bond->stats_lock);
4179         bond->params = bonding_defaults;
4180
4181         /* Initialize pointers */
4182         bond->dev = bond_dev;
4183
4184         /* Initialize the device entry points */
4185         ether_setup(bond_dev);
4186         bond_dev->netdev_ops = &bond_netdev_ops;
4187         bond_dev->ethtool_ops = &bond_ethtool_ops;
4188
4189         bond_dev->destructor = bond_destructor;
4190
4191         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4192
4193         /* Initialize the device options */
4194         bond_dev->flags |= IFF_MASTER;
4195         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4196         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4197
4198         /* don't acquire bond device's netif_tx_lock when transmitting */
4199         bond_dev->features |= NETIF_F_LLTX;
4200
4201         /* By default, we declare the bond to be fully
4202          * VLAN hardware accelerated capable. Special
4203          * care is taken in the various xmit functions
4204          * when there are slaves that are not hw accel
4205          * capable
4206          */
4207
4208         /* Don't allow bond devices to change network namespaces. */
4209         bond_dev->features |= NETIF_F_NETNS_LOCAL;
4210
4211         bond_dev->hw_features = BOND_VLAN_FEATURES |
4212                                 NETIF_F_HW_VLAN_CTAG_TX |
4213                                 NETIF_F_HW_VLAN_CTAG_RX |
4214                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4215
4216         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4217         bond_dev->features |= bond_dev->hw_features;
4218 }
4219
4220 /* Destroy a bonding device.
4221  * Must be under rtnl_lock when this function is called.
4222  */
4223 static void bond_uninit(struct net_device *bond_dev)
4224 {
4225         struct bonding *bond = netdev_priv(bond_dev);
4226         struct list_head *iter;
4227         struct slave *slave;
4228         struct bond_up_slave *arr;
4229
4230         bond_netpoll_cleanup(bond_dev);
4231
4232         /* Release the bonded slaves */
4233         bond_for_each_slave(bond, slave, iter)
4234                 __bond_release_one(bond_dev, slave->dev, true);
4235         netdev_info(bond_dev, "Released all slaves\n");
4236
4237         arr = rtnl_dereference(bond->slave_arr);
4238         if (arr) {
4239                 RCU_INIT_POINTER(bond->slave_arr, NULL);
4240                 kfree_rcu(arr, rcu);
4241         }
4242
4243         list_del(&bond->bond_list);
4244
4245         bond_debug_unregister(bond);
4246 }
4247
4248 /*------------------------- Module initialization ---------------------------*/
4249
4250 static int bond_check_params(struct bond_params *params)
4251 {
4252         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4253         struct bond_opt_value newval;
4254         const struct bond_opt_value *valptr;
4255         int arp_all_targets_value;
4256         u16 ad_actor_sys_prio = 0;
4257         u16 ad_user_port_key = 0;
4258
4259         /* Convert string parameters. */
4260         if (mode) {
4261                 bond_opt_initstr(&newval, mode);
4262                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4263                 if (!valptr) {
4264                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4265                         return -EINVAL;
4266                 }
4267                 bond_mode = valptr->value;
4268         }
4269
4270         if (xmit_hash_policy) {
4271                 if ((bond_mode != BOND_MODE_XOR) &&
4272                     (bond_mode != BOND_MODE_8023AD) &&
4273                     (bond_mode != BOND_MODE_TLB)) {
4274                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4275                                 bond_mode_name(bond_mode));
4276                 } else {
4277                         bond_opt_initstr(&newval, xmit_hash_policy);
4278                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4279                                                 &newval);
4280                         if (!valptr) {
4281                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4282                                        xmit_hash_policy);
4283                                 return -EINVAL;
4284                         }
4285                         xmit_hashtype = valptr->value;
4286                 }
4287         }
4288
4289         if (lacp_rate) {
4290                 if (bond_mode != BOND_MODE_8023AD) {
4291                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4292                                 bond_mode_name(bond_mode));
4293                 } else {
4294                         bond_opt_initstr(&newval, lacp_rate);
4295                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4296                                                 &newval);
4297                         if (!valptr) {
4298                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4299                                        lacp_rate);
4300                                 return -EINVAL;
4301                         }
4302                         lacp_fast = valptr->value;
4303                 }
4304         }
4305
4306         if (ad_select) {
4307                 bond_opt_initstr(&newval, ad_select);
4308                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4309                                         &newval);
4310                 if (!valptr) {
4311                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4312                         return -EINVAL;
4313                 }
4314                 params->ad_select = valptr->value;
4315                 if (bond_mode != BOND_MODE_8023AD)
4316                         pr_warn("ad_select param only affects 802.3ad mode\n");
4317         } else {
4318                 params->ad_select = BOND_AD_STABLE;
4319         }
4320
4321         if (max_bonds < 0) {
4322                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4323                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4324                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4325         }
4326
4327         if (miimon < 0) {
4328                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4329                         miimon, INT_MAX);
4330                 miimon = 0;
4331         }
4332
4333         if (updelay < 0) {
4334                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4335                         updelay, INT_MAX);
4336                 updelay = 0;
4337         }
4338
4339         if (downdelay < 0) {
4340                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4341                         downdelay, INT_MAX);
