kernel/drivers/net/bonding/bond_main.c

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 #include <linux/kernel.h>
  35 #include <linux/module.h>
  36 #include <linux/types.h>
  37 #include <linux/fcntl.h>
  38 #include <linux/interrupt.h>
  39 #include <linux/ptrace.h>
  40 #include <linux/ioport.h>
  41 #include <linux/in.h>
  42 #include <net/ip.h>
  43 #include <linux/ip.h>
  44 #include <linux/tcp.h>
  45 #include <linux/udp.h>
  46 #include <linux/slab.h>
  47 #include <linux/string.h>
  48 #include <linux/init.h>
  49 #include <linux/timer.h>
  50 #include <linux/socket.h>
  51 #include <linux/ctype.h>
  52 #include <linux/inet.h>
  53 #include <linux/bitops.h>
  54 #include <linux/io.h>
  55 #include <asm/dma.h>
  56 #include <linux/uaccess.h>
  57 #include <linux/errno.h>
  58 #include <linux/netdevice.h>
  59 #include <linux/inetdevice.h>
  60 #include <linux/igmp.h>
  61 #include <linux/etherdevice.h>
  62 #include <linux/skbuff.h>
  63 #include <net/sock.h>
  64 #include <linux/rtnetlink.h>
  65 #include <linux/smp.h>
  66 #include <linux/if_ether.h>
  67 #include <net/arp.h>
  68 #include <linux/mii.h>
  69 #include <linux/ethtool.h>
  70 #include <linux/if_vlan.h>
  71 #include <linux/if_bonding.h>
  72 #include <linux/jiffies.h>
  73 #include <linux/preempt.h>
  74 #include <net/route.h>
  75 #include <net/net_namespace.h>
  76 #include <net/netns/generic.h>
  77 #include <net/pkt_sched.h>
  78 #include <linux/rculist.h>
  79 #include <net/flow_keys.h>
  80 #include <net/switchdev.h>
  81 #include <net/bonding.h>
  82 #include <net/bond_3ad.h>
  83 #include <net/bond_alb.h>
  84
  85 #include "bonding_priv.h"
  86
  87 /*---------------------------- Module parameters ----------------------------*/
  88
  89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  90
  91 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  92 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  93 static int num_peer_notif = 1;
  94 static int miimon;
  95 static int updelay;
  96 static int downdelay;
  97 static int use_carrier  = 1;
  98 static char *mode;
  99 static char *primary;
 100 static char *primary_reselect;
 101 static char *lacp_rate;
 102 static int min_links;
 103 static char *ad_select;
 104 static char *xmit_hash_policy;
 105 static int arp_interval;
 106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 107 static char *arp_validate;
 108 static char *arp_all_targets;
 109 static char *fail_over_mac;
 110 static int all_slaves_active;
 111 static struct bond_params bonding_defaults;
 112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 113 static int packets_per_slave = 1;
 114 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
 115
 116 module_param(max_bonds, int, 0);
 117 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 118 module_param(tx_queues, int, 0);
 119 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 120 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 121 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 122                                "failover event (alias of num_unsol_na)");
 123 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 124 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 125                                "failover event (alias of num_grat_arp)");
 126 module_param(miimon, int, 0);
 127 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 128 module_param(updelay, int, 0);
 129 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 130 module_param(downdelay, int, 0);
 131 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 132                             "in milliseconds");
 133 module_param(use_carrier, int, 0);
 134 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 135                               "0 for off, 1 for on (default)");
 136 module_param(mode, charp, 0);
 137 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 138                        "1 for active-backup, 2 for balance-xor, "
 139                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 140                        "6 for balance-alb");
 141 module_param(primary, charp, 0);
 142 MODULE_PARM_DESC(primary, "Primary network device to use");
 143 module_param(primary_reselect, charp, 0);
 144 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 145                                    "once it comes up; "
 146                                    "0 for always (default), "
 147                                    "1 for only if speed of primary is "
 148                                    "better, "
 149                                    "2 for only on active slave "
 150                                    "failure");
 151 module_param(lacp_rate, charp, 0);
 152 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 153                             "0 for slow, 1 for fast");
 154 module_param(ad_select, charp, 0);
 155 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 156                             "0 for stable (default), 1 for bandwidth, "
 157                             "2 for count");
 158 module_param(min_links, int, 0);
 159 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 160
 161 module_param(xmit_hash_policy, charp, 0);
 162 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 163                                    "0 for layer 2 (default), 1 for layer 3+4, "
 164                                    "2 for layer 2+3, 3 for encap layer 2+3, "
 165                                    "4 for encap layer 3+4");
 166 module_param(arp_interval, int, 0);
 167 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 168 module_param_array(arp_ip_target, charp, NULL, 0);
 169 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 170 module_param(arp_validate, charp, 0);
 171 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 172                                "0 for none (default), 1 for active, "
 173                                "2 for backup, 3 for all");
 174 module_param(arp_all_targets, charp, 0);
 175 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 176 module_param(fail_over_mac, charp, 0);
 177 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 178                                 "the same MAC; 0 for none (default), "
 179                                 "1 for active, 2 for follow");
 180 module_param(all_slaves_active, int, 0);
 181 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
 182                                      "by setting active flag for all slaves; "
 183                                      "0 for never (default), 1 for always.");
 184 module_param(resend_igmp, int, 0);
 185 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 186                               "link failure");
 187 module_param(packets_per_slave, int, 0);
 188 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
 189                                     "mode; 0 for a random slave, 1 packet per "
 190                                     "slave (default), >1 packets per slave.");
 191 module_param(lp_interval, uint, 0);
 192 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
 193                               "the bonding driver sends learning packets to "
 194                               "each slaves peer switch. The default is 1.");
 195
 196 /*----------------------------- Global variables ----------------------------*/
 197
 198 #ifdef CONFIG_NET_POLL_CONTROLLER
 199 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 200 #endif
 201
 202 int bond_net_id __read_mostly;
 203
 204 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 205 static int arp_ip_count;
 206 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 207 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 208 static int lacp_fast;
 209
 210 /*-------------------------- Forward declarations ---------------------------*/
 211
 212 static int bond_init(struct net_device *bond_dev);
 213 static void bond_uninit(struct net_device *bond_dev);
 214 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
 215                                                 struct rtnl_link_stats64 *stats);
 216 static void bond_slave_arr_handler(struct work_struct *work);
 217
 218 /*---------------------------- General routines -----------------------------*/
 219
 220 const char *bond_mode_name(int mode)
 221 {
 222         static const char *names[] = {
 223                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 224                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 225                 [BOND_MODE_XOR] = "load balancing (xor)",
 226                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 227                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 228                 [BOND_MODE_TLB] = "transmit load balancing",
 229                 [BOND_MODE_ALB] = "adaptive load balancing",
 230         };
 231
 232         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 233                 return "unknown";
 234
 235         return names[mode];
 236 }
 237
 238 /*---------------------------------- VLAN -----------------------------------*/
 239
 240 /**
 241  * bond_dev_queue_xmit - Prepare skb for xmit.
 242  *
 243  * @bond: bond device that got this skb for tx.
 244  * @skb: hw accel VLAN tagged skb to transmit
 245  * @slave_dev: slave that is supposed to xmit this skbuff
 246  */
 247 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 248                         struct net_device *slave_dev)
 249 {
 250         skb->dev = slave_dev;
 251
 252         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 253                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 254         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 255
 256         if (unlikely(netpoll_tx_running(bond->dev)))
 257                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 258         else
 259                 dev_queue_xmit(skb);
 260 }
 261
 262 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 263  * We don't protect the slave list iteration with a lock because:
 264  * a. This operation is performed in IOCTL context,
 265  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 266  * c. Holding a lock with BH disabled while directly calling a base driver
 267  *    entry point is generally a BAD idea.
 268  *
 269  * The design of synchronization/protection for this operation in the 8021q
 270  * module is good for one or more VLAN devices over a single physical device
 271  * and cannot be extended for a teaming solution like bonding, so there is a
 272  * potential race condition here where a net device from the vlan group might
 273  * be referenced (either by a base driver or the 8021q code) while it is being
 274  * removed from the system. However, it turns out we're not making matters
 275  * worse, and if it works for regular VLAN usage it will work here too.
 276 */
 277
 278 /**
 279  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 280  * @bond_dev: bonding net device that got called
 281  * @vid: vlan id being added
 282  */
 283 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 284                                 __be16 proto, u16 vid)
 285 {
 286         struct bonding *bond = netdev_priv(bond_dev);
 287         struct slave *slave, *rollback_slave;
 288         struct list_head *iter;
 289         int res;
 290
 291         bond_for_each_slave(bond, slave, iter) {
 292                 res = vlan_vid_add(slave->dev, proto, vid);
 293                 if (res)
 294                         goto unwind;
 295         }
 296
 297         return 0;
 298
 299 unwind:
 300         /* unwind to the slave that failed */
 301         bond_for_each_slave(bond, rollback_slave, iter) {
 302                 if (rollback_slave == slave)
 303                         break;
 304
 305                 vlan_vid_del(rollback_slave->dev, proto, vid);
 306         }
 307
 308         return res;
 309 }
 310
 311 /**
 312  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 313  * @bond_dev: bonding net device that got called
 314  * @vid: vlan id being removed
 315  */
 316 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 317                                  __be16 proto, u16 vid)
 318 {
 319         struct bonding *bond = netdev_priv(bond_dev);
 320         struct list_head *iter;
 321         struct slave *slave;
 322
 323         bond_for_each_slave(bond, slave, iter)
 324                 vlan_vid_del(slave->dev, proto, vid);
 325
 326         if (bond_is_lb(bond))
 327                 bond_alb_clear_vlan(bond, vid);
 328
 329         return 0;
 330 }
 331
 332 /*------------------------------- Link status -------------------------------*/
 333
 334 /* Set the carrier state for the master according to the state of its
 335  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 336  * do special 802.3ad magic.
 337  *
 338  * Returns zero if carrier state does not change, nonzero if it does.
 339  */
 340 int bond_set_carrier(struct bonding *bond)
 341 {
 342         struct list_head *iter;
 343         struct slave *slave;
 344
 345         if (!bond_has_slaves(bond))
 346                 goto down;
 347
 348         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 349                 return bond_3ad_set_carrier(bond);
 350
 351         bond_for_each_slave(bond, slave, iter) {
 352                 if (slave->link == BOND_LINK_UP) {
 353                         if (!netif_carrier_ok(bond->dev)) {
 354                                 netif_carrier_on(bond->dev);
 355                                 return 1;
 356                         }
 357                         return 0;
 358                 }
 359         }
 360
 361 down:
 362         if (netif_carrier_ok(bond->dev)) {
 363                 netif_carrier_off(bond->dev);
 364                 return 1;
 365         }
 366         return 0;
 367 }
 368
 369 /* Get link speed and duplex from the slave's base driver
 370  * using ethtool. If for some reason the call fails or the
 371  * values are invalid, set speed and duplex to -1,
 372  * and return.
 373  */
 374 static void bond_update_speed_duplex(struct slave *slave)
 375 {
 376         struct net_device *slave_dev = slave->dev;
 377         struct ethtool_cmd ecmd;
 378         u32 slave_speed;
 379         int res;
 380
 381         slave->speed = SPEED_UNKNOWN;
 382         slave->duplex = DUPLEX_UNKNOWN;
 383
 384         res = __ethtool_get_settings(slave_dev, &ecmd);
 385         if (res < 0)
 386                 return;
 387
 388         slave_speed = ethtool_cmd_speed(&ecmd);
 389         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 390                 return;
 391
 392         switch (ecmd.duplex) {
 393         case DUPLEX_FULL:
 394         case DUPLEX_HALF:
 395                 break;
 396         default:
 397                 return;
 398         }
 399
 400         slave->speed = slave_speed;
 401         slave->duplex = ecmd.duplex;
 402
 403         return;
 404 }
 405
 406 const char *bond_slave_link_status(s8 link)
 407 {
 408         switch (link) {
 409         case BOND_LINK_UP:
 410                 return "up";
 411         case BOND_LINK_FAIL:
 412                 return "going down";
 413         case BOND_LINK_DOWN:
 414                 return "down";
 415         case BOND_LINK_BACK:
 416                 return "going back";
 417         default:
 418                 return "unknown";
 419         }
 420 }
 421
 422 /* if <dev> supports MII link status reporting, check its link status.
 423  *
 424  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 425  * depending upon the setting of the use_carrier parameter.
 426  *
 427  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 428  * can't tell and just pretend it is), or 0, meaning that the link is
 429  * down.
 430  *
 431  * If reporting is non-zero, instead of faking link up, return -1 if
 432  * both ETHTOOL and MII ioctls fail (meaning the device does not
 433  * support them).  If use_carrier is set, return whatever it says.
 434  * It'd be nice if there was a good way to tell if a driver supports
 435  * netif_carrier, but there really isn't.
 436  */
 437 static int bond_check_dev_link(struct bonding *bond,
 438                                struct net_device *slave_dev, int reporting)
 439 {
 440         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 441         int (*ioctl)(struct net_device *, struct ifreq *, int);
 442         struct ifreq ifr;
 443         struct mii_ioctl_data *mii;
 444
 445         if (!reporting && !netif_running(slave_dev))
 446                 return 0;
 447
 448         if (bond->params.use_carrier)
 449                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 450
 451         /* Try to get link status using Ethtool first. */
 452         if (slave_dev->ethtool_ops->get_link)
 453                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 454                         BMSR_LSTATUS : 0;
 455
 456         /* Ethtool can't be used, fallback to MII ioctls. */
 457         ioctl = slave_ops->ndo_do_ioctl;
 458         if (ioctl) {
 459                 /* TODO: set pointer to correct ioctl on a per team member
 460                  *       bases to make this more efficient. that is, once
 461                  *       we determine the correct ioctl, we will always
 462                  *       call it and not the others for that team
 463                  *       member.
 464                  */
 465
 466                 /* We cannot assume that SIOCGMIIPHY will also read a
 467                  * register; not all network drivers (e.g., e100)
 468                  * support that.
 469                  */
 470
 471                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 472                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 473                 mii = if_mii(&ifr);
 474                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 475                         mii->reg_num = MII_BMSR;
 476                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 477                                 return mii->val_out & BMSR_LSTATUS;
 478                 }
 479         }
 480
 481         /* If reporting, report that either there's no dev->do_ioctl,
 482          * or both SIOCGMIIREG and get_link failed (meaning that we
 483          * cannot report link status).  If not reporting, pretend
 484          * we're ok.
 485          */
 486         return reporting ? -1 : BMSR_LSTATUS;
 487 }
 488
 489 /*----------------------------- Multicast list ------------------------------*/
 490
 491 /* Push the promiscuity flag down to appropriate slaves */
 492 static int bond_set_promiscuity(struct bonding *bond, int inc)
 493 {
 494         struct list_head *iter;
 495         int err = 0;
 496
 497         if (bond_uses_primary(bond)) {
 498                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 499
 500                 if (curr_active)
 501                         err = dev_set_promiscuity(curr_active->dev, inc);
 502         } else {
 503                 struct slave *slave;
 504
 505                 bond_for_each_slave(bond, slave, iter) {
 506                         err = dev_set_promiscuity(slave->dev, inc);
 507                         if (err)
 508                                 return err;
 509                 }
 510         }
 511         return err;
 512 }
 513
 514 /* Push the allmulti flag down to all slaves */
 515 static int bond_set_allmulti(struct bonding *bond, int inc)
 516 {
 517         struct list_head *iter;
 518         int err = 0;
 519
 520         if (bond_uses_primary(bond)) {
 521                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 522
 523                 if (curr_active)
 524                         err = dev_set_allmulti(curr_active->dev, inc);
 525         } else {
 526                 struct slave *slave;
 527
 528                 bond_for_each_slave(bond, slave, iter) {
 529                         err = dev_set_allmulti(slave->dev, inc);
 530                         if (err)
 531                                 return err;
 532                 }
 533         }
 534         return err;
 535 }
 536
 537 /* Retrieve the list of registered multicast addresses for the bonding
 538  * device and retransmit an IGMP JOIN request to the current active
 539  * slave.
 540  */
 541 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 542 {
 543         struct bonding *bond = container_of(work, struct bonding,
 544                                             mcast_work.work);
 545
 546         if (!rtnl_trylock()) {
 547                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 548                 return;
 549         }
 550         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 551
 552         if (bond->igmp_retrans > 1) {
 553                 bond->igmp_retrans--;
 554                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 555         }
 556         rtnl_unlock();
 557 }
 558
 559 /* Flush bond's hardware addresses from slave */
 560 static void bond_hw_addr_flush(struct net_device *bond_dev,
 561                                struct net_device *slave_dev)
 562 {
 563         struct bonding *bond = netdev_priv(bond_dev);
 564
 565         dev_uc_unsync(slave_dev, bond_dev);
 566         dev_mc_unsync(slave_dev, bond_dev);
 567
 568         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 569                 /* del lacpdu mc addr from mc list */
 570                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 571
 572                 dev_mc_del(slave_dev, lacpdu_multicast);
 573         }
 574 }
 575
 576 /*--------------------------- Active slave change ---------------------------*/
 577
 578 /* Update the hardware address list and promisc/allmulti for the new and
 579  * old active slaves (if any).  Modes that are not using primary keep all
 580  * slaves up date at all times; only the modes that use primary need to call
 581  * this function to swap these settings during a failover.
 582  */
 583 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 584                               struct slave *old_active)
 585 {
 586         if (old_active) {
 587                 if (bond->dev->flags & IFF_PROMISC)
 588                         dev_set_promiscuity(old_active->dev, -1);
 589
 590                 if (bond->dev->flags & IFF_ALLMULTI)
 591                         dev_set_allmulti(old_active->dev, -1);
 592
 593                 bond_hw_addr_flush(bond->dev, old_active->dev);
 594         }
 595
 596         if (new_active) {
 597                 /* FIXME: Signal errors upstream. */
 598                 if (bond->dev->flags & IFF_PROMISC)
 599                         dev_set_promiscuity(new_active->dev, 1);
 600
 601                 if (bond->dev->flags & IFF_ALLMULTI)
 602                         dev_set_allmulti(new_active->dev, 1);
 603
 604                 netif_addr_lock_bh(bond->dev);
 605                 dev_uc_sync(new_active->dev, bond->dev);
 606                 dev_mc_sync(new_active->dev, bond->dev);
 607                 netif_addr_unlock_bh(bond->dev);
 608         }
 609 }
 610
 611 /**
 612  * bond_set_dev_addr - clone slave's address to bond
 613  * @bond_dev: bond net device
 614  * @slave_dev: slave net device
 615  *
 616  * Should be called with RTNL held.
