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 "bonding.h"
  81 #include "bond_3ad.h"
  82 #include "bond_alb.h"
  83
  84 /*---------------------------- Module parameters ----------------------------*/
  85
  86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  87
  88 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  89 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  90 static int num_peer_notif = 1;
  91 static int miimon;
  92 static int updelay;
  93 static int downdelay;
  94 static int use_carrier  = 1;
  95 static char *mode;
  96 static char *primary;
  97 static char *primary_reselect;
  98 static char *lacp_rate;
  99 static int min_links;
 100 static char *ad_select;
 101 static char *xmit_hash_policy;
 102 static int arp_interval;
 103 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 104 static char *arp_validate;
 105 static char *arp_all_targets;
 106 static char *fail_over_mac;
 107 static int all_slaves_active;
 108 static struct bond_params bonding_defaults;
 109 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 110 static int packets_per_slave = 1;
 111 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
 112
 113 module_param(max_bonds, int, 0);
 114 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 115 module_param(tx_queues, int, 0);
 116 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 117 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 118 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 119                                "failover event (alias of num_unsol_na)");
 120 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 121 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 122                                "failover event (alias of num_grat_arp)");
 123 module_param(miimon, int, 0);
 124 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 125 module_param(updelay, int, 0);
 126 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 127 module_param(downdelay, int, 0);
 128 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 129                             "in milliseconds");
 130 module_param(use_carrier, int, 0);
 131 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 132                               "0 for off, 1 for on (default)");
 133 module_param(mode, charp, 0);
 134 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 135                        "1 for active-backup, 2 for balance-xor, "
 136                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 137                        "6 for balance-alb");
 138 module_param(primary, charp, 0);
 139 MODULE_PARM_DESC(primary, "Primary network device to use");
 140 module_param(primary_reselect, charp, 0);
 141 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 142                                    "once it comes up; "
 143                                    "0 for always (default), "
 144                                    "1 for only if speed of primary is "
 145                                    "better, "
 146                                    "2 for only on active slave "
 147                                    "failure");
 148 module_param(lacp_rate, charp, 0);
 149 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 150                             "0 for slow, 1 for fast");
 151 module_param(ad_select, charp, 0);
 152 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 153                             "0 for stable (default), 1 for bandwidth, "
 154                             "2 for count");
 155 module_param(min_links, int, 0);
 156 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 157
 158 module_param(xmit_hash_policy, charp, 0);
 159 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 160                                    "0 for layer 2 (default), 1 for layer 3+4, "
 161                                    "2 for layer 2+3, 3 for encap layer 2+3, "
 162                                    "4 for encap layer 3+4");
 163 module_param(arp_interval, int, 0);
 164 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 165 module_param_array(arp_ip_target, charp, NULL, 0);
 166 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 167 module_param(arp_validate, charp, 0);
 168 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 169                                "0 for none (default), 1 for active, "
 170                                "2 for backup, 3 for all");
 171 module_param(arp_all_targets, charp, 0);
 172 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 173 module_param(fail_over_mac, charp, 0);
 174 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 175                                 "the same MAC; 0 for none (default), "
 176                                 "1 for active, 2 for follow");
 177 module_param(all_slaves_active, int, 0);
 178 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
 179                                      "by setting active flag for all slaves; "
 180                                      "0 for never (default), 1 for always.");
 181 module_param(resend_igmp, int, 0);
 182 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 183                               "link failure");
 184 module_param(packets_per_slave, int, 0);
 185 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
 186                                     "mode; 0 for a random slave, 1 packet per "
 187                                     "slave (default), >1 packets per slave.");
 188 module_param(lp_interval, uint, 0);
 189 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
 190                               "the bonding driver sends learning packets to "
 191                               "each slaves peer switch. The default is 1.");
 192
 193 /*----------------------------- Global variables ----------------------------*/
 194
 195 #ifdef CONFIG_NET_POLL_CONTROLLER
 196 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 197 #endif
 198
 199 int bond_net_id __read_mostly;
 200
 201 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 202 static int arp_ip_count;
 203 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 204 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 205 static int lacp_fast;
 206
 207 /*-------------------------- Forward declarations ---------------------------*/
 208
 209 static int bond_init(struct net_device *bond_dev);
 210 static void bond_uninit(struct net_device *bond_dev);
 211
 212 /*---------------------------- General routines -----------------------------*/
 213
 214 const char *bond_mode_name(int mode)
 215 {
 216         static const char *names[] = {
 217                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 218                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 219                 [BOND_MODE_XOR] = "load balancing (xor)",
 220                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 221                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 222                 [BOND_MODE_TLB] = "transmit load balancing",
 223                 [BOND_MODE_ALB] = "adaptive load balancing",
 224         };
 225
 226         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 227                 return "unknown";
 228
 229         return names[mode];
 230 }
 231
 232 /*---------------------------------- VLAN -----------------------------------*/
 233
 234 /**
 235  * bond_dev_queue_xmit - Prepare skb for xmit.