4342                 downdelay = 0;
4343         }
4344
4345         if ((use_carrier != 0) && (use_carrier != 1)) {
4346                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4347                         use_carrier);
4348                 use_carrier = 1;
4349         }
4350
4351         if (num_peer_notif < 0 || num_peer_notif > 255) {
4352                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4353                         num_peer_notif);
4354                 num_peer_notif = 1;
4355         }
4356
4357         /* reset values for 802.3ad/TLB/ALB */
4358         if (!bond_mode_uses_arp(bond_mode)) {
4359                 if (!miimon) {
4360                         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");
4361                         pr_warn("Forcing miimon to 100msec\n");
4362                         miimon = BOND_DEFAULT_MIIMON;
4363                 }
4364         }
4365
4366         if (tx_queues < 1 || tx_queues > 255) {
4367                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4368                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4369                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4370         }
4371
4372         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4373                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4374                         all_slaves_active);
4375                 all_slaves_active = 0;
4376         }
4377
4378         if (resend_igmp < 0 || resend_igmp > 255) {
4379                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4380                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4381                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4382         }
4383
4384         bond_opt_initval(&newval, packets_per_slave);
4385         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4386                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4387                         packets_per_slave, USHRT_MAX);
4388                 packets_per_slave = 1;
4389         }
4390
4391         if (bond_mode == BOND_MODE_ALB) {
4392                 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",
4393                           updelay);
4394         }
4395
4396         if (!miimon) {
4397                 if (updelay || downdelay) {
4398                         /* just warn the user the up/down delay will have
4399                          * no effect since miimon is zero...
4400                          */
4401                         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",
4402                                 updelay, downdelay);
4403                 }
4404         } else {
4405                 /* don't allow arp monitoring */
4406                 if (arp_interval) {
4407                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4408                                 miimon, arp_interval);
4409                         arp_interval = 0;
4410                 }
4411
4412                 if ((updelay % miimon) != 0) {
4413                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4414                                 updelay, miimon, (updelay / miimon) * miimon);
4415                 }
4416
4417                 updelay /= miimon;
4418
4419                 if ((downdelay % miimon) != 0) {
4420                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4421                                 downdelay, miimon,
4422                                 (downdelay / miimon) * miimon);
4423                 }
4424
4425                 downdelay /= miimon;
4426         }
4427
4428         if (arp_interval < 0) {
4429                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4430                         arp_interval, INT_MAX);
4431                 arp_interval = 0;
4432         }
4433
4434         for (arp_ip_count = 0, i = 0;
4435              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4436                 __be32 ip;
4437
4438                 /* not a complete check, but good enough to catch mistakes */
4439                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4440                     !bond_is_ip_target_ok(ip)) {
4441                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4442                                 arp_ip_target[i]);
4443                         arp_interval = 0;
4444                 } else {
4445                         if (bond_get_targets_ip(arp_target, ip) == -1)
4446                                 arp_target[arp_ip_count++] = ip;
4447                         else
4448                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4449                                         &ip);
4450                 }
4451         }
4452
4453         if (arp_interval && !arp_ip_count) {
4454                 /* don't allow arping if no arp_ip_target given... */
4455                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4456                         arp_interval);
4457                 arp_interval = 0;
4458         }
4459
4460         if (arp_validate) {
4461                 if (!arp_interval) {
4462                         pr_err("arp_validate requires arp_interval\n");
4463                         return -EINVAL;
4464                 }
4465
4466                 bond_opt_initstr(&newval, arp_validate);
4467                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4468                                         &newval);
4469                 if (!valptr) {
4470                         pr_err("Error: invalid arp_validate \"%s\"\n",
4471                                arp_validate);
4472                         return -EINVAL;
4473                 }
4474                 arp_validate_value = valptr->value;
4475         } else {
4476                 arp_validate_value = 0;
4477         }
4478
4479         arp_all_targets_value = 0;
4480         if (arp_all_targets) {
4481                 bond_opt_initstr(&newval, arp_all_targets);
4482                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4483                                         &newval);
4484                 if (!valptr) {
4485                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4486                                arp_all_targets);
4487                         arp_all_targets_value = 0;
4488                 } else {
4489                         arp_all_targets_value = valptr->value;
4490                 }
4491         }
4492
4493         if (miimon) {
4494                 pr_info("MII link monitoring set to %d ms\n", miimon);
4495         } else if (arp_interval) {