 617  */
 618 static void bond_set_dev_addr(struct net_device *bond_dev,
 619                               struct net_device *slave_dev)
 620 {
 621         netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 622                    bond_dev, slave_dev, slave_dev->addr_len);
 623         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 624         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 625         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 626 }
 627
 628 /* bond_do_fail_over_mac
 629  *
 630  * Perform special MAC address swapping for fail_over_mac settings
 631  *
 632  * Called with RTNL
 633  */
 634 static void bond_do_fail_over_mac(struct bonding *bond,
 635                                   struct slave *new_active,
 636                                   struct slave *old_active)
 637 {
 638         u8 tmp_mac[ETH_ALEN];
 639         struct sockaddr saddr;
 640         int rv;
 641
 642         switch (bond->params.fail_over_mac) {
 643         case BOND_FOM_ACTIVE:
 644                 if (new_active)
 645                         bond_set_dev_addr(bond->dev, new_active->dev);
 646                 break;
 647         case BOND_FOM_FOLLOW:
 648                 /* if new_active && old_active, swap them
 649                  * if just old_active, do nothing (going to no active slave)
 650                  * if just new_active, set new_active to bond's MAC
 651                  */
 652                 if (!new_active)
 653                         return;
 654
 655                 if (old_active) {
 656                         ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
 657                         ether_addr_copy(saddr.sa_data,
 658                                         old_active->dev->dev_addr);
 659                         saddr.sa_family = new_active->dev->type;
 660                 } else {
 661                         ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
 662                         saddr.sa_family = bond->dev->type;
 663                 }
 664
 665                 rv = dev_set_mac_address(new_active->dev, &saddr);
 666                 if (rv) {
 667                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 668                                    -rv, new_active->dev->name);
 669                         goto out;
 670                 }
 671
 672                 if (!old_active)
 673                         goto out;
 674
 675                 ether_addr_copy(saddr.sa_data, tmp_mac);
 676                 saddr.sa_family = old_active->dev->type;
 677
 678                 rv = dev_set_mac_address(old_active->dev, &saddr);
 679                 if (rv)
 680                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 681                                    -rv, new_active->dev->name);
 682 out:
 683                 break;
 684         default:
 685                 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
 686                            bond->params.fail_over_mac);
 687                 break;
 688         }
 689
 690 }
 691
 692 static bool bond_should_change_active(struct bonding *bond)
 693 {
 694         struct slave *prim = rtnl_dereference(bond->primary_slave);
 695         struct slave *curr = rtnl_dereference(bond->curr_active_slave);
 696
 697         if (!prim || !curr || curr->link != BOND_LINK_UP)
 698                 return true;
 699         if (bond->force_primary) {
 700                 bond->force_primary = false;
 701                 return true;
 702         }
 703         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 704             (prim->speed < curr->speed ||
 705              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 706                 return false;
 707         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 708                 return false;
 709         return true;
 710 }
 711
 712 /**
 713  * find_best_interface - select the best available slave to be the active one
 714  * @bond: our bonding struct
 715  */
 716 static struct slave *bond_find_best_slave(struct bonding *bond)
 717 {
 718         struct slave *slave, *bestslave = NULL, *primary;
 719         struct list_head *iter;
 720         int mintime = bond->params.updelay;
 721
 722         primary = rtnl_dereference(bond->primary_slave);
 723         if (primary && primary->link == BOND_LINK_UP &&
 724             bond_should_change_active(bond))
 725                 return primary;
 726
 727         bond_for_each_slave(bond, slave, iter) {
 728                 if (slave->link == BOND_LINK_UP)
 729                         return slave;
 730                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
 731                     slave->delay < mintime) {
 732                         mintime = slave->delay;
 733                         bestslave = slave;
 734                 }
 735         }
 736
 737         return bestslave;
 738 }
 739
 740 static bool bond_should_notify_peers(struct bonding *bond)
 741 {
 742         struct slave *slave;
 743
 744         rcu_read_lock();
 745         slave = rcu_dereference(bond->curr_active_slave);
 746         rcu_read_unlock();
 747
 748         netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
 749                    slave ? slave->dev->name : "NULL");
 750
 751         if (!slave || !bond->send_peer_notif ||
 752             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 753                 return false;
 754
 755         return true;
 756 }
 757
 758 /**
 759  * change_active_interface - change the active slave into the specified one
 760  * @bond: our bonding struct
 761  * @new: the new slave to make the active one
 762  *
 763  * Set the new slave to the bond's settings and unset them on the old
 764  * curr_active_slave.
 765  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 766  *
 767  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 768  * because it is apparently the best available slave we have, even though its
 769  * updelay hasn't timed out yet.
 770  *
 771  * Caller must hold RTNL.
 772  */
 773 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 774 {
 775         struct slave *old_active;
 776
 777         ASSERT_RTNL();
 778
 779         old_active = rtnl_dereference(bond->curr_active_slave);
 780
 781         if (old_active == new_active)
 782                 return;
 783
 784         if (new_active) {
 785                 new_active->last_link_up = jiffies;
 786
 787                 if (new_active->link == BOND_LINK_BACK) {
 788                         if (bond_uses_primary(bond)) {
 789                                 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
 790                                             new_active->dev->name,
 791                                             (bond->params.updelay - new_active->delay) * bond->params.miimon);
 792                         }
 793
 794                         new_active->delay = 0;
 795                         bond_set_slave_link_state(new_active, BOND_LINK_UP);
 796
 797                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 798                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 799
 800                         if (bond_is_lb(bond))
 801                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 802                 } else {
 803                         if (bond_uses_primary(bond)) {
 804                                 netdev_info(bond->dev, "making interface %s the new active one\n",
 805                                             new_active->dev->name);
 806                         }
 807                 }
 808         }
 809
 810         if (bond_uses_primary(bond))
 811                 bond_hw_addr_swap(bond, new_active, old_active);
 812
 813         if (bond_is_lb(bond)) {
 814                 bond_alb_handle_active_change(bond, new_active);
 815                 if (old_active)
 816                         bond_set_slave_inactive_flags(old_active,
 817                                                       BOND_SLAVE_NOTIFY_NOW);
 818                 if (new_active)
 819                         bond_set_slave_active_flags(new_active,
 820                                                     BOND_SLAVE_NOTIFY_NOW);
 821         } else {
 822                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 823         }
 824
 825         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
 826                 if (old_active)
 827                         bond_set_slave_inactive_flags(old_active,
 828                                                       BOND_SLAVE_NOTIFY_NOW);
 829
 830                 if (new_active) {
 831                         bool should_notify_peers = false;
 832
 833                         bond_set_slave_active_flags(new_active,
 834                                                     BOND_SLAVE_NOTIFY_NOW);
 835
 836                         if (bond->params.fail_over_mac)
 837                                 bond_do_fail_over_mac(bond, new_active,
 838                                                       old_active);
 839
 840                         if (netif_running(bond->dev)) {
 841                                 bond->send_peer_notif =
 842                                         bond->params.num_peer_notif;
 843                                 should_notify_peers =
 844                                         bond_should_notify_peers(bond);
 845                         }
 846
 847                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 848                         if (should_notify_peers)
 849                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 850                                                          bond->dev);
 851                 }
 852         }
 853
 854         /* resend IGMP joins since active slave has changed or
 855          * all were sent on curr_active_slave.
 856          * resend only if bond is brought up with the affected
 857          * bonding modes and the retransmission is enabled
 858          */
 859         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 860             ((bond_uses_primary(bond) && new_active) ||
 861              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
 862                 bond->igmp_retrans = bond->params.resend_igmp;
 863                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 864         }
 865 }
 866
 867 /**
 868  * bond_select_active_slave - select a new active slave, if needed
 869  * @bond: our bonding struct
 870  *
 871  * This functions should be called when one of the following occurs:
 872  * - The old curr_active_slave has been released or lost its link.
 873  * - The primary_slave has got its link back.
 874  * - A slave has got its link back and there's no old curr_active_slave.
 875  *
 876  * Caller must hold RTNL.
 877  */
 878 void bond_select_active_slave(struct bonding *bond)
 879 {
 880         struct slave *best_slave;
 881         int rv;
 882
 883         ASSERT_RTNL();
 884
 885         best_slave = bond_find_best_slave(bond);
 886         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
 887                 bond_change_active_slave(bond, best_slave);
 888                 rv = bond_set_carrier(bond);
 889                 if (!rv)
 890                         return;
 891
 892                 if (netif_carrier_ok(bond->dev)) {
 893                         netdev_info(bond->dev, "first active interface up!\n");
 894                 } else {
 895                         netdev_info(bond->dev, "now running without any active interface!\n");
 896                 }
 897         }
 898 }
 899
 900 #ifdef CONFIG_NET_POLL_CONTROLLER
 901 static inline int slave_enable_netpoll(struct slave *slave)
 902 {
 903         struct netpoll *np;
 904         int err = 0;
 905
 906         np = kzalloc(sizeof(*np), GFP_KERNEL);
 907         err = -ENOMEM;
 908         if (!np)
 909                 goto out;
 910
 911         err = __netpoll_setup(np, slave->dev);
 912         if (err) {
 913                 kfree(np);
 914                 goto out;
 915         }
 916         slave->np = np;
 917 out:
 918         return err;
 919 }
 920 static inline void slave_disable_netpoll(struct slave *slave)
 921 {
 922         struct netpoll *np = slave->np;
 923
 924         if (!np)
 925                 return;
 926
 927         slave->np = NULL;
 928         __netpoll_free_async(np);
 929 }
 930
 931 static void bond_poll_controller(struct net_device *bond_dev)
 932 {
 933         struct bonding *bond = netdev_priv(bond_dev);
 934         struct slave *slave = NULL;
 935         struct list_head *iter;
 936         struct ad_info ad_info;
 937         struct netpoll_info *ni;
 938         const struct net_device_ops *ops;
 939
 940         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 941                 if (bond_3ad_get_active_agg_info(bond, &ad_info))
 942                         return;
 943
 944         rcu_read_lock_bh();
 945         bond_for_each_slave_rcu(bond, slave, iter) {
 946                 ops = slave->dev->netdev_ops;
 947                 if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller)
 948                         continue;
 949
 950                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 951                         struct aggregator *agg =
 952                             SLAVE_AD_INFO(slave)->port.aggregator;
 953
 954                         if (agg &&
 955                             agg->aggregator_identifier != ad_info.aggregator_id)
 956                                 continue;
 957                 }
 958
 959                 ni = rcu_dereference_bh(slave->dev->npinfo);
 960                 if (down_trylock(&ni->dev_lock))
 961                         continue;
 962                 ops->ndo_poll_controller(slave->dev);
 963                 up(&ni->dev_lock);
 964         }
 965         rcu_read_unlock_bh();
 966 }
 967
 968 static void bond_netpoll_cleanup(struct net_device *bond_dev)
 969 {
 970         struct bonding *bond = netdev_priv(bond_dev);
 971         struct list_head *iter;
 972         struct slave *slave;
 973
 974         bond_for_each_slave(bond, slave, iter)
 975                 if (bond_slave_is_up(slave))
 976                         slave_disable_netpoll(slave);
 977 }
 978
 979 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
 980 {
 981         struct bonding *bond = netdev_priv(dev);
 982         struct list_head *iter;
 983         struct slave *slave;
 984         int err = 0;
 985
 986         bond_for_each_slave(bond, slave, iter) {
 987                 err = slave_enable_netpoll(slave);
 988                 if (err) {
 989                         bond_netpoll_cleanup(dev);
 990                         break;
 991                 }
 992         }
 993         return err;
 994 }
 995 #else
 996 static inline int slave_enable_netpoll(struct slave *slave)
 997 {
 998         return 0;
 999 }
1000 static inline void slave_disable_netpoll(struct slave *slave)
1001 {
1002 }
1003 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1004 {
1005 }
1006 #endif
1007
1008 /*---------------------------------- IOCTL ----------------------------------*/
1009
1010 static netdev_features_t bond_fix_features(struct net_device *dev,
1011                                            netdev_features_t features)
1012 {
1013         struct bonding *bond = netdev_priv(dev);
1014         struct list_head *iter;
1015         netdev_features_t mask;
1016         struct slave *slave;
1017
1018         /* If any slave has the offload feature flag set,
1019          * set the offload flag on the bond.
1020          */
1021         mask = features | NETIF_F_HW_SWITCH_OFFLOAD;
1022
1023         features &= ~NETIF_F_ONE_FOR_ALL;
1024         features |= NETIF_F_ALL_FOR_ALL;
1025
1026         bond_for_each_slave(bond, slave, iter) {
1027                 features = netdev_increment_features(features,
1028                                                      slave->dev->features,
1029                                                      mask);
1030         }
1031         features = netdev_add_tso_features(features, mask);
1032
1033         return features;
1034 }
1035
1036 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1037                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1038                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1039
1040 #define BOND_ENC_FEATURES       (NETIF_F_ALL_CSUM | NETIF_F_SG | NETIF_F_RXCSUM |\
1041                                  NETIF_F_TSO)
1042
1043 static void bond_compute_features(struct bonding *bond)
1044 {
1045         unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1046                                         IFF_XMIT_DST_RELEASE_PERM;
1047         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1048         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1049         struct net_device *bond_dev = bond->dev;
1050         struct list_head *iter;
1051         struct slave *slave;
1052         unsigned short max_hard_header_len = ETH_HLEN;
1053         unsigned int gso_max_size = GSO_MAX_SIZE;
1054         u16 gso_max_segs = GSO_MAX_SEGS;
1055
1056         if (!bond_has_slaves(bond))
1057                 goto done;
1058         vlan_features &= NETIF_F_ALL_FOR_ALL;
1059
1060         bond_for_each_slave(bond, slave, iter) {
1061                 vlan_features = netdev_increment_features(vlan_features,
1062                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1063
1064                 enc_features = netdev_increment_features(enc_features,
1065                                                          slave->dev->hw_enc_features,
1066                                                          BOND_ENC_FEATURES);
1067                 dst_release_flag &= slave->dev->priv_flags;
1068                 if (slave->dev->hard_header_len > max_hard_header_len)
1069                         max_hard_header_len = slave->dev->hard_header_len;
1070
1071                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1072                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1073         }
1074
1075 done:
1076         bond_dev->vlan_features = vlan_features;
1077         bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
1078         bond_dev->hard_header_len = max_hard_header_len;
1079         bond_dev->gso_max_segs = gso_max_segs;
1080         netif_set_gso_max_size(bond_dev, gso_max_size);
1081
1082         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1083         if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1084             dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1085                 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1086
1087         netdev_change_features(bond_dev);
1088 }
1089
1090 static void bond_setup_by_slave(struct net_device *bond_dev,
1091                                 struct net_device *slave_dev)
1092 {
1093         bond_dev->header_ops        = slave_dev->header_ops;
1094
1095         bond_dev->type              = slave_dev->type;
1096         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1097         bond_dev->addr_len          = slave_dev->addr_len;
1098
1099         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1100                 slave_dev->addr_len);
1101 }
1102
1103 /* On bonding slaves other than the currently active slave, suppress
1104  * duplicates except for alb non-mcast/bcast.