 236  *
 237  * @bond: bond device that got this skb for tx.
 238  * @skb: hw accel VLAN tagged skb to transmit
 239  * @slave_dev: slave that is supposed to xmit this skbuff
 240  */
 241 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 242                         struct net_device *slave_dev)
 243 {
 244         skb->dev = slave_dev;
 245
 246         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 247                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 248         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 249
 250         if (unlikely(netpoll_tx_running(bond->dev)))
 251                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 252         else
 253                 dev_queue_xmit(skb);
 254 }
 255
 256 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 257  * We don't protect the slave list iteration with a lock because:
 258  * a. This operation is performed in IOCTL context,
 259  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 260  * c. Holding a lock with BH disabled while directly calling a base driver
 261  *    entry point is generally a BAD idea.
 262  *
 263  * The design of synchronization/protection for this operation in the 8021q
 264  * module is good for one or more VLAN devices over a single physical device
 265  * and cannot be extended for a teaming solution like bonding, so there is a
 266  * potential race condition here where a net device from the vlan group might
 267  * be referenced (either by a base driver or the 8021q code) while it is being
 268  * removed from the system. However, it turns out we're not making matters
 269  * worse, and if it works for regular VLAN usage it will work here too.
 270 */
 271
 272 /**
 273  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 274  * @bond_dev: bonding net device that got called
 275  * @vid: vlan id being added
 276  */
 277 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 278                                 __be16 proto, u16 vid)
 279 {
 280         struct bonding *bond = netdev_priv(bond_dev);
 281         struct slave *slave, *rollback_slave;
 282         struct list_head *iter;
 283         int res;
 284
 285         bond_for_each_slave(bond, slave, iter) {
 286                 res = vlan_vid_add(slave->dev, proto, vid);
 287                 if (res)
 288                         goto unwind;
 289         }
 290
 291         return 0;
 292
 293 unwind:
 294         /* unwind to the slave that failed */
 295         bond_for_each_slave(bond, rollback_slave, iter) {
 296                 if (rollback_slave == slave)
 297                         break;
 298
 299                 vlan_vid_del(rollback_slave->dev, proto, vid);
 300         }
 301
 302         return res;
 303 }
 304
 305 /**
 306  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 307  * @bond_dev: bonding net device that got called
 308  * @vid: vlan id being removed
 309  */
 310 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 311                                  __be16 proto, u16 vid)
 312 {
 313         struct bonding *bond = netdev_priv(bond_dev);
 314         struct list_head *iter;
 315         struct slave *slave;
 316
 317         bond_for_each_slave(bond, slave, iter)
 318                 vlan_vid_del(slave->dev, proto, vid);
 319
 320         if (bond_is_lb(bond))
 321                 bond_alb_clear_vlan(bond, vid);
 322
 323         return 0;
 324 }
 325
 326 /*------------------------------- Link status -------------------------------*/
 327
 328 /* Set the carrier state for the master according to the state of its
 329  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 330  * do special 802.3ad magic.
 331  *
 332  * Returns zero if carrier state does not change, nonzero if it does.
 333  */
 334 static int bond_set_carrier(struct bonding *bond)
 335 {
 336         struct list_head *iter;
 337         struct slave *slave;
 338
 339         if (!bond_has_slaves(bond))
 340                 goto down;
 341
 342         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 343                 return bond_3ad_set_carrier(bond);
 344
 345         bond_for_each_slave(bond, slave, iter) {
 346                 if (slave->link == BOND_LINK_UP) {
 347                         if (!netif_carrier_ok(bond->dev)) {
 348                                 netif_carrier_on(bond->dev);
 349                                 return 1;
 350                         }
 351                         return 0;
 352                 }
 353         }
 354
 355 down:
 356         if (netif_carrier_ok(bond->dev)) {
 357                 netif_carrier_off(bond->dev);
 358                 return 1;
 359         }
 360         return 0;
 361 }
 362
 363 /* Get link speed and duplex from the slave's base driver
 364  * using ethtool. If for some reason the call fails or the
 365  * values are invalid, set speed and duplex to -1,
 366  * and return.