4496                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4497                                           arp_validate_value);
4498                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4499                         arp_interval, valptr->string, arp_ip_count);
4500
4501                 for (i = 0; i < arp_ip_count; i++)
4502                         pr_cont(" %s", arp_ip_target[i]);
4503
4504                 pr_cont("\n");
4505
4506         } else if (max_bonds) {
4507                 /* miimon and arp_interval not set, we need one so things
4508                  * work as expected, see bonding.txt for details
4509                  */
4510                 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");
4511         }
4512
4513         if (primary && !bond_mode_uses_primary(bond_mode)) {
4514                 /* currently, using a primary only makes sense
4515                  * in active backup, TLB or ALB modes
4516                  */
4517                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4518                         primary, bond_mode_name(bond_mode));
4519                 primary = NULL;
4520         }
4521
4522         if (primary && primary_reselect) {
4523                 bond_opt_initstr(&newval, primary_reselect);
4524                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4525                                         &newval);
4526                 if (!valptr) {
4527                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4528                                primary_reselect);
4529                         return -EINVAL;
4530                 }
4531                 primary_reselect_value = valptr->value;
4532         } else {
4533                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4534         }
4535
4536         if (fail_over_mac) {
4537                 bond_opt_initstr(&newval, fail_over_mac);
4538                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4539                                         &newval);
4540                 if (!valptr) {
4541                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4542                                fail_over_mac);
4543                         return -EINVAL;
4544                 }
4545                 fail_over_mac_value = valptr->value;
4546                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4547                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4548         } else {
4549                 fail_over_mac_value = BOND_FOM_NONE;
4550         }
4551
4552         bond_opt_initstr(&newval, "default");
4553         valptr = bond_opt_parse(
4554                         bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4555                                      &newval);
4556         if (!valptr) {
4557                 pr_err("Error: No ad_actor_sys_prio default value");
4558                 return -EINVAL;
4559         }
4560         ad_actor_sys_prio = valptr->value;
4561
4562         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4563                                 &newval);
4564         if (!valptr) {
4565                 pr_err("Error: No ad_user_port_key default value");
4566                 return -EINVAL;
4567         }
4568         ad_user_port_key = valptr->value;
4569
4570         if (lp_interval == 0) {
4571                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4572                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4573                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4574         }
4575
4576         /* fill params struct with the proper values */
4577         params->mode = bond_mode;
4578         params->xmit_policy = xmit_hashtype;
4579         params->miimon = miimon;
4580         params->num_peer_notif = num_peer_notif;
4581         params->arp_interval = arp_interval;
4582         params->arp_validate = arp_validate_value;
4583         params->arp_all_targets = arp_all_targets_value;
4584         params->updelay = updelay;
4585         params->downdelay = downdelay;
4586         params->use_carrier = use_carrier;
4587         params->lacp_fast = lacp_fast;
4588         params->primary[0] = 0;
4589         params->primary_reselect = primary_reselect_value;
4590         params->fail_over_mac = fail_over_mac_value;
4591         params->tx_queues = tx_queues;
4592         params->all_slaves_active = all_slaves_active;
4593         params->resend_igmp = resend_igmp;
4594         params->min_links = min_links;
4595         params->lp_interval = lp_interval;
4596         params->packets_per_slave = packets_per_slave;
4597         params->tlb_dynamic_lb = 1; /* Default value */
4598         params->ad_actor_sys_prio = ad_actor_sys_prio;
4599         eth_zero_addr(params->ad_actor_system);
4600         params->ad_user_port_key = ad_user_port_key;
4601         if (packets_per_slave > 0) {
4602                 params->reciprocal_packets_per_slave =
4603                         reciprocal_value(packets_per_slave);
4604         } else {
4605                 /* reciprocal_packets_per_slave is unused if
4606                  * packets_per_slave is 0 or 1, just initialize it
4607                  */
4608                 params->reciprocal_packets_per_slave =
4609                         (struct reciprocal_value) { 0 };
4610         }
4611
4612         if (primary) {
4613                 strncpy(params->primary, primary, IFNAMSIZ);
4614                 params->primary[IFNAMSIZ - 1] = 0;
4615         }
4616
4617         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4618
4619         return 0;
4620 }
4621
4622 /* Called from registration process */
4623 static int bond_init(struct net_device *bond_dev)
4624 {
4625         struct bonding *bond = netdev_priv(bond_dev);
4626         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4627
4628         netdev_dbg(bond_dev, "Begin bond_init\n");
4629
4630         bond->wq = create_singlethread_workqueue(bond_dev->name);
4631         if (!bond->wq)
4632                 return -ENOMEM;
4633
4634         netdev_lockdep_set_classes(bond_dev);
4635
4636         list_add_tail(&bond->bond_list, &bn->dev_list);
4637
4638         bond_prepare_sysfs_group(bond);
4639
4640         bond_debug_register(bond);
4641
4642         /* Ensure valid dev_addr */
4643         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4644             bond_dev->addr_assign_type == NET_ADDR_PERM)
4645                 eth_hw_addr_random(bond_dev);
4646
4647         return 0;
4648 }
4649
4650 unsigned int bond_get_num_tx_queues(void)
4651 {
4652         return tx_queues;
4653 }
4654
4655 /* Create a new bond based on the specified name and bonding parameters.