1105  */
1106 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1107                                             struct slave *slave,
1108                                             struct bonding *bond)
1109 {
1110         if (bond_is_slave_inactive(slave)) {
1111                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1112                     skb->pkt_type != PACKET_BROADCAST &&
1113                     skb->pkt_type != PACKET_MULTICAST)
1114                         return false;
1115                 return true;
1116         }
1117         return false;
1118 }
1119
1120 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1121 {
1122         struct sk_buff *skb = *pskb;
1123         struct slave *slave;
1124         struct bonding *bond;
1125         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1126                           struct slave *);
1127         int ret = RX_HANDLER_ANOTHER;
1128
1129         skb = skb_share_check(skb, GFP_ATOMIC);
1130         if (unlikely(!skb))
1131                 return RX_HANDLER_CONSUMED;
1132
1133         *pskb = skb;
1134
1135         slave = bond_slave_get_rcu(skb->dev);
1136         bond = slave->bond;
1137
1138         recv_probe = ACCESS_ONCE(bond->recv_probe);
1139         if (recv_probe) {
1140                 ret = recv_probe(skb, bond, slave);
1141                 if (ret == RX_HANDLER_CONSUMED) {
1142                         consume_skb(skb);
1143                         return ret;
1144                 }
1145         }
1146
1147         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1148                 return RX_HANDLER_EXACT;
1149         }
1150
1151         skb->dev = bond->dev;
1152
1153         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1154             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1155             skb->pkt_type == PACKET_HOST) {
1156
1157                 if (unlikely(skb_cow_head(skb,
1158                                           skb->data - skb_mac_header(skb)))) {
1159                         kfree_skb(skb);
1160                         return RX_HANDLER_CONSUMED;
1161                 }
1162                 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1163         }
1164
1165         return ret;
1166 }
1167
1168 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1169                                       struct net_device *slave_dev,
1170                                       struct slave *slave)
1171 {
1172         int err;
1173
1174         err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1175         if (err)
1176                 return err;
1177         slave_dev->flags |= IFF_SLAVE;
1178         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1179         return 0;
1180 }
1181
1182 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1183                                   struct net_device *slave_dev)
1184 {
1185         netdev_upper_dev_unlink(slave_dev, bond_dev);
1186         slave_dev->flags &= ~IFF_SLAVE;
1187         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1188 }
1189
1190 static struct slave *bond_alloc_slave(struct bonding *bond)
1191 {
1192         struct slave *slave = NULL;
1193
1194         slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1195         if (!slave)
1196                 return NULL;
1197
1198         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1199                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1200                                                GFP_KERNEL);
1201                 if (!SLAVE_AD_INFO(slave)) {
1202                         kfree(slave);
1203                         return NULL;
1204                 }
1205         }
1206         return slave;
1207 }
1208
1209 static void bond_free_slave(struct slave *slave)
1210 {
1211         struct bonding *bond = bond_get_bond_by_slave(slave);
1212
1213         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1214                 kfree(SLAVE_AD_INFO(slave));
1215
1216         kfree(slave);
1217 }
1218
1219 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1220 {
1221         info->bond_mode = BOND_MODE(bond);
1222         info->miimon = bond->params.miimon;
1223         info->num_slaves = bond->slave_cnt;
1224 }
1225
1226 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1227 {
1228         strcpy(info->slave_name, slave->dev->name);
1229         info->link = slave->link;
1230         info->state = bond_slave_state(slave);
1231         info->link_failure_count = slave->link_failure_count;
1232 }
1233
1234 static void bond_netdev_notify(struct net_device *dev,
1235                                struct netdev_bonding_info *info)
1236 {
1237         rtnl_lock();
1238         netdev_bonding_info_change(dev, info);
1239         rtnl_unlock();
1240 }
1241
1242 static void bond_netdev_notify_work(struct work_struct *_work)
1243 {
1244         struct netdev_notify_work *w =
1245                 container_of(_work, struct netdev_notify_work, work.work);
1246
1247         bond_netdev_notify(w->dev, &w->bonding_info);
1248         dev_put(w->dev);
1249         kfree(w);
1250 }
1251
1252 void bond_queue_slave_event(struct slave *slave)
1253 {
1254         struct bonding *bond = slave->bond;
1255         struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1256
1257         if (!nnw)
1258                 return;
1259
1260         dev_hold(slave->dev);
1261         nnw->dev = slave->dev;
1262         bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1263         bond_fill_ifbond(bond, &nnw->bonding_info.master);
1264         INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1265
1266         queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1267 }
1268
1269 /* enslave device <slave> to bond device <master> */
1270 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1271 {
1272         struct bonding *bond = netdev_priv(bond_dev);
1273         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1274         struct slave *new_slave = NULL, *prev_slave;
1275         struct sockaddr addr;
1276         int link_reporting;
1277         int res = 0, i;
1278
1279         if (!bond->params.use_carrier &&
1280             slave_dev->ethtool_ops->get_link == NULL &&
1281             slave_ops->ndo_do_ioctl == NULL) {
1282                 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1283                             slave_dev->name);
1284         }
1285
1286         /* already enslaved */
1287         if (slave_dev->flags & IFF_SLAVE) {
1288                 netdev_dbg(bond_dev, "Error: Device was already enslaved\n");
1289                 return -EBUSY;
1290         }
1291
1292         if (bond_dev == slave_dev) {
1293                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1294                 return -EPERM;
1295         }
1296
1297         /* vlan challenged mutual exclusion */
1298         /* no need to lock since we're protected by rtnl_lock */
1299         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1300                 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1301                            slave_dev->name);
1302                 if (vlan_uses_dev(bond_dev)) {
1303                         netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1304                                    slave_dev->name, bond_dev->name);
1305                         return -EPERM;
1306                 } else {
1307                         netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1308                                     slave_dev->name, slave_dev->name,
1309                                     bond_dev->name);
1310                 }
1311         } else {
1312                 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1313                            slave_dev->name);
1314         }
1315
1316         /* Old ifenslave binaries are no longer supported.  These can
1317          * be identified with moderate accuracy by the state of the slave:
1318          * the current ifenslave will set the interface down prior to
1319          * enslaving it; the old ifenslave will not.
1320          */
1321         if ((slave_dev->flags & IFF_UP)) {
1322                 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1323                            slave_dev->name);
1324                 res = -EPERM;
1325                 goto err_undo_flags;
1326         }
1327
1328         /* set bonding device ether type by slave - bonding netdevices are
1329          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1330          * there is a need to override some of the type dependent attribs/funcs.
1331          *
1332          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1333          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1334          */
1335         if (!bond_has_slaves(bond)) {
1336                 if (bond_dev->type != slave_dev->type) {
1337                         netdev_dbg(bond_dev, "change device type from %d to %d\n",
1338                                    bond_dev->type, slave_dev->type);
1339
1340                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1341                                                        bond_dev);
1342                         res = notifier_to_errno(res);
1343                         if (res) {
1344                                 netdev_err(bond_dev, "refused to change device type\n");
1345                                 res = -EBUSY;
1346                                 goto err_undo_flags;
1347                         }
1348
1349                         /* Flush unicast and multicast addresses */
1350                         dev_uc_flush(bond_dev);
1351                         dev_mc_flush(bond_dev);
1352
1353                         if (slave_dev->type != ARPHRD_ETHER)
1354                                 bond_setup_by_slave(bond_dev, slave_dev);
1355                         else {
1356                                 ether_setup(bond_dev);
1357                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1358                         }
1359
1360                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1361                                                  bond_dev);
1362                 }
1363         } else if (bond_dev->type != slave_dev->type) {
1364                 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1365                            slave_dev->name, slave_dev->type, bond_dev->type);
1366                 res = -EINVAL;
1367                 goto err_undo_flags;
1368         }
1369
1370         if (slave_ops->ndo_set_mac_address == NULL) {
1371                 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1372                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1373                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1374                         if (!bond_has_slaves(bond)) {
1375                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1376                                 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1377                         } else {
1378                                 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");
1379                                 res = -EOPNOTSUPP;
1380                                 goto err_undo_flags;
1381                         }
1382                 }
1383         }
1384
1385         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1386
1387         /* If this is the first slave, then we need to set the master's hardware
1388          * address to be the same as the slave's.
1389          */
1390         if (!bond_has_slaves(bond) &&
1391             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1392                 bond_set_dev_addr(bond->dev, slave_dev);
1393
1394         new_slave = bond_alloc_slave(bond);
1395         if (!new_slave) {
1396                 res = -ENOMEM;
1397                 goto err_undo_flags;
1398         }
1399
1400         new_slave->bond = bond;
1401         new_slave->dev = slave_dev;
1402         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1403          * is set via sysfs or module option if desired.
1404          */
1405         new_slave->queue_id = 0;
1406
1407         /* Save slave's original mtu and then set it to match the bond */
1408         new_slave->original_mtu = slave_dev->mtu;
1409         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1410         if (res) {
1411                 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1412                 goto err_free;
1413         }
1414
1415         /* Save slave's original ("permanent") mac address for modes
1416          * that need it, and for restoring it upon release, and then
1417          * set it to the master's address
1418          */
1419         ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1420
1421         if (!bond->params.fail_over_mac ||
1422             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1423                 /* Set slave to master's mac address.  The application already
1424                  * set the master's mac address to that of the first slave
1425                  */
1426                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1427                 addr.sa_family = slave_dev->type;
1428                 res = dev_set_mac_address(slave_dev, &addr);
1429                 if (res) {
1430                         netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1431                         goto err_restore_mtu;
1432                 }
1433         }
1434
1435         /* open the slave since the application closed it */
1436         res = dev_open(slave_dev);
1437         if (res) {
1438                 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1439                 goto err_restore_mac;
1440         }
1441
1442         slave_dev->priv_flags |= IFF_BONDING;
1443         /* initialize slave stats */
1444         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1445
1446         if (bond_is_lb(bond)) {
1447                 /* bond_alb_init_slave() must be called before all other stages since
1448                  * it might fail and we do not want to have to undo everything
1449                  */
1450                 res = bond_alb_init_slave(bond, new_slave);
1451                 if (res)
1452                         goto err_close;
1453         }
1454
1455         /* If the mode uses primary, then the following is handled by
1456          * bond_change_active_slave().
1457          */
1458         if (!bond_uses_primary(bond)) {
1459                 /* set promiscuity level to new slave */
1460                 if (bond_dev->flags & IFF_PROMISC) {
1461                         res = dev_set_promiscuity(slave_dev, 1);
1462                         if (res)
1463                                 goto err_close;
1464                 }
1465
1466                 /* set allmulti level to new slave */
1467                 if (bond_dev->flags & IFF_ALLMULTI) {
1468                         res = dev_set_allmulti(slave_dev, 1);
1469                         if (res)
1470                                 goto err_close;
1471                 }
1472
1473                 netif_addr_lock_bh(bond_dev);
1474
1475                 dev_mc_sync_multiple(slave_dev, bond_dev);
1476                 dev_uc_sync_multiple(slave_dev, bond_dev);
1477
1478                 netif_addr_unlock_bh(bond_dev);
1479         }
1480
1481         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1482                 /* add lacpdu mc addr to mc list */
1483                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1484
1485                 dev_mc_add(slave_dev, lacpdu_multicast);
1486         }
1487
1488         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1489         if (res) {
1490                 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1491                            slave_dev->name);
1492                 goto err_close;
1493         }
1494
1495         prev_slave = bond_last_slave(bond);
1496
1497         new_slave->delay = 0;
1498         new_slave->link_failure_count = 0;
1499
1500         bond_update_speed_duplex(new_slave);
1501
1502         new_slave->last_rx = jiffies -
1503                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1504         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1505                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1506
1507         if (bond->params.miimon && !bond->params.use_carrier) {
1508                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1509
1510                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1511                         /* miimon is set but a bonded network driver
1512                          * does not support ETHTOOL/MII and
1513                          * arp_interval is not set.  Note: if
1514                          * use_carrier is enabled, we will never go
1515                          * here (because netif_carrier is always
1516                          * supported); thus, we don't need to change
1517                          * the messages for netif_carrier.
1518                          */
1519                         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",
1520                                     slave_dev->name);
1521                 } else if (link_reporting == -1) {
1522                         /* unable get link status using mii/ethtool */
1523                         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",
1524                                     slave_dev->name);
1525                 }
1526         }
1527
1528         /* check for initial state */
1529         if (bond->params.miimon) {
1530                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1531                         if (bond->params.updelay) {
1532                                 bond_set_slave_link_state(new_slave,
1533                                                           BOND_LINK_BACK);
1534                                 new_slave->delay = bond->params.updelay;
1535                         } else {
1536                                 bond_set_slave_link_state(new_slave,
1537                                                           BOND_LINK_UP);
1538                         }
1539                 } else {
1540                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN);
1541                 }
1542         } else if (bond->params.arp_interval) {
1543                 bond_set_slave_link_state(new_slave,
1544                                           (netif_carrier_ok(slave_dev) ?
1545                                           BOND_LINK_UP : BOND_LINK_DOWN));
1546         } else {
1547                 bond_set_slave_link_state(new_slave, BOND_LINK_UP);
1548         }
1549
1550         if (new_slave->link != BOND_LINK_DOWN)
1551                 new_slave->last_link_up = jiffies;
1552         netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1553                    new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1554                    (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1555
1556         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1557                 /* if there is a primary slave, remember it */
1558                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1559                         rcu_assign_pointer(bond->primary_slave, new_slave);
1560                         bond->force_primary = true;
1561                 }
1562         }
1563
1564         switch (BOND_MODE(bond)) {
1565         case BOND_MODE_ACTIVEBACKUP:
1566                 bond_set_slave_inactive_flags(new_slave,
1567                                               BOND_SLAVE_NOTIFY_NOW);
1568                 break;
1569         case BOND_MODE_8023AD:
1570                 /* in 802.3ad mode, the internal mechanism
1571                  * will activate the slaves in the selected
1572                  * aggregator
1573                  */
1574                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1575                 /* if this is the first slave */
1576                 if (!prev_slave) {
1577                         SLAVE_AD_INFO(new_slave)->id = 1;
1578                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1579                          * can be called only after the mac address of the bond is set
1580                          */
1581                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1582                 } else {
1583                         SLAVE_AD_INFO(new_slave)->id =
1584                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1585                 }
1586
1587                 bond_3ad_bind_slave(new_slave);
1588                 break;
1589         case BOND_MODE_TLB:
1590         case BOND_MODE_ALB:
1591                 bond_set_active_slave(new_slave);
1592                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1593                 break;
1594         default:
1595                 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1596
1597                 /* always active in trunk mode */
1598                 bond_set_active_slave(new_slave);
1599
1600                 /* In trunking mode there is little meaning to curr_active_slave
1601                  * anyway (it holds no special properties of the bond device),
1602                  * so we can change it without calling change_active_interface()
1603                  */
1604                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1605                     new_slave->link == BOND_LINK_UP)
1606                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1607
1608                 break;
1609         } /* switch(bond_mode) */
1610
1611 #ifdef CONFIG_NET_POLL_CONTROLLER
1612         slave_dev->npinfo = bond->dev->npinfo;
1613         if (slave_dev->npinfo) {
1614                 if (slave_enable_netpoll(new_slave)) {
1615                         netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1616                         res = -EBUSY;
1617                         goto err_detach;
1618                 }
1619         }
1620 #endif
1621
1622         if (!(bond_dev->features & NETIF_F_LRO))
1623                 dev_disable_lro(slave_dev);
1624
1625         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1626                                          new_slave);
1627         if (res) {
1628                 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1629                 goto err_detach;
1630         }
1631
1632         res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1633         if (res) {
1634                 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1635                 goto err_unregister;
1636         }
1637
1638         res = bond_sysfs_slave_add(new_slave);
1639         if (res) {
1640                 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1641                 goto err_upper_unlink;
1642         }
1643
1644         bond->slave_cnt++;
1645         bond_compute_features(bond);
1646         bond_set_carrier(bond);
1647
1648         if (bond_uses_primary(bond)) {
1649                 block_netpoll_tx();
1650                 bond_select_active_slave(bond);
1651                 unblock_netpoll_tx();
1652         }
1653
1654         if (bond_mode_uses_xmit_hash(bond))
1655                 bond_update_slave_arr(bond, NULL);
1656
1657         netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1658                     slave_dev->name,
1659                     bond_is_active_slave(new_slave) ? "an active" : "a backup",
1660                     new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1661
1662         /* enslave is successful */
1663         bond_queue_slave_event(new_slave);
1664         return 0;
1665
1666 /* Undo stages on error */
1667 err_upper_unlink:
1668         bond_upper_dev_unlink(bond_dev, slave_dev);
1669
1670 err_unregister:
1671         netdev_rx_handler_unregister(slave_dev);
1672
1673 err_detach:
1674         if (!bond_uses_primary(bond))
1675                 bond_hw_addr_flush(bond_dev, slave_dev);
1676
1677         vlan_vids_del_by_dev(slave_dev, bond_dev);
1678         if (rcu_access_pointer(bond->primary_slave) == new_slave)
1679                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1680         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1681                 block_netpoll_tx();
1682                 bond_change_active_slave(bond, NULL);
1683                 bond_select_active_slave(bond);
1684                 unblock_netpoll_tx();
1685         }
1686         /* either primary_slave or curr_active_slave might've changed */
1687         synchronize_rcu();
1688         slave_disable_netpoll(new_slave);
1689
1690 err_close:
1691         slave_dev->priv_flags &= ~IFF_BONDING;
1692         dev_close(slave_dev);
1693
1694 err_restore_mac:
1695         if (!bond->params.fail_over_mac ||
1696             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1697                 /* XXX TODO - fom follow mode needs to change master's
1698                  * MAC if this slave's MAC is in use by the bond, or at
1699                  * least print a warning.
1700                  */
1701                 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1702                 addr.sa_family = slave_dev->type;
1703                 dev_set_mac_address(slave_dev, &addr);
1704         }
1705
1706 err_restore_mtu:
1707         dev_set_mtu(slave_dev, new_slave->original_mtu);
1708
1709 err_free:
1710         bond_free_slave(new_slave);
1711
1712 err_undo_flags:
1713         /* Enslave of first slave has failed and we need to fix master's mac */
1714         if (!bond_has_slaves(bond) &&
1715             ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
1716                 eth_hw_addr_random(bond_dev);
1717
1718         return res;
1719 }
1720
1721 /* Try to release the slave device <slave> from the bond device <master>
1722  * It is legal to access curr_active_slave without a lock because all the function
1723  * is RTNL-locked. If "all" is true it means that the function is being called
1724  * while destroying a bond interface and all slaves are being released.
1725  *
1726  * The rules for slave state should be:
1727  *   for Active/Backup:
1728  *     Active stays on all backups go down
1729  *   for Bonded connections:
1730  *     The first up interface should be left on and all others downed.
1731  */
1732 static int __bond_release_one(struct net_device *bond_dev,
1733                               struct net_device *slave_dev,
1734                               bool all)
1735 {
1736         struct bonding *bond = netdev_priv(bond_dev);
1737         struct slave *slave, *oldcurrent;
1738         struct sockaddr addr;
1739         int old_flags = bond_dev->flags;
1740         netdev_features_t old_features = bond_dev->features;
1741
1742         /* slave is not a slave or master is not master of this slave */
1743         if (!(slave_dev->flags & IFF_SLAVE) ||
1744             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1745                 netdev_dbg(bond_dev, "cannot release %s\n",
1746                            slave_dev->name);
1747                 return -EINVAL;
1748         }
1749
1750         block_netpoll_tx();
1751
1752         slave = bond_get_slave_by_dev(bond, slave_dev);
1753         if (!slave) {
1754                 /* not a slave of this bond */
1755                 netdev_info(bond_dev, "%s not enslaved\n",
1756                             slave_dev->name);
1757                 unblock_netpoll_tx();
1758                 return -EINVAL;
1759         }
1760
1761         bond_sysfs_slave_del(slave);
1762
1763         /* recompute stats just before removing the slave */
1764         bond_get_stats(bond->dev, &bond->bond_stats);
1765
1766         bond_upper_dev_unlink(bond_dev, slave_dev);
1767         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1768          * for this slave anymore.
1769          */
1770         netdev_rx_handler_unregister(slave_dev);
1771
1772         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1773                 bond_3ad_unbind_slave(slave);
1774
1775         if (bond_mode_uses_xmit_hash(bond))
1776                 bond_update_slave_arr(bond, slave);
1777
1778         netdev_info(bond_dev, "Releasing %s interface %s\n",
1779                     bond_is_active_slave(slave) ? "active" : "backup",
1780                     slave_dev->name);
1781
1782         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1783
1784         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1785
1786         if (!all && (!bond->params.fail_over_mac ||
1787                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1788                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1789                     bond_has_slaves(bond))
1790                         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",
1791                                     slave_dev->name, slave->perm_hwaddr,
1792                                     bond_dev->name, slave_dev->name);
1793         }
1794
1795         if (rtnl_dereference(bond->primary_slave) == slave)
1796                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1797
1798         if (oldcurrent == slave)
1799                 bond_change_active_slave(bond, NULL);
1800
1801         if (bond_is_lb(bond)) {
1802                 /* Must be called only after the slave has been
1803                  * detached from the list and the curr_active_slave
1804                  * has been cleared (if our_slave == old_current),
1805                  * but before a new active slave is selected.