 367  */
 368 static void bond_update_speed_duplex(struct slave *slave)
 369 {
 370         struct net_device *slave_dev = slave->dev;
 371         struct ethtool_cmd ecmd;
 372         u32 slave_speed;
 373         int res;
 374
 375         slave->speed = SPEED_UNKNOWN;
 376         slave->duplex = DUPLEX_UNKNOWN;
 377
 378         res = __ethtool_get_settings(slave_dev, &ecmd);
 379         if (res < 0)
 380                 return;
 381
 382         slave_speed = ethtool_cmd_speed(&ecmd);
 383         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 384                 return;
 385
 386         switch (ecmd.duplex) {
 387         case DUPLEX_FULL:
 388         case DUPLEX_HALF:
 389                 break;
 390         default:
 391                 return;
 392         }
 393
 394         slave->speed = slave_speed;
 395         slave->duplex = ecmd.duplex;
 396
 397         return;
 398 }
 399
 400 const char *bond_slave_link_status(s8 link)
 401 {
 402         switch (link) {
 403         case BOND_LINK_UP:
 404                 return "up";
 405         case BOND_LINK_FAIL:
 406                 return "going down";
 407         case BOND_LINK_DOWN:
 408                 return "down";
 409         case BOND_LINK_BACK:
 410                 return "going back";
 411         default:
 412                 return "unknown";
 413         }
 414 }
 415
 416 /* if <dev> supports MII link status reporting, check its link status.
 417  *
 418  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 419  * depending upon the setting of the use_carrier parameter.
 420  *
 421  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 422  * can't tell and just pretend it is), or 0, meaning that the link is
 423  * down.
 424  *
 425  * If reporting is non-zero, instead of faking link up, return -1 if
 426  * both ETHTOOL and MII ioctls fail (meaning the device does not
 427  * support them).  If use_carrier is set, return whatever it says.
 428  * It'd be nice if there was a good way to tell if a driver supports
 429  * netif_carrier, but there really isn't.
 430  */
 431 static int bond_check_dev_link(struct bonding *bond,
 432                                struct net_device *slave_dev, int reporting)
 433 {
 434         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 435         int (*ioctl)(struct net_device *, struct ifreq *, int);
 436         struct ifreq ifr;
 437         struct mii_ioctl_data *mii;
 438
 439         if (!reporting && !netif_running(slave_dev))
 440                 return 0;
 441
 442         if (bond->params.use_carrier)
 443                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 444
 445         /* Try to get link status using Ethtool first. */
 446         if (slave_dev->ethtool_ops->get_link)
 447                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 448                         BMSR_LSTATUS : 0;
 449
 450         /* Ethtool can't be used, fallback to MII ioctls. */
 451         ioctl = slave_ops->ndo_do_ioctl;
 452         if (ioctl) {
 453                 /* TODO: set pointer to correct ioctl on a per team member
 454                  *       bases to make this more efficient. that is, once
 455                  *       we determine the correct ioctl, we will always
 456                  *       call it and not the others for that team
 457                  *       member.
 458                  */
 459
 460                 /* We cannot assume that SIOCGMIIPHY will also read a
 461                  * register; not all network drivers (e.g., e100)
 462                  * support that.
 463                  */
 464
 465                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 466                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 467                 mii = if_mii(&ifr);
 468                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 469                         mii->reg_num = MII_BMSR;
 470                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 471                                 return mii->val_out & BMSR_LSTATUS;
 472                 }
 473         }
 474
 475         /* If reporting, report that either there's no dev->do_ioctl,
 476          * or both SIOCGMIIREG and get_link failed (meaning that we
 477          * cannot report link status).  If not reporting, pretend
 478          * we're ok.