4656  * If name is NULL, obtain a suitable "bond%d" name for us.
4657  * Caller must NOT hold rtnl_lock; we need to release it here before we
4658  * set up our sysfs entries.
4659  */
4660 int bond_create(struct net *net, const char *name)
4661 {
4662         struct net_device *bond_dev;
4663         struct bonding *bond;
4664         struct alb_bond_info *bond_info;
4665         int res;
4666
4667         rtnl_lock();
4668
4669         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4670                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
4671                                    bond_setup, tx_queues);
4672         if (!bond_dev) {
4673                 pr_err("%s: eek! can't alloc netdev!\n", name);
4674                 rtnl_unlock();
4675                 return -ENOMEM;
4676         }
4677
4678         /*
4679          * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4680          * It is set to 0 by default which is wrong.
4681          */
4682         bond = netdev_priv(bond_dev);
4683         bond_info = &(BOND_ALB_INFO(bond));
4684         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4685
4686         dev_net_set(bond_dev, net);
4687         bond_dev->rtnl_link_ops = &bond_link_ops;
4688
4689         res = register_netdevice(bond_dev);
4690
4691         netif_carrier_off(bond_dev);
4692
4693         rtnl_unlock();
4694         if (res < 0)
4695                 bond_destructor(bond_dev);
4696         return res;
4697 }
4698
4699 static int __net_init bond_net_init(struct net *net)
4700 {
4701         struct bond_net *bn = net_generic(net, bond_net_id);
4702
4703         bn->net = net;
4704         INIT_LIST_HEAD(&bn->dev_list);
4705
4706         bond_create_proc_dir(bn);
4707         bond_create_sysfs(bn);
4708
4709         return 0;
4710 }
4711
4712 static void __net_exit bond_net_exit(struct net *net)
4713 {
4714         struct bond_net *bn = net_generic(net, bond_net_id);
4715         struct bonding *bond, *tmp_bond;
4716         LIST_HEAD(list);
4717
4718         bond_destroy_sysfs(bn);
4719
4720         /* Kill off any bonds created after unregistering bond rtnl ops */
4721         rtnl_lock();
4722         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4723                 unregister_netdevice_queue(bond->dev, &list);
4724         unregister_netdevice_many(&list);
4725         rtnl_unlock();
4726
4727         bond_destroy_proc_dir(bn);
4728 }
4729
4730 static struct pernet_operations bond_net_ops = {
4731         .init = bond_net_init,
4732         .exit = bond_net_exit,
4733         .id   = &bond_net_id,
4734         .size = sizeof(struct bond_net),
4735 };
4736
4737 static int __init bonding_init(void)
4738 {
4739         int i;
4740         int res;
4741
4742         pr_info("%s", bond_version);
4743
4744         res = bond_check_params(&bonding_defaults);
4745         if (res)
4746                 goto out;
4747
4748         res = register_pernet_subsys(&bond_net_ops);
4749         if (res)
4750                 goto out;
4751
4752         res = bond_netlink_init();
4753         if (res)
4754                 goto err_link;
4755
4756         bond_create_debugfs();
4757
4758         for (i = 0; i < max_bonds; i++) {
4759                 res = bond_create(&init_net, NULL);
4760                 if (res)
4761                         goto err;
4762         }
4763
4764         register_netdevice_notifier(&bond_netdev_notifier);
4765 out:
4766         return res;
4767 err:
4768         bond_destroy_debugfs();
4769         bond_netlink_fini();
4770 err_link:
4771         unregister_pernet_subsys(&bond_net_ops);
4772         goto out;
4773
4774 }
4775
4776 static void __exit bonding_exit(void)
4777 {
4778         unregister_netdevice_notifier(&bond_netdev_notifier);
4779
4780         bond_destroy_debugfs();
4781
4782         bond_netlink_fini();
4783         unregister_pernet_subsys(&bond_net_ops);
4784
4785 #ifdef CONFIG_NET_POLL_CONTROLLER
4786         /* Make sure we don't have an imbalance on our netpoll blocking */
4787         WARN_ON(atomic_read(&netpoll_block_tx));
4788 #endif
4789 }
4790
4791 module_init(bonding_init);
4792 module_exit(bonding_exit);
4793 MODULE_LICENSE("GPL");
4794 MODULE_VERSION(DRV_VERSION);
4795 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4796 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");