1806                  */
1807                 bond_alb_deinit_slave(bond, slave);
1808         }
1809
1810         if (all) {
1811                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1812         } else if (oldcurrent == slave) {
1813                 /* Note that we hold RTNL over this sequence, so there
1814                  * is no concern that another slave add/remove event
1815                  * will interfere.
1816                  */
1817                 bond_select_active_slave(bond);
1818         }
1819
1820         if (!bond_has_slaves(bond)) {
1821                 bond_set_carrier(bond);
1822                 eth_hw_addr_random(bond_dev);
1823         }
1824
1825         unblock_netpoll_tx();
1826         synchronize_rcu();
1827         bond->slave_cnt--;
1828
1829         if (!bond_has_slaves(bond)) {
1830                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1831                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1832         }
1833
1834         bond_compute_features(bond);
1835         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1836             (old_features & NETIF_F_VLAN_CHALLENGED))
1837                 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1838                             slave_dev->name, bond_dev->name);
1839
1840         vlan_vids_del_by_dev(slave_dev, bond_dev);
1841
1842         /* If the mode uses primary, then this case was handled above by
1843          * bond_change_active_slave(..., NULL)
1844          */
1845         if (!bond_uses_primary(bond)) {
1846                 /* unset promiscuity level from slave
1847                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1848                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1849                  * value of that flag before that change, as that was the value
1850                  * when this slave was attached, so we cache at the start of the
1851                  * function and use it here. Same goes for ALLMULTI below
1852                  */
1853                 if (old_flags & IFF_PROMISC)
1854                         dev_set_promiscuity(slave_dev, -1);
1855
1856                 /* unset allmulti level from slave */
1857                 if (old_flags & IFF_ALLMULTI)
1858                         dev_set_allmulti(slave_dev, -1);
1859
1860                 bond_hw_addr_flush(bond_dev, slave_dev);
1861         }
1862
1863         slave_disable_netpoll(slave);
1864
1865         /* close slave before restoring its mac address */
1866         dev_close(slave_dev);
1867
1868         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1869             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1870                 /* restore original ("permanent") mac address */
1871                 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1872                 addr.sa_family = slave_dev->type;
1873                 dev_set_mac_address(slave_dev, &addr);
1874         }
1875
1876         dev_set_mtu(slave_dev, slave->original_mtu);
1877
1878         slave_dev->priv_flags &= ~IFF_BONDING;
1879
1880         bond_free_slave(slave);
1881
1882         return 0;
1883 }
1884
1885 /* A wrapper used because of ndo_del_link */
1886 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1887 {
1888         return __bond_release_one(bond_dev, slave_dev, false);
1889 }
1890
1891 /* First release a slave and then destroy the bond if no more slaves are left.
1892  * Must be under rtnl_lock when this function is called.
1893  */
1894 static int  bond_release_and_destroy(struct net_device *bond_dev,
1895                                      struct net_device *slave_dev)
1896 {
1897         struct bonding *bond = netdev_priv(bond_dev);
1898         int ret;
1899
1900         ret = bond_release(bond_dev, slave_dev);
1901         if (ret == 0 && !bond_has_slaves(bond)) {
1902                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1903                 netdev_info(bond_dev, "Destroying bond %s\n",
1904                             bond_dev->name);
1905                 unregister_netdevice(bond_dev);
1906         }
1907         return ret;
1908 }
1909
1910 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1911 {
1912         struct bonding *bond = netdev_priv(bond_dev);
1913         bond_fill_ifbond(bond, info);
1914         return 0;
1915 }
1916
1917 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1918 {
1919         struct bonding *bond = netdev_priv(bond_dev);
1920         struct list_head *iter;
1921         int i = 0, res = -ENODEV;
1922         struct slave *slave;
1923
1924         bond_for_each_slave(bond, slave, iter) {
1925                 if (i++ == (int)info->slave_id) {
1926                         res = 0;
1927                         bond_fill_ifslave(slave, info);
1928                         break;
1929                 }
1930         }
1931
1932         return res;
1933 }
1934
1935 /*-------------------------------- Monitoring -------------------------------*/
1936
1937 /* called with rcu_read_lock() */
1938 static int bond_miimon_inspect(struct bonding *bond)
1939 {
1940         int link_state, commit = 0;
1941         struct list_head *iter;
1942         struct slave *slave;
1943         bool ignore_updelay;
1944
1945         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
1946
1947         bond_for_each_slave_rcu(bond, slave, iter) {
1948                 slave->new_link = BOND_LINK_NOCHANGE;
1949
1950                 link_state = bond_check_dev_link(bond, slave->dev, 0);
1951
1952                 switch (slave->link) {
1953                 case BOND_LINK_UP:
1954                         if (link_state)
1955                                 continue;
1956
1957                         bond_set_slave_link_state(slave, BOND_LINK_FAIL);
1958                         slave->delay = bond->params.downdelay;
1959                         if (slave->delay) {
1960                                 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
1961                                             (BOND_MODE(bond) ==
1962                                              BOND_MODE_ACTIVEBACKUP) ?
1963                                              (bond_is_active_slave(slave) ?
1964                                               "active " : "backup ") : "",
1965                                             slave->dev->name,
1966                                             bond->params.downdelay * bond->params.miimon);
1967                         }
1968                         /*FALLTHRU*/
1969                 case BOND_LINK_FAIL:
1970                         if (link_state) {
1971                                 /* recovered before downdelay expired */
1972                                 bond_set_slave_link_state(slave, BOND_LINK_UP);
1973                                 slave->last_link_up = jiffies;
1974                                 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
1975                                             (bond->params.downdelay - slave->delay) *
1976                                             bond->params.miimon,
1977                                             slave->dev->name);
1978                                 continue;
1979                         }
1980
1981                         if (slave->delay <= 0) {
1982                                 slave->new_link = BOND_LINK_DOWN;
1983                                 commit++;
1984                                 continue;
1985                         }
1986
1987                         slave->delay--;
1988                         break;
1989
1990                 case BOND_LINK_DOWN:
1991                         if (!link_state)
1992                                 continue;
1993
1994                         bond_set_slave_link_state(slave, BOND_LINK_BACK);
1995                         slave->delay = bond->params.updelay;
1996
1997                         if (slave->delay) {
1998                                 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
1999                                             slave->dev->name,
2000                                             ignore_updelay ? 0 :
2001                                             bond->params.updelay *
2002                                             bond->params.miimon);
2003                         }
2004                         /*FALLTHRU*/
2005                 case BOND_LINK_BACK:
2006                         if (!link_state) {
2007                                 bond_set_slave_link_state(slave,
2008                                                           BOND_LINK_DOWN);
2009                                 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2010                                             (bond->params.updelay - slave->delay) *
2011                                             bond->params.miimon,
2012                                             slave->dev->name);
2013
2014                                 continue;
2015                         }
2016
2017                         if (ignore_updelay)
2018                                 slave->delay = 0;
2019
2020                         if (slave->delay <= 0) {
2021                                 slave->new_link = BOND_LINK_UP;
2022                                 commit++;
2023                                 ignore_updelay = false;
2024                                 continue;
2025                         }
2026
2027                         slave->delay--;
2028                         break;
2029                 }
2030         }
2031
2032         return commit;
2033 }
2034
2035 static void bond_miimon_commit(struct bonding *bond)
2036 {
2037         struct list_head *iter;
2038         struct slave *slave, *primary;
2039
2040         bond_for_each_slave(bond, slave, iter) {
2041                 switch (slave->new_link) {
2042                 case BOND_LINK_NOCHANGE:
2043                         continue;
2044
2045                 case BOND_LINK_UP:
2046                         bond_set_slave_link_state(slave, BOND_LINK_UP);
2047                         slave->last_link_up = jiffies;
2048
2049                         primary = rtnl_dereference(bond->primary_slave);
2050                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2051                                 /* prevent it from being the active one */
2052                                 bond_set_backup_slave(slave);
2053                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2054                                 /* make it immediately active */
2055                                 bond_set_active_slave(slave);
2056                         } else if (slave != primary) {
2057                                 /* prevent it from being the active one */
2058                                 bond_set_backup_slave(slave);
2059                         }
2060
2061                         netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2062                                     slave->dev->name,
2063                                     slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2064                                     slave->duplex ? "full" : "half");
2065
2066                         /* notify ad that the link status has changed */
2067                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2068                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2069
2070                         if (bond_is_lb(bond))
2071                                 bond_alb_handle_link_change(bond, slave,
2072                                                             BOND_LINK_UP);
2073
2074                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2075                                 bond_update_slave_arr(bond, NULL);
2076
2077                         if (!bond->curr_active_slave || slave == primary)
2078                                 goto do_failover;
2079
2080                         continue;
2081
2082                 case BOND_LINK_DOWN:
2083                         if (slave->link_failure_count < UINT_MAX)
2084                                 slave->link_failure_count++;
2085
2086                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2087
2088                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2089                             BOND_MODE(bond) == BOND_MODE_8023AD)
2090                                 bond_set_slave_inactive_flags(slave,
2091                                                               BOND_SLAVE_NOTIFY_NOW);
2092
2093                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2094                                     slave->dev->name);
2095
2096                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2097                                 bond_3ad_handle_link_change(slave,
2098                                                             BOND_LINK_DOWN);
2099
2100                         if (bond_is_lb(bond))
2101                                 bond_alb_handle_link_change(bond, slave,
2102                                                             BOND_LINK_DOWN);
2103
2104                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2105                                 bond_update_slave_arr(bond, NULL);
2106
2107                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2108                                 goto do_failover;
2109
2110                         continue;
2111
2112                 default:
2113                         netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2114                                    slave->new_link, slave->dev->name);
2115                         slave->new_link = BOND_LINK_NOCHANGE;
2116
2117                         continue;
2118                 }
2119
2120 do_failover:
2121                 block_netpoll_tx();
2122                 bond_select_active_slave(bond);
2123                 unblock_netpoll_tx();
2124         }
2125
2126         bond_set_carrier(bond);
2127 }
2128
2129 /* bond_mii_monitor
2130  *
2131  * Really a wrapper that splits the mii monitor into two phases: an
2132  * inspection, then (if inspection indicates something needs to be done)
2133  * an acquisition of appropriate locks followed by a commit phase to
2134  * implement whatever link state changes are indicated.
2135  */
2136 static void bond_mii_monitor(struct work_struct *work)
2137 {
2138         struct bonding *bond = container_of(work, struct bonding,
2139                                             mii_work.work);
2140         bool should_notify_peers = false;
2141         unsigned long delay;
2142
2143         delay = msecs_to_jiffies(bond->params.miimon);
2144
2145         if (!bond_has_slaves(bond))
2146                 goto re_arm;
2147
2148         rcu_read_lock();
2149
2150         should_notify_peers = bond_should_notify_peers(bond);
2151
2152         if (bond_miimon_inspect(bond)) {
2153                 rcu_read_unlock();
2154
2155                 /* Race avoidance with bond_close cancel of workqueue */
2156                 if (!rtnl_trylock()) {
2157                         delay = 1;
2158                         should_notify_peers = false;
2159                         goto re_arm;
2160                 }
2161
2162                 bond_miimon_commit(bond);
2163
2164                 rtnl_unlock();  /* might sleep, hold no other locks */
2165         } else
2166                 rcu_read_unlock();
2167
2168 re_arm:
2169         if (bond->params.miimon)
2170                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2171
2172         if (should_notify_peers) {
2173                 if (!rtnl_trylock())
2174                         return;
2175                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2176                 rtnl_unlock();
2177         }
2178 }
2179
2180 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2181 {
2182         struct net_device *upper;
2183         struct list_head *iter;
2184         bool ret = false;
2185
2186         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2187                 return true;
2188
2189         rcu_read_lock();
2190         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2191                 if (ip == bond_confirm_addr(upper, 0, ip)) {
2192                         ret = true;
2193                         break;
2194                 }
2195         }
2196         rcu_read_unlock();
2197
2198         return ret;
2199 }
2200
2201 /* We go to the (large) trouble of VLAN tagging ARP frames because
2202  * switches in VLAN mode (especially if ports are configured as
2203  * "native" to a VLAN) might not pass non-tagged frames.
2204  */
2205 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2206                           __be32 dest_ip, __be32 src_ip,
2207                           struct bond_vlan_tag *tags)
2208 {
2209         struct sk_buff *skb;
2210         struct bond_vlan_tag *outer_tag = tags;
2211
2212         netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2213                    arp_op, slave_dev->name, &dest_ip, &src_ip);
2214
2215         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2216                          NULL, slave_dev->dev_addr, NULL);
2217
2218         if (!skb) {
2219                 net_err_ratelimited("ARP packet allocation failed\n");
2220                 return;
2221         }
2222
2223         if (!tags || tags->vlan_proto == VLAN_N_VID)
2224                 goto xmit;
2225
2226         tags++;
2227
2228         /* Go through all the tags backwards and add them to the packet */
2229         while (tags->vlan_proto != VLAN_N_VID) {
2230                 if (!tags->vlan_id) {
2231                         tags++;
2232                         continue;
2233                 }
2234
2235                 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2236                            ntohs(outer_tag->vlan_proto), tags->vlan_id);
2237                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2238                                                 tags->vlan_id);
2239                 if (!skb) {
2240                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2241                         return;
2242                 }
2243
2244                 tags++;
2245         }
2246         /* Set the outer tag */
2247         if (outer_tag->vlan_id) {
2248                 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2249                            ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2250                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2251                                        outer_tag->vlan_id);
2252         }
2253
2254 xmit:
2255         arp_xmit(skb);
2256 }
2257
2258 /* Validate the device path between the @start_dev and the @end_dev.
2259  * The path is valid if the @end_dev is reachable through device
2260  * stacking.
2261  * When the path is validated, collect any vlan information in the
2262  * path.
2263  */
2264 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2265                                               struct net_device *end_dev,
2266                                               int level)
2267 {
2268         struct bond_vlan_tag *tags;
2269         struct net_device *upper;
2270         struct list_head  *iter;
2271
2272         if (start_dev == end_dev) {
2273                 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2274                 if (!tags)
2275                         return ERR_PTR(-ENOMEM);
2276                 tags[level].vlan_proto = VLAN_N_VID;
2277                 return tags;
2278         }
2279
2280         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2281                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2282                 if (IS_ERR_OR_NULL(tags)) {
2283                         if (IS_ERR(tags))
2284                                 return tags;
2285                         continue;
2286                 }
2287                 if (is_vlan_dev(upper)) {
2288                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2289                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2290                 }
2291
2292                 return tags;
2293         }
2294
2295         return NULL;
2296 }
2297
2298 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2299 {
2300         struct rtable *rt;
2301         struct bond_vlan_tag *tags;
2302         __be32 *targets = bond->params.arp_targets, addr;
2303         int i;
2304
2305         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2306                 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2307                 tags = NULL;
2308
2309                 /* Find out through which dev should the packet go */
2310                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2311                                      RTO_ONLINK, 0);
2312                 if (IS_ERR(rt)) {
2313                         /* there's no route to target - try to send arp
2314                          * probe to generate any traffic (arp_validate=0)
2315                          */
2316                         if (bond->params.arp_validate)
2317                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2318                                                      bond->dev->name,
2319                                                      &targets[i]);
2320                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2321                                       0, tags);
2322                         continue;
2323                 }
2324
2325                 /* bond device itself */
2326                 if (rt->dst.dev == bond->dev)
2327                         goto found;
2328
2329                 rcu_read_lock();
2330                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2331                 rcu_read_unlock();
2332
2333                 if (!IS_ERR_OR_NULL(tags))
2334                         goto found;
2335
2336                 /* Not our device - skip */
2337                 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2338                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2339
2340                 ip_rt_put(rt);
2341                 continue;
2342
2343 found:
2344                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2345                 ip_rt_put(rt);
2346                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2347                               addr, tags);
2348                 kfree(tags);
2349         }
2350 }
2351
2352 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2353 {
2354         int i;
2355
2356         if (!sip || !bond_has_this_ip(bond, tip)) {
2357                 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2358                            &sip, &tip);
2359                 return;
2360         }
2361
2362         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2363         if (i == -1) {
2364                 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2365                            &sip);
2366                 return;
2367         }
2368         slave->last_rx = jiffies;
2369         slave->target_last_arp_rx[i] = jiffies;
2370 }
2371
2372 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2373                  struct slave *slave)
2374 {
2375         struct arphdr *arp = (struct arphdr *)skb->data;
2376         struct slave *curr_active_slave;
2377         unsigned char *arp_ptr;
2378         __be32 sip, tip;
2379         int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2380
2381         if (!slave_do_arp_validate(bond, slave)) {
2382                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2383                     !slave_do_arp_validate_only(bond))
2384                         slave->last_rx = jiffies;
2385                 return RX_HANDLER_ANOTHER;
2386         } else if (!is_arp) {
2387                 return RX_HANDLER_ANOTHER;
2388         }
2389
2390         alen = arp_hdr_len(bond->dev);
2391
2392         netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2393                    skb->dev->name);
2394
2395         if (alen > skb_headlen(skb)) {
2396                 arp = kmalloc(alen, GFP_ATOMIC);
2397                 if (!arp)
2398                         goto out_unlock;
2399                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2400                         goto out_unlock;
2401         }
2402
2403         if (arp->ar_hln != bond->dev->addr_len ||
2404             skb->pkt_type == PACKET_OTHERHOST ||
2405             skb->pkt_type == PACKET_LOOPBACK ||
2406             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2407             arp->ar_pro != htons(ETH_P_IP) ||
2408             arp->ar_pln != 4)
2409                 goto out_unlock;
2410
2411         arp_ptr = (unsigned char *)(arp + 1);
2412         arp_ptr += bond->dev->addr_len;
2413         memcpy(&sip, arp_ptr, 4);
2414         arp_ptr += 4 + bond->dev->addr_len;
2415         memcpy(&tip, arp_ptr, 4);
2416
2417         netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2418                    slave->dev->name, bond_slave_state(slave),
2419                      bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2420                      &sip, &tip);
2421
2422         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2423
2424         /* Backup slaves won't see the ARP reply, but do come through
2425          * here for each ARP probe (so we swap the sip/tip to validate
2426          * the probe).  In a "redundant switch, common router" type of
2427          * configuration, the ARP probe will (hopefully) travel from
2428          * the active, through one switch, the router, then the other
2429          * switch before reaching the backup.
2430          *
2431          * We 'trust' the arp requests if there is an active slave and
2432          * it received valid arp reply(s) after it became active. This
2433          * is done to avoid endless looping when we can't reach the
2434          * arp_ip_target and fool ourselves with our own arp requests.
2435          */
2436
2437         if (bond_is_active_slave(slave))
2438                 bond_validate_arp(bond, slave, sip, tip);
2439         else if (curr_active_slave &&
2440                  time_after(slave_last_rx(bond, curr_active_slave),
2441                             curr_active_slave->last_link_up))
2442                 bond_validate_arp(bond, slave, tip, sip);
2443
2444 out_unlock:
2445         if (arp != (struct arphdr *)skb->data)
2446                 kfree(arp);
2447         return RX_HANDLER_ANOTHER;
2448 }
2449
2450 /* function to verify if we're in the arp_interval timeslice, returns true if
2451  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2452  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2453  */
2454 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2455                                   int mod)
2456 {
2457         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2458
2459         return time_in_range(jiffies,
2460                              last_act - delta_in_ticks,
2461                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2462 }
2463
2464 /* This function is called regularly to monitor each slave's link
2465  * ensuring that traffic is being sent and received when arp monitoring
2466  * is used in load-balancing mode. if the adapter has been dormant, then an
2467  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2468  * arp monitoring in active backup mode.
2469  */
2470 static void bond_loadbalance_arp_mon(struct work_struct *work)
2471 {
2472         struct bonding *bond = container_of(work, struct bonding,
2473                                             arp_work.work);
2474         struct slave *slave, *oldcurrent;
2475         struct list_head *iter;
2476         int do_failover = 0, slave_state_changed = 0;
2477
2478         if (!bond_has_slaves(bond))
2479                 goto re_arm;
2480
2481         rcu_read_lock();
2482
2483         oldcurrent = rcu_dereference(bond->curr_active_slave);
2484         /* see if any of the previous devices are up now (i.e. they have
2485          * xmt and rcv traffic). the curr_active_slave does not come into
2486          * the picture unless it is null. also, slave->last_link_up is not
2487          * needed here because we send an arp on each slave and give a slave
2488          * as long as it needs to get the tx/rx within the delta.
2489          * TODO: what about up/down delay in arp mode? it wasn't here before
2490          *       so it can wait
2491          */
2492         bond_for_each_slave_rcu(bond, slave, iter) {
2493                 unsigned long trans_start = dev_trans_start(slave->dev);
2494
2495                 if (slave->link != BOND_LINK_UP) {
2496                         if (bond_time_in_interval(bond, trans_start, 1) &&
2497                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2498
2499                                 slave->link  = BOND_LINK_UP;
2500                                 slave_state_changed = 1;
2501
2502                                 /* primary_slave has no meaning in round-robin
2503                                  * mode. the window of a slave being up and
2504                                  * curr_active_slave being null after enslaving
2505                                  * is closed.
2506                                  */
2507                                 if (!oldcurrent) {
2508                                         netdev_info(bond->dev, "link status definitely up for interface %s\n",
2509                                                     slave->dev->name);
2510                                         do_failover = 1;
2511                                 } else {
2512                                         netdev_info(bond->dev, "interface %s is now up\n",
2513                                                     slave->dev->name);
2514                                 }
2515                         }
2516                 } else {
2517                         /* slave->link == BOND_LINK_UP */
2518
2519                         /* not all switches will respond to an arp request
2520                          * when the source ip is 0, so don't take the link down
2521                          * if we don't know our ip yet
2522                          */
2523                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2524                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2525
2526                                 slave->link  = BOND_LINK_DOWN;
2527                                 slave_state_changed = 1;
2528
2529                                 if (slave->link_failure_count < UINT_MAX)
2530                                         slave->link_failure_count++;
2531
2532                                 netdev_info(bond->dev, "interface %s is now down\n",
2533                                             slave->dev->name);
2534
2535                                 if (slave == oldcurrent)
2536                                         do_failover = 1;
2537                         }
2538                 }
2539
2540                 /* note: if switch is in round-robin mode, all links
2541                  * must tx arp to ensure all links rx an arp - otherwise
2542                  * links may oscillate or not come up at all; if switch is
2543                  * in something like xor mode, there is nothing we can
2544                  * do - all replies will be rx'ed on same link causing slaves
2545                  * to be unstable during low/no traffic periods
2546                  */
2547                 if (bond_slave_is_up(slave))
2548                         bond_arp_send_all(bond, slave);
2549         }
2550
2551         rcu_read_unlock();
2552
2553         if (do_failover || slave_state_changed) {
2554                 if (!rtnl_trylock())
2555                         goto re_arm;
2556
2557                 if (slave_state_changed) {
2558                         bond_slave_state_change(bond);
2559                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2560                                 bond_update_slave_arr(bond, NULL);
2561                 }
2562                 if (do_failover) {
2563                         block_netpoll_tx();
2564                         bond_select_active_slave(bond);
2565                         unblock_netpoll_tx();
2566                 }
2567                 rtnl_unlock();
2568         }
2569
2570 re_arm:
2571         if (bond->params.arp_interval)
2572                 queue_delayed_work(bond->wq, &bond->arp_work,
2573                                    msecs_to_jiffies(bond->params.arp_interval));
2574 }
2575
2576 /* Called to inspect slaves for active-backup mode ARP monitor link state
2577  * changes.  Sets new_link in slaves to specify what action should take
2578  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2579  * to link states must be committed.
2580  *
2581  * Called with rcu_read_lock held.
2582  */
2583 static int bond_ab_arp_inspect(struct bonding *bond)
2584 {
2585         unsigned long trans_start, last_rx;
2586         struct list_head *iter;
2587         struct slave *slave;
2588         int commit = 0;
2589
2590         bond_for_each_slave_rcu(bond, slave, iter) {
2591                 slave->new_link = BOND_LINK_NOCHANGE;
2592                 last_rx = slave_last_rx(bond, slave);
2593
2594                 if (slave->link != BOND_LINK_UP) {
2595                         if (bond_time_in_interval(bond, last_rx, 1)) {
2596                                 slave->new_link = BOND_LINK_UP;
2597                                 commit++;
2598                         }
2599                         continue;
2600                 }
2601
2602                 /* Give slaves 2*delta after being enslaved or made
2603                  * active.  This avoids bouncing, as the last receive
2604                  * times need a full ARP monitor cycle to be updated.
2605                  */
2606                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2607                         continue;
2608
2609                 /* Backup slave is down if:
2610                  * - No current_arp_slave AND
2611                  * - more than 3*delta since last receive AND
2612                  * - the bond has an IP address
2613                  *
2614                  * Note: a non-null current_arp_slave indicates
2615                  * the curr_active_slave went down and we are
2616                  * searching for a new one; under this condition
2617                  * we only take the curr_active_slave down - this
2618                  * gives each slave a chance to tx/rx traffic
2619                  * before being taken out
2620                  */
2621                 if (!bond_is_active_slave(slave) &&
2622                     !rcu_access_pointer(bond->current_arp_slave) &&
2623                     !bond_time_in_interval(bond, last_rx, 3)) {
2624                         slave->new_link = BOND_LINK_DOWN;
2625                         commit++;
2626                 }
2627
2628                 /* Active slave is down if:
2629                  * - more than 2*delta since transmitting OR
2630                  * - (more than 2*delta since receive AND
2631                  *    the bond has an IP address)
2632                  */
2633                 trans_start = dev_trans_start(slave->dev);
2634                 if (bond_is_active_slave(slave) &&
2635                     (!bond_time_in_interval(bond, trans_start, 2) ||
2636                      !bond_time_in_interval(bond, last_rx, 2))) {
2637                         slave->new_link = BOND_LINK_DOWN;
2638                         commit++;
2639                 }
2640         }
2641
2642         return commit;
2643 }
2644
2645 /* Called to commit link state changes noted by inspection step of
2646  * active-backup mode ARP monitor.
2647  *
2648  * Called with RTNL hold.
2649  */
2650 static void bond_ab_arp_commit(struct bonding *bond)
2651 {
2652         unsigned long trans_start;
2653         struct list_head *iter;
2654         struct slave *slave;
2655
2656         bond_for_each_slave(bond, slave, iter) {
2657                 switch (slave->new_link) {
2658                 case BOND_LINK_NOCHANGE:
2659                         continue;
2660
2661                 case BOND_LINK_UP:
2662                         trans_start = dev_trans_start(slave->dev);
2663                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
2664                             (!rtnl_dereference(bond->curr_active_slave) &&
2665                              bond_time_in_interval(bond, trans_start, 1))) {
2666                                 struct slave *current_arp_slave;
2667
2668                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2669                                 bond_set_slave_link_state(slave, BOND_LINK_UP);
2670                                 if (current_arp_slave) {
2671                                         bond_set_slave_inactive_flags(
2672                                                 current_arp_slave,
2673                                                 BOND_SLAVE_NOTIFY_NOW);
2674                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2675                                 }
2676
2677                                 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2678                                             slave->dev->name);
2679
2680                                 if (!rtnl_dereference(bond->curr_active_slave) ||
2681                                     slave == rtnl_dereference(bond->primary_slave))
2682                                         goto do_failover;
2683
2684                         }
2685
2686                         continue;
2687
2688                 case BOND_LINK_DOWN:
2689                         if (slave->link_failure_count < UINT_MAX)
2690                                 slave->link_failure_count++;
2691
2692                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2693                         bond_set_slave_inactive_flags(slave,
2694                                                       BOND_SLAVE_NOTIFY_NOW);
2695
2696                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2697                                     slave->dev->name);
2698
2699                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
2700                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2701                                 goto do_failover;
2702                         }
2703
2704                         continue;
2705
2706                 default:
2707                         netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2708                                    slave->new_link, slave->dev->name);
2709                         continue;
2710                 }
2711
2712 do_failover:
2713                 block_netpoll_tx();
2714                 bond_select_active_slave(bond);
2715                 unblock_netpoll_tx();
2716         }
2717
2718         bond_set_carrier(bond);
2719 }
2720
2721 /* Send ARP probes for active-backup mode ARP monitor.
2722  *
2723  * Called with rcu_read_lock held.
2724  */
2725 static bool bond_ab_arp_probe(struct bonding *bond)
2726 {
2727         struct slave *slave, *before = NULL, *new_slave = NULL,
2728                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2729                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2730         struct list_head *iter;
2731         bool found = false;
2732         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2733
2734         if (curr_arp_slave && curr_active_slave)
2735                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2736                             curr_arp_slave->dev->name,
2737                             curr_active_slave->dev->name);
2738
2739         if (curr_active_slave) {
2740                 bond_arp_send_all(bond, curr_active_slave);
2741                 return should_notify_rtnl;
2742         }
2743
2744         /* if we don't have a curr_active_slave, search for the next available
2745          * backup slave from the current_arp_slave and make it the candidate
2746          * for becoming the curr_active_slave
2747          */
2748
2749         if (!curr_arp_slave) {
2750                 curr_arp_slave = bond_first_slave_rcu(bond);
2751                 if (!curr_arp_slave)
2752                         return should_notify_rtnl;
2753         }
2754
2755         bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2756
2757         bond_for_each_slave_rcu(bond, slave, iter) {
2758                 if (!found && !before && bond_slave_is_up(slave))
2759                         before = slave;
2760
2761                 if (found && !new_slave && bond_slave_is_up(slave))
2762                         new_slave = slave;
2763                 /* if the link state is up at this point, we
2764                  * mark it down - this can happen if we have
2765                  * simultaneous link failures and
2766                  * reselect_active_interface doesn't make this
2767                  * one the current slave so it is still marked
2768                  * up when it is actually down
2769                  */
2770                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2771                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2772                         if (slave->link_failure_count < UINT_MAX)
2773                                 slave->link_failure_count++;
2774
2775                         bond_set_slave_inactive_flags(slave,
2776                                                       BOND_SLAVE_NOTIFY_LATER);
2777
2778                         netdev_info(bond->dev, "backup interface %s is now down\n",
2779                                     slave->dev->name);
2780                 }
2781                 if (slave == curr_arp_slave)
2782                         found = true;
2783         }
2784
2785         if (!new_slave && before)
2786                 new_slave = before;
2787
2788         if (!new_slave)
2789                 goto check_state;
2790
2791         bond_set_slave_link_state(new_slave, BOND_LINK_BACK);
2792         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2793         bond_arp_send_all(bond, new_slave);
2794         new_slave->last_link_up = jiffies;
2795         rcu_assign_pointer(bond->current_arp_slave, new_slave);
2796
2797 check_state:
2798         bond_for_each_slave_rcu(bond, slave, iter) {
2799                 if (slave->should_notify) {
2800                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2801                         break;
2802                 }
2803         }
2804         return should_notify_rtnl;
2805 }
2806
2807 static void bond_activebackup_arp_mon(struct work_struct *work)
2808 {
2809         struct bonding *bond = container_of(work, struct bonding,
2810                                             arp_work.work);
2811         bool should_notify_peers = false;
2812         bool should_notify_rtnl = false;
2813         int delta_in_ticks;
2814
2815         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2816
2817         if (!bond_has_slaves(bond))
2818                 goto re_arm;
2819
2820         rcu_read_lock();
2821
2822         should_notify_peers = bond_should_notify_peers(bond);
2823
2824         if (bond_ab_arp_inspect(bond)) {
2825                 rcu_read_unlock();
2826
2827                 /* Race avoidance with bond_close flush of workqueue */
2828                 if (!rtnl_trylock()) {
2829                         delta_in_ticks = 1;
2830                         should_notify_peers = false;
2831                         goto re_arm;
2832                 }
2833
2834                 bond_ab_arp_commit(bond);
2835
2836                 rtnl_unlock();
2837                 rcu_read_lock();
2838         }
2839
2840         should_notify_rtnl = bond_ab_arp_probe(bond);
2841         rcu_read_unlock();
2842
2843 re_arm:
2844         if (bond->params.arp_interval)
2845                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2846
2847         if (should_notify_peers || should_notify_rtnl) {
2848                 if (!rtnl_trylock())
2849                         return;
2850
2851                 if (should_notify_peers)
2852                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2853                                                  bond->dev);
2854                 if (should_notify_rtnl)
2855                         bond_slave_state_notify(bond);
2856
2857                 rtnl_unlock();
2858         }
2859 }
2860
2861 /*-------------------------- netdev event handling --------------------------*/
2862
2863 /* Change device name */
2864 static int bond_event_changename(struct bonding *bond)
2865 {
2866         bond_remove_proc_entry(bond);
2867         bond_create_proc_entry(bond);
2868
2869         bond_debug_reregister(bond);
2870
2871         return NOTIFY_DONE;
2872 }
2873
2874 static int bond_master_netdev_event(unsigned long event,
2875                                     struct net_device *bond_dev)
2876 {
2877         struct bonding *event_bond = netdev_priv(bond_dev);
2878
2879         switch (event) {
2880         case NETDEV_CHANGENAME:
2881                 return bond_event_changename(event_bond);
2882         case NETDEV_UNREGISTER:
2883                 bond_remove_proc_entry(event_bond);
2884                 break;
2885         case NETDEV_REGISTER:
2886                 bond_create_proc_entry(event_bond);
2887                 break;
2888         case NETDEV_NOTIFY_PEERS:
2889                 if (event_bond->send_peer_notif)
2890                         event_bond->send_peer_notif--;
2891                 break;
2892         default:
2893                 break;
2894         }
2895
2896         return NOTIFY_DONE;
2897 }
2898
2899 static int bond_slave_netdev_event(unsigned long event,
2900                                    struct net_device *slave_dev)
2901 {
2902         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
2903         struct bonding *bond;
2904         struct net_device *bond_dev;
2905         u32 old_speed;
2906         u8 old_duplex;
2907
2908         /* A netdev event can be generated while enslaving a device
2909          * before netdev_rx_handler_register is called in which case
2910          * slave will be NULL
2911          */
2912         if (!slave)
2913                 return NOTIFY_DONE;
2914         bond_dev = slave->bond->dev;
2915         bond = slave->bond;
2916         primary = rtnl_dereference(bond->primary_slave);
2917
2918         switch (event) {
2919         case NETDEV_UNREGISTER:
2920                 if (bond_dev->type != ARPHRD_ETHER)
2921                         bond_release_and_destroy(bond_dev, slave_dev);
2922                 else
2923                         bond_release(bond_dev, slave_dev);
2924                 break;
2925         case NETDEV_UP:
2926         case NETDEV_CHANGE:
2927                 old_speed = slave->speed;
2928                 old_duplex = slave->duplex;
2929
2930                 bond_update_speed_duplex(slave);
2931
2932                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2933                         if (old_speed != slave->speed)
2934                                 bond_3ad_adapter_speed_changed(slave);
2935                         if (old_duplex != slave->duplex)
2936                                 bond_3ad_adapter_duplex_changed(slave);
2937                 }
2938                 /* Fallthrough */
2939         case NETDEV_DOWN:
2940                 /* Refresh slave-array if applicable!
2941                  * If the setup does not use miimon or arpmon (mode-specific!),
2942                  * then these events will not cause the slave-array to be
2943                  * refreshed. This will cause xmit to use a slave that is not
2944                  * usable. Avoid such situation by refeshing the array at these
2945                  * events. If these (miimon/arpmon) parameters are configured
2946                  * then array gets refreshed twice and that should be fine!
2947                  */
2948                 if (bond_mode_uses_xmit_hash(bond))
2949                         bond_update_slave_arr(bond, NULL);
2950                 break;
2951         case NETDEV_CHANGEMTU:
2952                 /* TODO: Should slaves be allowed to
2953                  * independently alter their MTU?  For
2954                  * an active-backup bond, slaves need
2955                  * not be the same type of device, so
2956                  * MTUs may vary.  For other modes,
2957                  * slaves arguably should have the
2958                  * same MTUs. To do this, we'd need to
2959                  * take over the slave's change_mtu
2960                  * function for the duration of their
2961                  * servitude.
2962                  */
2963                 break;
2964         case NETDEV_CHANGENAME:
2965                 /* we don't care if we don't have primary set */
2966                 if (!bond_uses_primary(bond) ||
2967                     !bond->params.primary[0])
2968                         break;
2969
2970                 if (slave == primary) {
2971                         /* slave's name changed - he's no longer primary */
2972                         RCU_INIT_POINTER(bond->primary_slave, NULL);
2973                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2974                         /* we have a new primary slave */
2975                         rcu_assign_pointer(bond->primary_slave, slave);
2976                 } else { /* we didn't change primary - exit */
2977                         break;
2978                 }
2979
2980                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
2981                             primary ? slave_dev->name : "none");
2982
2983                 block_netpoll_tx();
2984                 bond_select_active_slave(bond);
2985                 unblock_netpoll_tx();
2986                 break;
2987         case NETDEV_FEAT_CHANGE:
2988                 bond_compute_features(bond);
2989                 break;
2990         case NETDEV_RESEND_IGMP:
2991                 /* Propagate to master device */
2992                 call_netdevice_notifiers(event, slave->bond->dev);
2993                 break;
2994         default:
2995                 break;
2996         }
2997
2998         return NOTIFY_DONE;
2999 }
3000
3001 /* bond_netdev_event: handle netdev notifier chain events.
3002  *
3003  * This function receives events for the netdev chain.  The caller (an
3004  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3005  * locks for us to safely manipulate the slave devices (RTNL lock,
3006  * dev_probe_lock).
3007  */
3008 static int bond_netdev_event(struct notifier_block *this,
3009                              unsigned long event, void *ptr)
3010 {
3011         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3012
3013         netdev_dbg(event_dev, "event: %lx\n", event);
3014
3015         if (!(event_dev->priv_flags & IFF_BONDING))
3016                 return NOTIFY_DONE;
3017
3018         if (event_dev->flags & IFF_MASTER) {
3019                 netdev_dbg(event_dev, "IFF_MASTER\n");
3020                 return bond_master_netdev_event(event, event_dev);
3021         }
3022
3023         if (event_dev->flags & IFF_SLAVE) {
3024                 netdev_dbg(event_dev, "IFF_SLAVE\n");
3025                 return bond_slave_netdev_event(event, event_dev);
3026         }
3027
3028         return NOTIFY_DONE;
3029 }
3030
3031 static struct notifier_block bond_netdev_notifier = {
3032         .notifier_call = bond_netdev_event,
3033 };
3034
3035 /*---------------------------- Hashing Policies -----------------------------*/
3036
3037 /* L2 hash helper */
3038 static inline u32 bond_eth_hash(struct sk_buff *skb)
3039 {
3040         struct ethhdr *ep, hdr_tmp;
3041
3042         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3043         if (ep)
3044                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3045         return 0;
3046 }
3047
3048 /* Extract the appropriate headers based on bond's xmit policy */
3049 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3050                               struct flow_keys *fk)
3051 {
3052         const struct ipv6hdr *iph6;
3053         const struct iphdr *iph;
3054         int noff, proto = -1;
3055
3056         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3057                 return skb_flow_dissect(skb, fk);
3058
3059         fk->ports = 0;
3060         noff = skb_network_offset(skb);
3061         if (skb->protocol == htons(ETH_P_IP)) {
3062                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3063                         return false;
3064                 iph = ip_hdr(skb);
3065                 fk->src = iph->saddr;
3066                 fk->dst = iph->daddr;
3067                 noff += iph->ihl << 2;
3068                 if (!ip_is_fragment(iph))
3069                         proto = iph->protocol;
3070         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3071                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3072                         return false;
3073                 iph6 = ipv6_hdr(skb);
3074                 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
3075                 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
3076                 noff += sizeof(*iph6);
3077                 proto = iph6->nexthdr;
3078         } else {
3079                 return false;
3080         }
3081         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3082                 fk->ports = skb_flow_get_ports(skb, noff, proto);
3083
3084         return true;
3085 }
3086
3087 /**
3088  * bond_xmit_hash - generate a hash value based on the xmit policy
3089  * @bond: bonding device
3090  * @skb: buffer to use for headers
3091  *
3092  * This function will extract the necessary headers from the skb buffer and use
3093  * them to generate a hash based on the xmit_policy set in the bonding device
3094  */
3095 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3096 {
3097         struct flow_keys flow;
3098         u32 hash;
3099
3100         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3101             !bond_flow_dissect(bond, skb, &flow))
3102                 return bond_eth_hash(skb);
3103
3104         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3105             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3106                 hash = bond_eth_hash(skb);
3107         else
3108                 hash = (__force u32)flow.ports;
3109         hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
3110         hash ^= (hash >> 16);
3111         hash ^= (hash >> 8);
3112
3113         return hash;
3114 }
3115
3116 /*-------------------------- Device entry points ----------------------------*/
3117
3118 static void bond_work_init_all(struct bonding *bond)
3119 {
3120         INIT_DELAYED_WORK(&bond->mcast_work,
3121                           bond_resend_igmp_join_requests_delayed);
3122         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3123         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3124         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3125                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3126         else
3127                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3128         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3129         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3130 }
3131
3132 static void bond_work_cancel_all(struct bonding *bond)
3133 {
3134         cancel_delayed_work_sync(&bond->mii_work);
3135         cancel_delayed_work_sync(&bond->arp_work);
3136         cancel_delayed_work_sync(&bond->alb_work);
3137         cancel_delayed_work_sync(&bond->ad_work);
3138         cancel_delayed_work_sync(&bond->mcast_work);
3139         cancel_delayed_work_sync(&bond->slave_arr_work);
3140 }
3141
3142 static int bond_open(struct net_device *bond_dev)
3143 {
3144         struct bonding *bond = netdev_priv(bond_dev);
3145         struct list_head *iter;
3146         struct slave *slave;
3147
3148         /* reset slave->backup and slave->inactive */
3149         if (bond_has_slaves(bond)) {
3150                 bond_for_each_slave(bond, slave, iter) {
3151                         if (bond_uses_primary(bond) &&
3152                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3153                                 bond_set_slave_inactive_flags(slave,
3154                                                               BOND_SLAVE_NOTIFY_NOW);
3155                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3156                                 bond_set_slave_active_flags(slave,
3157                                                             BOND_SLAVE_NOTIFY_NOW);
3158                         }
3159                 }
3160         }
3161
3162         bond_work_init_all(bond);
3163
3164         if (bond_is_lb(bond)) {
3165                 /* bond_alb_initialize must be called before the timer
3166                  * is started.
3167                  */
3168                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3169                         return -ENOMEM;
3170                 if (bond->params.tlb_dynamic_lb)
3171                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3172         }
3173
3174         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3175                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3176
3177         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3178                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3179                 bond->recv_probe = bond_arp_rcv;
3180         }
3181
3182         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3183                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3184                 /* register to receive LACPDUs */
3185                 bond->recv_probe = bond_3ad_lacpdu_recv;
3186                 bond_3ad_initiate_agg_selection(bond, 1);
3187         }
3188
3189         if (bond_mode_uses_xmit_hash(bond))
3190                 bond_update_slave_arr(bond, NULL);
3191
3192         return 0;
3193 }
3194
3195 static int bond_close(struct net_device *bond_dev)
3196 {
3197         struct bonding *bond = netdev_priv(bond_dev);
3198
3199         bond_work_cancel_all(bond);
3200         bond->send_peer_notif = 0;
3201         if (bond_is_lb(bond))
3202                 bond_alb_deinitialize(bond);
3203         bond->recv_probe = NULL;
3204
3205         return 0;
3206 }
3207
3208 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3209                                                 struct rtnl_link_stats64 *stats)
3210 {
3211         struct bonding *bond = netdev_priv(bond_dev);
3212         struct rtnl_link_stats64 temp;
3213         struct list_head *iter;
3214         struct slave *slave;
3215
3216         memcpy(stats, &bond->bond_stats, sizeof(*stats));
3217
3218         bond_for_each_slave(bond, slave, iter) {
3219                 const struct rtnl_link_stats64 *sstats =
3220                         dev_get_stats(slave->dev, &temp);
3221                 struct rtnl_link_stats64 *pstats = &slave->slave_stats;
3222
3223                 stats->rx_packets +=  sstats->rx_packets - pstats->rx_packets;
3224                 stats->rx_bytes += sstats->rx_bytes - pstats->rx_bytes;
3225                 stats->rx_errors += sstats->rx_errors - pstats->rx_errors;
3226                 stats->rx_dropped += sstats->rx_dropped - pstats->rx_dropped;
3227
3228                 stats->tx_packets += sstats->tx_packets - pstats->tx_packets;;
3229                 stats->tx_bytes += sstats->tx_bytes - pstats->tx_bytes;
3230                 stats->tx_errors += sstats->tx_errors - pstats->tx_errors;
3231                 stats->tx_dropped += sstats->tx_dropped - pstats->tx_dropped;
3232
3233                 stats->multicast += sstats->multicast - pstats->multicast;
3234                 stats->collisions += sstats->collisions - pstats->collisions;
3235
3236                 stats->rx_length_errors += sstats->rx_length_errors - pstats->rx_length_errors;
3237                 stats->rx_over_errors += sstats->rx_over_errors - pstats->rx_over_errors;
3238                 stats->rx_crc_errors += sstats->rx_crc_errors - pstats->rx_crc_errors;
3239                 stats->rx_frame_errors += sstats->rx_frame_errors - pstats->rx_frame_errors;
3240                 stats->rx_fifo_errors += sstats->rx_fifo_errors - pstats->rx_fifo_errors;
3241                 stats->rx_missed_errors += sstats->rx_missed_errors - pstats->rx_missed_errors;
3242
3243                 stats->tx_aborted_errors += sstats->tx_aborted_errors - pstats->tx_aborted_errors;
3244                 stats->tx_carrier_errors += sstats->tx_carrier_errors - pstats->tx_carrier_errors;
3245                 stats->tx_fifo_errors += sstats->tx_fifo_errors - pstats->tx_fifo_errors;
3246                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors - pstats->tx_heartbeat_errors;
3247                 stats->tx_window_errors += sstats->tx_window_errors - pstats->tx_window_errors;
3248
3249                 /* save off the slave stats for the next run */
3250                 memcpy(pstats, sstats, sizeof(*sstats));
3251         }
3252         memcpy(&bond->bond_stats, stats, sizeof(*stats));
3253
3254         return stats;
3255 }
3256
3257 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3258 {
3259         struct bonding *bond = netdev_priv(bond_dev);
3260         struct net_device *slave_dev = NULL;
3261         struct ifbond k_binfo;
3262         struct ifbond __user *u_binfo = NULL;
3263         struct ifslave k_sinfo;
3264         struct ifslave __user *u_sinfo = NULL;
3265         struct mii_ioctl_data *mii = NULL;
3266         struct bond_opt_value newval;
3267         struct net *net;
3268         int res = 0;
3269
3270         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3271
3272         switch (cmd) {
3273         case SIOCGMIIPHY:
3274                 mii = if_mii(ifr);
3275                 if (!mii)
3276                         return -EINVAL;
3277
3278                 mii->phy_id = 0;
3279                 /* Fall Through */
3280         case SIOCGMIIREG:
3281                 /* We do this again just in case we were called by SIOCGMIIREG
3282                  * instead of SIOCGMIIPHY.
3283                  */
3284                 mii = if_mii(ifr);
3285                 if (!mii)
3286                         return -EINVAL;
3287
3288                 if (mii->reg_num == 1) {
3289                         mii->val_out = 0;
3290                         if (netif_carrier_ok(bond->dev))
3291                                 mii->val_out = BMSR_LSTATUS;
3292                 }
3293
3294                 return 0;
3295         case BOND_INFO_QUERY_OLD:
3296         case SIOCBONDINFOQUERY:
3297                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3298
3299                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3300                         return -EFAULT;
3301
3302                 res = bond_info_query(bond_dev, &k_binfo);
3303                 if (res == 0 &&
3304                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3305                         return -EFAULT;
3306
3307                 return res;
3308         case BOND_SLAVE_INFO_QUERY_OLD:
3309         case SIOCBONDSLAVEINFOQUERY:
3310                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3311
3312                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3313                         return -EFAULT;
3314
3315                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3316                 if (res == 0 &&
3317                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3318                         return -EFAULT;
3319
3320                 return res;
3321         default:
3322                 break;
3323         }
3324
3325         net = dev_net(bond_dev);
3326
3327         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3328                 return -EPERM;
3329
3330         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3331
3332         netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3333
3334         if (!slave_dev)
3335                 return -ENODEV;
3336
3337         netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3338         switch (cmd) {
3339         case BOND_ENSLAVE_OLD:
3340         case SIOCBONDENSLAVE:
3341                 res = bond_enslave(bond_dev, slave_dev);
3342                 break;
3343         case BOND_RELEASE_OLD:
3344         case SIOCBONDRELEASE:
3345                 res = bond_release(bond_dev, slave_dev);
3346                 break;
3347         case BOND_SETHWADDR_OLD:
3348         case SIOCBONDSETHWADDR:
3349                 bond_set_dev_addr(bond_dev, slave_dev);
3350                 res = 0;
3351                 break;
3352         case BOND_CHANGE_ACTIVE_OLD:
3353         case SIOCBONDCHANGEACTIVE:
3354                 bond_opt_initstr(&newval, slave_dev->name);
3355                 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3356                 break;
3357         default:
3358                 res = -EOPNOTSUPP;
3359         }
3360
3361         return res;
3362 }
3363
3364 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3365 {
3366         struct bonding *bond = netdev_priv(bond_dev);
3367
3368         if (change & IFF_PROMISC)
3369                 bond_set_promiscuity(bond,
3370                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3371
3372         if (change & IFF_ALLMULTI)
3373                 bond_set_allmulti(bond,
3374                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3375 }
3376
3377 static void bond_set_rx_mode(struct net_device *bond_dev)
3378 {
3379         struct bonding *bond = netdev_priv(bond_dev);
3380         struct list_head *iter;
3381         struct slave *slave;
3382
3383         rcu_read_lock();
3384         if (bond_uses_primary(bond)) {
3385                 slave = rcu_dereference(bond->curr_active_slave);
3386                 if (slave) {
3387                         dev_uc_sync(slave->dev, bond_dev);
3388                         dev_mc_sync(slave->dev, bond_dev);
3389                 }
3390         } else {
3391                 bond_for_each_slave_rcu(bond, slave, iter) {
3392                         dev_uc_sync_multiple(slave->dev, bond_dev);
3393                         dev_mc_sync_multiple(slave->dev, bond_dev);
3394                 }
3395         }
3396         rcu_read_unlock();
3397 }
3398
3399 static int bond_neigh_init(struct neighbour *n)
3400 {
3401         struct bonding *bond = netdev_priv(n->dev);
3402         const struct net_device_ops *slave_ops;
3403         struct neigh_parms parms;
3404         struct slave *slave;
3405         int ret;
3406
3407         slave = bond_first_slave(bond);
3408         if (!slave)
3409                 return 0;
3410         slave_ops = slave->dev->netdev_ops;
3411         if (!slave_ops->ndo_neigh_setup)
3412                 return 0;
3413
3414         parms.neigh_setup = NULL;
3415         parms.neigh_cleanup = NULL;
3416         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3417         if (ret)
3418                 return ret;
3419
3420         /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3421          * after the last slave has been detached.  Assumes that all slaves
3422          * utilize the same neigh_cleanup (true at this writing as only user
3423          * is ipoib).
3424          */
3425         n->parms->neigh_cleanup = parms.neigh_cleanup;
3426
3427         if (!parms.neigh_setup)
3428                 return 0;
3429
3430         return parms.neigh_setup(n);
3431 }
3432
3433 /* The bonding ndo_neigh_setup is called at init time beofre any
3434  * slave exists. So we must declare proxy setup function which will
3435  * be used at run time to resolve the actual slave neigh param setup.
3436  *
3437  * It's also called by master devices (such as vlans) to setup their
3438  * underlying devices. In that case - do nothing, we're already set up from
3439  * our init.
3440  */
3441 static int bond_neigh_setup(struct net_device *dev,
3442                             struct neigh_parms *parms)
3443 {
3444         /* modify only our neigh_parms */
3445         if (parms->dev == dev)
3446                 parms->neigh_setup = bond_neigh_init;
3447
3448         return 0;
3449 }
3450
3451 /* Change the MTU of all of a master's slaves to match the master */
3452 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3453 {
3454         struct bonding *bond = netdev_priv(bond_dev);
3455         struct slave *slave, *rollback_slave;
3456         struct list_head *iter;
3457         int res = 0;
3458
3459         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3460
3461         bond_for_each_slave(bond, slave, iter) {
3462                 netdev_dbg(bond_dev, "s %p c_m %p\n",
3463                            slave, slave->dev->netdev_ops->ndo_change_mtu);
3464
3465                 res = dev_set_mtu(slave->dev, new_mtu);
3466
3467                 if (res) {
3468                         /* If we failed to set the slave's mtu to the new value
3469                          * we must abort the operation even in ACTIVE_BACKUP
3470                          * mode, because if we allow the backup slaves to have
3471                          * different mtu values than the active slave we'll
3472                          * need to change their mtu when doing a failover. That
3473                          * means changing their mtu from timer context, which
3474                          * is probably not a good idea.
3475                          */
3476                         netdev_dbg(bond_dev, "err %d %s\n", res,
3477                                    slave->dev->name);
3478                         goto unwind;
3479                 }
3480         }
3481
3482         bond_dev->mtu = new_mtu;
3483
3484         return 0;
3485
3486 unwind:
3487         /* unwind from head to the slave that failed */
3488         bond_for_each_slave(bond, rollback_slave, iter) {
3489                 int tmp_res;
3490
3491                 if (rollback_slave == slave)
3492                         break;
3493
3494                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3495                 if (tmp_res) {
3496                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3497                                    tmp_res, rollback_slave->dev->name);
3498                 }
3499         }
3500
3501         return res;
3502 }
3503
3504 /* Change HW address
3505  *
3506  * Note that many devices must be down to change the HW address, and
3507  * downing the master releases all slaves.  We can make bonds full of
3508  * bonding devices to test this, however.
3509  */
3510 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3511 {
3512         struct bonding *bond = netdev_priv(bond_dev);
3513         struct slave *slave, *rollback_slave;
3514         struct sockaddr *sa = addr, tmp_sa;
3515         struct list_head *iter;
3516         int res = 0;
3517
3518         if (BOND_MODE(bond) == BOND_MODE_ALB)
3519                 return bond_alb_set_mac_address(bond_dev, addr);
3520
3521
3522         netdev_dbg(bond_dev, "bond=%p\n", bond);
3523
3524         /* If fail_over_mac is enabled, do nothing and return success.
3525          * Returning an error causes ifenslave to fail.
3526          */
3527         if (bond->params.fail_over_mac &&
3528             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3529                 return 0;
3530
3531         if (!is_valid_ether_addr(sa->sa_data))
3532                 return -EADDRNOTAVAIL;
3533
3534         bond_for_each_slave(bond, slave, iter) {
3535                 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3536                 res = dev_set_mac_address(slave->dev, addr);
3537                 if (res) {
3538                         /* TODO: consider downing the slave
3539                          * and retry ?
3540                          * User should expect communications
3541                          * breakage anyway until ARP finish
3542                          * updating, so...
3543                          */
3544                         netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3545                         goto unwind;
3546                 }
3547         }
3548
3549         /* success */
3550         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3551         return 0;
3552
3553 unwind:
3554         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3555         tmp_sa.sa_family = bond_dev->type;
3556
3557         /* unwind from head to the slave that failed */
3558         bond_for_each_slave(bond, rollback_slave, iter) {
3559                 int tmp_res;
3560
3561                 if (rollback_slave == slave)
3562                         break;
3563
3564                 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3565                 if (tmp_res) {
3566                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3567                                    tmp_res, rollback_slave->dev->name);
3568                 }
3569         }
3570
3571         return res;
3572 }
3573
3574 /**
3575  * bond_xmit_slave_id - transmit skb through slave with slave_id
3576  * @bond: bonding device that is transmitting
3577  * @skb: buffer to transmit
3578  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3579  *
3580  * This function tries to transmit through slave with slave_id but in case
3581  * it fails, it tries to find the first available slave for transmission.
3582  * The skb is consumed in all cases, thus the function is void.
3583  */
3584 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3585 {
3586         struct list_head *iter;
3587         struct slave *slave;
3588         int i = slave_id;
3589
3590         /* Here we start from the slave with slave_id */
3591         bond_for_each_slave_rcu(bond, slave, iter) {
3592                 if (--i < 0) {
3593                         if (bond_slave_can_tx(slave)) {
3594                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3595                                 return;
3596                         }
3597                 }
3598         }
3599
3600         /* Here we start from the first slave up to slave_id */
3601         i = slave_id;
3602         bond_for_each_slave_rcu(bond, slave, iter) {
3603                 if (--i < 0)
3604                         break;
3605                 if (bond_slave_can_tx(slave)) {
3606                         bond_dev_queue_xmit(bond, skb, slave->dev);
3607                         return;
3608                 }
3609         }
3610         /* no slave that can tx has been found */
3611         bond_tx_drop(bond->dev, skb);
3612 }
3613
3614 /**
3615  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3616  * @bond: bonding device to use
3617  *
3618  * Based on the value of the bonding device's packets_per_slave parameter
3619  * this function generates a slave id, which is usually used as the next
3620  * slave to transmit through.
3621  */
3622 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3623 {
3624         u32 slave_id;
3625         struct reciprocal_value reciprocal_packets_per_slave;
3626         int packets_per_slave = bond->params.packets_per_slave;
3627
3628         switch (packets_per_slave) {
3629         case 0:
3630                 slave_id = prandom_u32();
3631                 break;
3632         case 1:
3633                 slave_id = bond->rr_tx_counter;
3634                 break;
3635         default:
3636                 reciprocal_packets_per_slave =
3637                         bond->params.reciprocal_packets_per_slave;
3638                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3639                                              reciprocal_packets_per_slave);
3640                 break;
3641         }
3642         bond->rr_tx_counter++;
3643
3644         return slave_id;
3645 }
3646
3647 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3648 {
3649         struct bonding *bond = netdev_priv(bond_dev);
3650         struct iphdr *iph = ip_hdr(skb);
3651         struct slave *slave;
3652         u32 slave_id;
3653
3654         /* Start with the curr_active_slave that joined the bond as the
3655          * default for sending IGMP traffic.  For failover purposes one
3656          * needs to maintain some consistency for the interface that will
3657          * send the join/membership reports.  The curr_active_slave found
3658          * will send all of this type of traffic.
3659          */
3660         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3661                 slave = rcu_dereference(bond->curr_active_slave);
3662                 if (slave)
3663                         bond_dev_queue_xmit(bond, skb, slave->dev);
3664                 else
3665                         bond_xmit_slave_id(bond, skb, 0);
3666         } else {
3667                 int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3668
3669                 if (likely(slave_cnt)) {
3670                         slave_id = bond_rr_gen_slave_id(bond);
3671                         bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3672                 } else {
3673                         bond_tx_drop(bond_dev, skb);
3674                 }
3675         }
3676
3677         return NETDEV_TX_OK;
3678 }
3679
3680 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3681  * the bond has a usable interface.
3682  */
3683 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3684 {
3685         struct bonding *bond = netdev_priv(bond_dev);
3686         struct slave *slave;
3687
3688         slave = rcu_dereference(bond->curr_active_slave);
3689         if (slave)
3690                 bond_dev_queue_xmit(bond, skb, slave->dev);
3691         else
3692                 bond_tx_drop(bond_dev, skb);
3693
3694         return NETDEV_TX_OK;
3695 }
3696
3697 /* Use this to update slave_array when (a) it's not appropriate to update
3698  * slave_array right away (note that update_slave_array() may sleep)
3699  * and / or (b) RTNL is not held.
3700  */
3701 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3702 {
3703         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3704 }
3705
3706 /* Slave array work handler. Holds only RTNL */
3707 static void bond_slave_arr_handler(struct work_struct *work)
3708 {
3709         struct bonding *bond = container_of(work, struct bonding,
3710                                             slave_arr_work.work);
3711         int ret;
3712
3713         if (!rtnl_trylock())
3714                 goto err;
3715
3716         ret = bond_update_slave_arr(bond, NULL);
3717         rtnl_unlock();
3718         if (ret) {
3719                 pr_warn_ratelimited("Failed to update slave array from WT\n");
3720                 goto err;
3721         }
3722         return;
3723
3724 err:
3725         bond_slave_arr_work_rearm(bond, 1);
3726 }
3727
3728 /* Build the usable slaves array in control path for modes that use xmit-hash
3729  * to determine the slave interface -
3730  * (a) BOND_MODE_8023AD
3731  * (b) BOND_MODE_XOR
3732  * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3733  *
3734  * The caller is expected to hold RTNL only and NO other lock!
3735  */
3736 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3737 {
3738         struct slave *slave;
3739         struct list_head *iter;
3740         struct bond_up_slave *new_arr, *old_arr;
3741         int slaves_in_agg;
3742         int agg_id = 0;
3743         int ret = 0;
3744
3745 #ifdef CONFIG_LOCKDEP
3746         WARN_ON(lockdep_is_held(&bond->mode_lock));
3747 #endif
3748
3749         new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3750                           GFP_KERNEL);
3751         if (!new_arr) {
3752                 ret = -ENOMEM;
3753                 pr_err("Failed to build slave-array.\n");
3754                 goto out;
3755         }
3756         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3757                 struct ad_info ad_info;
3758
3759                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3760                         pr_debug("bond_3ad_get_active_agg_info failed\n");
3761                         kfree_rcu(new_arr, rcu);
3762                         /* No active aggragator means it's not safe to use
3763                          * the previous array.
3764                          */
3765                         old_arr = rtnl_dereference(bond->slave_arr);
3766                         if (old_arr) {
3767                                 RCU_INIT_POINTER(bond->slave_arr, NULL);
3768                                 kfree_rcu(old_arr, rcu);
3769                         }
3770                         goto out;
3771                 }
3772                 slaves_in_agg = ad_info.ports;
3773                 agg_id = ad_info.aggregator_id;
3774         }
3775         bond_for_each_slave(bond, slave, iter) {
3776                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3777                         struct aggregator *agg;
3778
3779                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
3780                         if (!agg || agg->aggregator_identifier != agg_id)
3781                                 continue;
3782                 }
3783                 if (!bond_slave_can_tx(slave))
3784                         continue;
3785                 if (skipslave == slave)
3786                         continue;
3787                 new_arr->arr[new_arr->count++] = slave;
3788         }
3789
3790         old_arr = rtnl_dereference(bond->slave_arr);
3791         rcu_assign_pointer(bond->slave_arr, new_arr);
3792         if (old_arr)
3793                 kfree_rcu(old_arr, rcu);
3794 out:
3795         if (ret != 0 && skipslave) {
3796                 int idx;
3797
3798                 /* Rare situation where caller has asked to skip a specific
3799                  * slave but allocation failed (most likely!). BTW this is
3800                  * only possible when the call is initiated from
3801                  * __bond_release_one(). In this situation; overwrite the
3802                  * skipslave entry in the array with the last entry from the
3803                  * array to avoid a situation where the xmit path may choose
3804                  * this to-be-skipped slave to send a packet out.
3805                  */
3806                 old_arr = rtnl_dereference(bond->slave_arr);
3807                 for (idx = 0; idx < old_arr->count; idx++) {
3808                         if (skipslave == old_arr->arr[idx]) {
3809                                 old_arr->arr[idx] =
3810                                     old_arr->arr[old_arr->count-1];
3811                                 old_arr->count--;
3812                                 break;
3813                         }
3814                 }
3815         }
3816         return ret;
3817 }
3818
3819 /* Use this Xmit function for 3AD as well as XOR modes. The current
3820  * usable slave array is formed in the control path. The xmit function
3821  * just calculates hash and sends the packet out.
3822  */
3823 static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3824 {
3825         struct bonding *bond = netdev_priv(dev);
3826         struct slave *slave;
3827         struct bond_up_slave *slaves;
3828         unsigned int count;
3829
3830         slaves = rcu_dereference(bond->slave_arr);
3831         count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3832         if (likely(count)) {
3833                 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3834                 bond_dev_queue_xmit(bond, skb, slave->dev);
3835         } else {
3836                 bond_tx_drop(dev, skb);
3837         }
3838
3839         return NETDEV_TX_OK;
3840 }
3841
3842 /* in broadcast mode, we send everything to all usable interfaces. */
3843 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3844 {
3845         struct bonding *bond = netdev_priv(bond_dev);
3846         struct slave *slave = NULL;
3847         struct list_head *iter;
3848
3849         bond_for_each_slave_rcu(bond, slave, iter) {
3850                 if (bond_is_last_slave(bond, slave))
3851                         break;
3852                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3853                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3854
3855                         if (!skb2) {
3856                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3857                                                     bond_dev->name, __func__);
3858                                 continue;
3859                         }
3860                         bond_dev_queue_xmit(bond, skb2, slave->dev);
3861                 }
3862         }
3863         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3864                 bond_dev_queue_xmit(bond, skb, slave->dev);
3865         else
3866                 bond_tx_drop(bond_dev, skb);
3867
3868         return NETDEV_TX_OK;
3869 }
3870
3871 /*------------------------- Device initialization ---------------------------*/
3872
3873 /* Lookup the slave that corresponds to a qid */
3874 static inline int bond_slave_override(struct bonding *bond,
3875                                       struct sk_buff *skb)
3876 {
3877         struct slave *slave = NULL;
3878         struct list_head *iter;
3879
3880         if (!skb->queue_mapping)
3881                 return 1;
3882
3883         /* Find out if any slaves have the same mapping as this skb. */
3884         bond_for_each_slave_rcu(bond, slave, iter) {
3885                 if (slave->queue_id == skb->queue_mapping) {
3886                         if (bond_slave_is_up(slave) &&
3887                             slave->link == BOND_LINK_UP) {
3888                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3889                                 return 0;
3890                         }
3891                         /* If the slave isn't UP, use default transmit policy. */
3892                         break;
3893                 }
3894         }
3895
3896         return 1;
3897 }
3898
3899
3900 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3901                              void *accel_priv, select_queue_fallback_t fallback)
3902 {
3903         /* This helper function exists to help dev_pick_tx get the correct
3904          * destination queue.  Using a helper function skips a call to
3905          * skb_tx_hash and will put the skbs in the queue we expect on their
3906          * way down to the bonding driver.
3907          */
3908         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3909
3910         /* Save the original txq to restore before passing to the driver */
3911         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3912
3913         if (unlikely(txq >= dev->real_num_tx_queues)) {
3914                 do {
3915                         txq -= dev->real_num_tx_queues;
3916                 } while (txq >= dev->real_num_tx_queues);
3917         }
3918         return txq;
3919 }
3920
3921 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3922 {
3923         struct bonding *bond = netdev_priv(dev);
3924
3925         if (bond_should_override_tx_queue(bond) &&
3926             !bond_slave_override(bond, skb))
3927                 return NETDEV_TX_OK;
3928
3929         switch (BOND_MODE(bond)) {
3930         case BOND_MODE_ROUNDROBIN:
3931                 return bond_xmit_roundrobin(skb, dev);
3932         case BOND_MODE_ACTIVEBACKUP:
3933                 return bond_xmit_activebackup(skb, dev);
3934         case BOND_MODE_8023AD:
3935         case BOND_MODE_XOR:
3936                 return bond_3ad_xor_xmit(skb, dev);
3937         case BOND_MODE_BROADCAST:
3938                 return bond_xmit_broadcast(skb, dev);
3939         case BOND_MODE_ALB:
3940                 return bond_alb_xmit(skb, dev);
3941         case BOND_MODE_TLB:
3942                 return bond_tlb_xmit(skb, dev);
3943         default:
3944                 /* Should never happen, mode already checked */
3945                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
3946                 WARN_ON_ONCE(1);
3947                 bond_tx_drop(dev, skb);
3948                 return NETDEV_TX_OK;
3949         }
3950 }
3951
3952 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3953 {
3954         struct bonding *bond = netdev_priv(dev);
3955         netdev_tx_t ret = NETDEV_TX_OK;
3956
3957         /* If we risk deadlock from transmitting this in the
3958          * netpoll path, tell netpoll to queue the frame for later tx
3959          */
3960         if (unlikely(is_netpoll_tx_blocked(dev)))
3961                 return NETDEV_TX_BUSY;
3962
3963         rcu_read_lock();
3964         if (bond_has_slaves(bond))
3965                 ret = __bond_start_xmit(skb, dev);
3966         else
3967                 bond_tx_drop(dev, skb);
3968         rcu_read_unlock();
3969
3970         return ret;
3971 }
3972
3973 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3974                                      struct ethtool_cmd *ecmd)
3975 {
3976         struct bonding *bond = netdev_priv(bond_dev);
3977         unsigned long speed = 0;
3978         struct list_head *iter;
3979         struct slave *slave;
3980
3981         ecmd->duplex = DUPLEX_UNKNOWN;
3982         ecmd->port = PORT_OTHER;
3983
3984         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
3985          * do not need to check mode.  Though link speed might not represent
3986          * the true receive or transmit bandwidth (not all modes are symmetric)
3987          * this is an accurate maximum.
3988          */
3989         bond_for_each_slave(bond, slave, iter) {
3990                 if (bond_slave_can_tx(slave)) {
3991                         if (slave->speed != SPEED_UNKNOWN)
3992                                 speed += slave->speed;
3993                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
3994                             slave->duplex != DUPLEX_UNKNOWN)
3995                                 ecmd->duplex = slave->duplex;
3996                 }
3997         }
3998         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3999
4000         return 0;
4001 }
4002
4003 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4004                                      struct ethtool_drvinfo *drvinfo)
4005 {
4006         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4007         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4008         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4009                  BOND_ABI_VERSION);
4010 }
4011
4012 static const struct ethtool_ops bond_ethtool_ops = {
4013         .get_drvinfo            = bond_ethtool_get_drvinfo,
4014         .get_settings           = bond_ethtool_get_settings,
4015         .get_link               = ethtool_op_get_link,
4016 };
4017
4018 static const struct net_device_ops bond_netdev_ops = {
4019         .ndo_init               = bond_init,
4020         .ndo_uninit             = bond_uninit,
4021         .ndo_open               = bond_open,
4022         .ndo_stop               = bond_close,
4023         .ndo_start_xmit         = bond_start_xmit,
4024         .ndo_select_queue       = bond_select_queue,
4025         .ndo_get_stats64        = bond_get_stats,
4026         .ndo_do_ioctl           = bond_do_ioctl,
4027         .ndo_change_rx_flags    = bond_change_rx_flags,
4028         .ndo_set_rx_mode        = bond_set_rx_mode,
4029         .ndo_change_mtu         = bond_change_mtu,
4030         .ndo_set_mac_address    = bond_set_mac_address,
4031         .ndo_neigh_setup        = bond_neigh_setup,
4032         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4033         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4034 #ifdef CONFIG_NET_POLL_CONTROLLER
4035         .ndo_netpoll_setup      = bond_netpoll_setup,
4036         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4037         .ndo_poll_controller    = bond_poll_controller,
4038 #endif
4039         .ndo_add_slave          = bond_enslave,
4040         .ndo_del_slave          = bond_release,
4041         .ndo_fix_features       = bond_fix_features,
4042         .ndo_bridge_setlink     = ndo_dflt_netdev_switch_port_bridge_setlink,
4043         .ndo_bridge_dellink     = ndo_dflt_netdev_switch_port_bridge_dellink,
4044         .ndo_features_check     = passthru_features_check,
4045 };
4046
4047 static const struct device_type bond_type = {
4048         .name = "bond",
4049 };
4050
4051 static void bond_destructor(struct net_device *bond_dev)
4052 {
4053         struct bonding *bond = netdev_priv(bond_dev);
4054         if (bond->wq)
4055                 destroy_workqueue(bond->wq);
4056         free_netdev(bond_dev);
4057 }
4058
4059 void bond_setup(struct net_device *bond_dev)
4060 {
4061         struct bonding *bond = netdev_priv(bond_dev);
4062
4063         spin_lock_init(&bond->mode_lock);
4064         bond->params = bonding_defaults;
4065
4066         /* Initialize pointers */
4067         bond->dev = bond_dev;
4068
4069         /* Initialize the device entry points */
4070         ether_setup(bond_dev);
4071         bond_dev->netdev_ops = &bond_netdev_ops;
4072         bond_dev->ethtool_ops = &bond_ethtool_ops;
4073
4074         bond_dev->destructor = bond_destructor;
4075
4076         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4077
4078         /* Initialize the device options */
4079         bond_dev->tx_queue_len = 0;
4080         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4081         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT;
4082         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4083
4084         /* don't acquire bond device's netif_tx_lock when transmitting */
4085         bond_dev->features |= NETIF_F_LLTX;
4086
4087         /* By default, we declare the bond to be fully
4088          * VLAN hardware accelerated capable. Special
4089          * care is taken in the various xmit functions
4090          * when there are slaves that are not hw accel
4091          * capable
4092          */
4093
4094         /* Don't allow bond devices to change network namespaces. */
4095         bond_dev->features |= NETIF_F_NETNS_LOCAL;
4096
4097         bond_dev->hw_features = BOND_VLAN_FEATURES |
4098                                 NETIF_F_HW_VLAN_CTAG_TX |
4099                                 NETIF_F_HW_VLAN_CTAG_RX |
4100                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4101
4102         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4103         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4104         bond_dev->features |= bond_dev->hw_features;
4105 }
4106
4107 /* Destroy a bonding device.
4108  * Must be under rtnl_lock when this function is called.
4109  */
4110 static void bond_uninit(struct net_device *bond_dev)
4111 {
4112         struct bonding *bond = netdev_priv(bond_dev);
4113         struct list_head *iter;
4114         struct slave *slave;
4115         struct bond_up_slave *arr;
4116
4117         bond_netpoll_cleanup(bond_dev);
4118
4119         /* Release the bonded slaves */
4120         bond_for_each_slave(bond, slave, iter)
4121                 __bond_release_one(bond_dev, slave->dev, true);
4122         netdev_info(bond_dev, "Released all slaves\n");
4123
4124         arr = rtnl_dereference(bond->slave_arr);
4125         if (arr) {
4126                 RCU_INIT_POINTER(bond->slave_arr, NULL);
4127                 kfree_rcu(arr, rcu);
4128         }
4129
4130         list_del(&bond->bond_list);
4131
4132         bond_debug_unregister(bond);
4133 }
4134
4135 /*------------------------- Module initialization ---------------------------*/
4136
4137 static int bond_check_params(struct bond_params *params)
4138 {
4139         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4140         struct bond_opt_value newval;
4141         const struct bond_opt_value *valptr;
4142         int arp_all_targets_value;
4143
4144         /* Convert string parameters. */
4145         if (mode) {
4146                 bond_opt_initstr(&newval, mode);
4147                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4148                 if (!valptr) {
4149                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4150                         return -EINVAL;
4151                 }
4152                 bond_mode = valptr->value;
4153         }
4154
4155         if (xmit_hash_policy) {
4156                 if ((bond_mode != BOND_MODE_XOR) &&
4157                     (bond_mode != BOND_MODE_8023AD) &&
4158                     (bond_mode != BOND_MODE_TLB)) {
4159                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4160                                 bond_mode_name(bond_mode));
4161                 } else {
4162                         bond_opt_initstr(&newval, xmit_hash_policy);
4163                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4164                                                 &newval);
4165                         if (!valptr) {
4166                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4167                                        xmit_hash_policy);
4168                                 return -EINVAL;
4169                         }
4170                         xmit_hashtype = valptr->value;
4171                 }
4172         }
4173
4174         if (lacp_rate) {
4175                 if (bond_mode != BOND_MODE_8023AD) {
4176                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4177                                 bond_mode_name(bond_mode));
4178                 } else {
4179                         bond_opt_initstr(&newval, lacp_rate);
4180                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4181                                                 &newval);
4182                         if (!valptr) {
4183                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4184                                        lacp_rate);
4185                                 return -EINVAL;
4186                         }
4187                         lacp_fast = valptr->value;
4188                 }
4189         }
4190
4191         if (ad_select) {
4192                 bond_opt_initstr(&newval, ad_select);
4193                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4194                                         &newval);
4195                 if (!valptr) {
4196                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4197                         return -EINVAL;
4198                 }
4199                 params->ad_select = valptr->value;
4200                 if (bond_mode != BOND_MODE_8023AD)
4201                         pr_warn("ad_select param only affects 802.3ad mode\n");
4202         } else {
4203                 params->ad_select = BOND_AD_STABLE;
4204         }
4205
4206         if (max_bonds < 0) {
4207                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4208                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4209                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4210         }
4211
4212         if (miimon < 0) {
4213                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4214                         miimon, INT_MAX);
4215                 miimon = 0;
4216         }
4217
4218         if (updelay < 0) {
4219                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4220                         updelay, INT_MAX);
4221                 updelay = 0;
4222         }
4223
4224         if (downdelay < 0) {
4225                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4226                         downdelay, INT_MAX);
4227                 downdelay = 0;
4228         }
4229
4230         if ((use_carrier != 0) && (use_carrier != 1)) {
4231                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4232                         use_carrier);
4233                 use_carrier = 1;
4234         }
4235
4236         if (num_peer_notif < 0 || num_peer_notif > 255) {
4237                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4238                         num_peer_notif);
4239                 num_peer_notif = 1;
4240         }
4241
4242         /* reset values for 802.3ad/TLB/ALB */
4243         if (!bond_mode_uses_arp(bond_mode)) {
4244                 if (!miimon) {
4245                         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");
4246                         pr_warn("Forcing miimon to 100msec\n");
4247                         miimon = BOND_DEFAULT_MIIMON;
4248                 }
4249         }
4250
4251         if (tx_queues < 1 || tx_queues > 255) {
4252                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4253                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4254                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4255         }
4256
4257         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4258                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4259                         all_slaves_active);
4260                 all_slaves_active = 0;
4261         }
4262
4263         if (resend_igmp < 0 || resend_igmp > 255) {
4264                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4265                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4266                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4267         }
4268
4269         bond_opt_initval(&newval, packets_per_slave);
4270         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4271                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4272                         packets_per_slave, USHRT_MAX);
4273                 packets_per_slave = 1;
4274         }
4275
4276         if (bond_mode == BOND_MODE_ALB) {
4277                 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",
4278                           updelay);
4279         }
4280
4281         if (!miimon) {
4282                 if (updelay || downdelay) {
4283                         /* just warn the user the up/down delay will have
4284                          * no effect since miimon is zero...
4285                          */
4286                         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",
4287                                 updelay, downdelay);
4288                 }
4289         } else {
4290                 /* don't allow arp monitoring */
4291                 if (arp_interval) {
4292                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4293                                 miimon, arp_interval);
4294                         arp_interval = 0;
4295                 }
4296
4297                 if ((updelay % miimon) != 0) {
4298                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4299                                 updelay, miimon, (updelay / miimon) * miimon);
4300                 }
4301
4302                 updelay /= miimon;
4303
4304                 if ((downdelay % miimon) != 0) {
4305                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4306                                 downdelay, miimon,
4307                                 (downdelay / miimon) * miimon);
4308                 }
4309
4310                 downdelay /= miimon;
4311         }
4312
4313         if (arp_interval < 0) {
4314                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4315                         arp_interval, INT_MAX);
4316                 arp_interval = 0;
4317         }
4318
4319         for (arp_ip_count = 0, i = 0;
4320              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4321                 __be32 ip;
4322
4323                 /* not a complete check, but good enough to catch mistakes */
4324                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4325                     !bond_is_ip_target_ok(ip)) {
4326                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4327                                 arp_ip_target[i]);
4328                         arp_interval = 0;
4329                 } else {
4330                         if (bond_get_targets_ip(arp_target, ip) == -1)
4331                                 arp_target[arp_ip_count++] = ip;
4332                         else
4333                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4334                                         &ip);
4335                 }
4336         }
4337
4338         if (arp_interval && !arp_ip_count) {
4339                 /* don't allow arping if no arp_ip_target given... */
4340                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4341                         arp_interval);
4342                 arp_interval = 0;
4343         }
4344
4345         if (arp_validate) {
4346                 if (!arp_interval) {
4347                         pr_err("arp_validate requires arp_interval\n");
4348                         return -EINVAL;
4349                 }
4350
4351                 bond_opt_initstr(&newval, arp_validate);
4352                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4353                                         &newval);
4354                 if (!valptr) {
4355                         pr_err("Error: invalid arp_validate \"%s\"\n",
4356                                arp_validate);
4357                         return -EINVAL;
4358                 }
4359                 arp_validate_value = valptr->value;
4360         } else {
4361                 arp_validate_value = 0;
4362         }
4363
4364         arp_all_targets_value = 0;
4365         if (arp_all_targets) {
4366                 bond_opt_initstr(&newval, arp_all_targets);
4367                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4368                                         &newval);
4369                 if (!valptr) {
4370                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4371                                arp_all_targets);
4372                         arp_all_targets_value = 0;
4373                 } else {
4374                         arp_all_targets_value = valptr->value;
4375                 }
4376         }
4377
4378         if (miimon) {
4379                 pr_info("MII link monitoring set to %d ms\n", miimon);
4380         } else if (arp_interval) {
4381                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4382                                           arp_validate_value);
4383                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4384                         arp_interval, valptr->string, arp_ip_count);
4385
4386                 for (i = 0; i < arp_ip_count; i++)
4387                         pr_cont(" %s", arp_ip_target[i]);
4388
4389                 pr_cont("\n");
4390
4391         } else if (max_bonds) {
4392                 /* miimon and arp_interval not set, we need one so things
4393                  * work as expected, see bonding.txt for details
4394                  */
4395                 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");
4396         }
4397
4398         if (primary && !bond_mode_uses_primary(bond_mode)) {
4399                 /* currently, using a primary only makes sense
4400                  * in active backup, TLB or ALB modes
4401                  */
4402                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4403                         primary, bond_mode_name(bond_mode));
4404                 primary = NULL;
4405         }
4406
4407         if (primary && primary_reselect) {
4408                 bond_opt_initstr(&newval, primary_reselect);
4409                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4410                                         &newval);
4411                 if (!valptr) {
4412                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4413                                primary_reselect);
4414                         return -EINVAL;
4415                 }
4416                 primary_reselect_value = valptr->value;
4417         } else {
4418                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4419         }
4420
4421         if (fail_over_mac) {
4422                 bond_opt_initstr(&newval, fail_over_mac);
4423                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4424                                         &newval);
4425                 if (!valptr) {
4426                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4427                                fail_over_mac);
4428                         return -EINVAL;
4429                 }
4430                 fail_over_mac_value = valptr->value;
4431                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4432                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4433         } else {
4434                 fail_over_mac_value = BOND_FOM_NONE;
4435         }
4436
4437         if (lp_interval == 0) {
4438                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4439                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4440                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4441         }
4442
4443         /* fill params struct with the proper values */
4444         params->mode = bond_mode;
4445         params->xmit_policy = xmit_hashtype;
4446         params->miimon = miimon;
4447         params->num_peer_notif = num_peer_notif;
4448         params->arp_interval = arp_interval;
4449         params->arp_validate = arp_validate_value;
4450         params->arp_all_targets = arp_all_targets_value;
4451         params->updelay = updelay;
4452         params->downdelay = downdelay;
4453         params->use_carrier = use_carrier;
4454         params->lacp_fast = lacp_fast;
4455         params->primary[0] = 0;
4456         params->primary_reselect = primary_reselect_value;
4457         params->fail_over_mac = fail_over_mac_value;
4458         params->tx_queues = tx_queues;
4459         params->all_slaves_active = all_slaves_active;
4460         params->resend_igmp = resend_igmp;
4461         params->min_links = min_links;
4462         params->lp_interval = lp_interval;
4463         params->packets_per_slave = packets_per_slave;
4464         params->tlb_dynamic_lb = 1; /* Default value */
4465         if (packets_per_slave > 0) {
4466                 params->reciprocal_packets_per_slave =
4467                         reciprocal_value(packets_per_slave);
4468         } else {
4469                 /* reciprocal_packets_per_slave is unused if
4470                  * packets_per_slave is 0 or 1, just initialize it
4471                  */
4472                 params->reciprocal_packets_per_slave =
4473                         (struct reciprocal_value) { 0 };
4474         }
4475
4476         if (primary) {
4477                 strncpy(params->primary, primary, IFNAMSIZ);
4478                 params->primary[IFNAMSIZ - 1] = 0;
4479         }
4480
4481         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4482
4483         return 0;
4484 }
4485
4486 static struct lock_class_key bonding_netdev_xmit_lock_key;
4487 static struct lock_class_key bonding_netdev_addr_lock_key;
4488 static struct lock_class_key bonding_tx_busylock_key;
4489
4490 static void bond_set_lockdep_class_one(struct net_device *dev,
4491                                        struct netdev_queue *txq,
4492                                        void *_unused)
4493 {
4494         lockdep_set_class(&txq->_xmit_lock,
4495                           &bonding_netdev_xmit_lock_key);
4496 }
4497
4498 static void bond_set_lockdep_class(struct net_device *dev)
4499 {
4500         lockdep_set_class(&dev->addr_list_lock,
4501                           &bonding_netdev_addr_lock_key);
4502         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4503         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4504 }
4505
4506 /* Called from registration process */
4507 static int bond_init(struct net_device *bond_dev)
4508 {
4509         struct bonding *bond = netdev_priv(bond_dev);
4510         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4511
4512         netdev_dbg(bond_dev, "Begin bond_init\n");
4513
4514         bond->wq = create_singlethread_workqueue(bond_dev->name);
4515         if (!bond->wq)
4516                 return -ENOMEM;
4517
4518         bond_set_lockdep_class(bond_dev);
4519
4520         list_add_tail(&bond->bond_list, &bn->dev_list);
4521
4522         bond_prepare_sysfs_group(bond);
4523
4524         bond_debug_register(bond);
4525
4526         /* Ensure valid dev_addr */
4527         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4528             bond_dev->addr_assign_type == NET_ADDR_PERM)
4529                 eth_hw_addr_random(bond_dev);
4530
4531         return 0;
4532 }
4533
4534 unsigned int bond_get_num_tx_queues(void)
4535 {
4536         return tx_queues;
4537 }
4538
4539 /* Create a new bond based on the specified name and bonding parameters.
4540  * If name is NULL, obtain a suitable "bond%d" name for us.
4541  * Caller must NOT hold rtnl_lock; we need to release it here before we
4542  * set up our sysfs entries.
4543  */
4544 int bond_create(struct net *net, const char *name)
4545 {
4546         struct net_device *bond_dev;
4547         struct bonding *bond;
4548         struct alb_bond_info *bond_info;
4549         int res;
4550
4551         rtnl_lock();
4552
4553         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4554                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
4555                                    bond_setup, tx_queues);
4556         if (!bond_dev) {
4557                 pr_err("%s: eek! can't alloc netdev!\n", name);
4558                 rtnl_unlock();
4559                 return -ENOMEM;
4560         }
4561
4562         /*
4563          * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4564          * It is set to 0 by default which is wrong.
4565          */
4566         bond = netdev_priv(bond_dev);
4567         bond_info = &(BOND_ALB_INFO(bond));
4568         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4569
4570         dev_net_set(bond_dev, net);
4571         bond_dev->rtnl_link_ops = &bond_link_ops;
4572
4573         res = register_netdevice(bond_dev);
4574
4575         netif_carrier_off(bond_dev);
4576
4577         rtnl_unlock();
4578         if (res < 0)
4579                 bond_destructor(bond_dev);
4580         return res;
4581 }
4582
4583 static int __net_init bond_net_init(struct net *net)
4584 {
4585         struct bond_net *bn = net_generic(net, bond_net_id);
4586
4587         bn->net = net;
4588         INIT_LIST_HEAD(&bn->dev_list);
4589
4590         bond_create_proc_dir(bn);
4591         bond_create_sysfs(bn);
4592
4593         return 0;
4594 }
4595
4596 static void __net_exit bond_net_exit(struct net *net)
4597 {
4598         struct bond_net *bn = net_generic(net, bond_net_id);
4599         struct bonding *bond, *tmp_bond;
4600         LIST_HEAD(list);
4601
4602         bond_destroy_sysfs(bn);
4603
4604         /* Kill off any bonds created after unregistering bond rtnl ops */
4605         rtnl_lock();
4606         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4607                 unregister_netdevice_queue(bond->dev, &list);
4608         unregister_netdevice_many(&list);
4609         rtnl_unlock();
4610
4611         bond_destroy_proc_dir(bn);
4612 }
4613
4614 static struct pernet_operations bond_net_ops = {
4615         .init = bond_net_init,
4616         .exit = bond_net_exit,
4617         .id   = &bond_net_id,
4618         .size = sizeof(struct bond_net),
4619 };
4620
4621 static int __init bonding_init(void)
4622 {
4623         int i;
4624         int res;
4625
4626         pr_info("%s", bond_version);
4627
4628         res = bond_check_params(&bonding_defaults);
4629         if (res)
4630                 goto out;
4631
4632         res = register_pernet_subsys(&bond_net_ops);
4633         if (res)
4634                 goto out;
4635
4636         res = bond_netlink_init();
4637         if (res)
4638                 goto err_link;
4639
4640         bond_create_debugfs();
4641
4642         for (i = 0; i < max_bonds; i++) {
4643                 res = bond_create(&init_net, NULL);
4644                 if (res)
4645                         goto err;
4646         }
4647
4648         register_netdevice_notifier(&bond_netdev_notifier);
4649 out:
4650         return res;
4651 err:
4652         bond_destroy_debugfs();
4653         bond_netlink_fini();
4654 err_link:
4655         unregister_pernet_subsys(&bond_net_ops);
4656         goto out;
4657
4658 }
4659
4660 static void __exit bonding_exit(void)
4661 {
4662         unregister_netdevice_notifier(&bond_netdev_notifier);
4663
4664         bond_destroy_debugfs();
4665
4666         bond_netlink_fini();
4667         unregister_pernet_subsys(&bond_net_ops);
4668
4669 #ifdef CONFIG_NET_POLL_CONTROLLER
4670         /* Make sure we don't have an imbalance on our netpoll blocking */
4671         WARN_ON(atomic_read(&netpoll_block_tx));
4672 #endif
4673 }
4674
4675 module_init(bonding_init);
4676 module_exit(bonding_exit);
4677 MODULE_LICENSE("GPL");
4678 MODULE_VERSION(DRV_VERSION);
4679 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4680 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");