 479          */
 480         return reporting ? -1 : BMSR_LSTATUS;
 481 }
 482
 483 /*----------------------------- Multicast list ------------------------------*/
 484
 485 /* Push the promiscuity flag down to appropriate slaves */
 486 static int bond_set_promiscuity(struct bonding *bond, int inc)
 487 {
 488         struct list_head *iter;
 489         int err = 0;
 490
 491         if (bond_uses_primary(bond)) {
 492                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 493
 494                 if (curr_active)
 495                         err = dev_set_promiscuity(curr_active->dev, inc);
 496         } else {
 497                 struct slave *slave;
 498
 499                 bond_for_each_slave(bond, slave, iter) {
 500                         err = dev_set_promiscuity(slave->dev, inc);
 501                         if (err)
 502                                 return err;
 503                 }
 504         }
 505         return err;
 506 }
 507
 508 /* Push the allmulti flag down to all slaves */
 509 static int bond_set_allmulti(struct bonding *bond, int inc)
 510 {
 511         struct list_head *iter;
 512         int err = 0;
 513
 514         if (bond_uses_primary(bond)) {
 515                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 516
 517                 if (curr_active)
 518                         err = dev_set_allmulti(curr_active->dev, inc);
 519         } else {
 520                 struct slave *slave;
 521
 522                 bond_for_each_slave(bond, slave, iter) {
 523                         err = dev_set_allmulti(slave->dev, inc);
 524                         if (err)
 525                                 return err;
 526                 }
 527         }
 528         return err;
 529 }
 530
 531 /* Retrieve the list of registered multicast addresses for the bonding
 532  * device and retransmit an IGMP JOIN request to the current active
 533  * slave.
 534  */
 535 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 536 {
 537         struct bonding *bond = container_of(work, struct bonding,
 538                                             mcast_work.work);
 539
 540         if (!rtnl_trylock()) {
 541                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 542                 return;
 543         }
 544         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 545
 546         if (bond->igmp_retrans > 1) {
 547                 bond->igmp_retrans--;
 548                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 549         }
 550         rtnl_unlock();
 551 }
 552
 553 /* Flush bond's hardware addresses from slave */
 554 static void bond_hw_addr_flush(struct net_device *bond_dev,
 555                                struct net_device *slave_dev)
 556 {
 557         struct bonding *bond = netdev_priv(bond_dev);
 558
 559         dev_uc_unsync(slave_dev, bond_dev);
 560         dev_mc_unsync(slave_dev, bond_dev);
 561
 562         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 563                 /* del lacpdu mc addr from mc list */
 564                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 565
 566                 dev_mc_del(slave_dev, lacpdu_multicast);
 567         }
 568 }
 569
 570 /*--------------------------- Active slave change ---------------------------*/
 571
 572 /* Update the hardware address list and promisc/allmulti for the new and
 573  * old active slaves (if any).  Modes that are not using primary keep all
 574  * slaves up date at all times; only the modes that use primary need to call
 575  * this function to swap these settings during a failover.
 576  */
 577 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 578                               struct slave *old_active)
 579 {
 580         if (old_active) {
 581                 if (bond->dev->flags & IFF_PROMISC)
 582                         dev_set_promiscuity(old_active->dev, -1);
 583
 584                 if (bond->dev->flags & IFF_ALLMULTI)
 585                         dev_set_allmulti(old_active->dev, -1);
 586
 587                 bond_hw_addr_flush(bond->dev, old_active->dev);
 588         }
 589
 590         if (new_active) {
 591                 /* FIXME: Signal errors upstream. */
 592                 if (bond->dev->flags & IFF_PROMISC)
 593                         dev_set_promiscuity(new_active->dev, 1);
 594
 595                 if (bond->dev->flags & IFF_ALLMULTI)
 596                         dev_set_allmulti(new_active->dev, 1);
 597
 598                 netif_addr_lock_bh(bond->dev);
 599                 dev_uc_sync(new_active->dev, bond->dev);
 600                 dev_mc_sync(new_active->dev, bond->dev);
 601                 netif_addr_unlock_bh(bond->dev);
 602         }
 603 }
 604
 605 /**
 606  * bond_set_dev_addr - clone slave's address to bond
 607  * @bond_dev: bond net device
 608  * @slave_dev: slave net device
 609  *
 610  * Should be called with RTNL held.
 611  */
 612 static void bond_set_dev_addr(struct net_device *bond_dev,
 613                               struct net_device *slave_dev)
 614 {
 615         netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 616                    bond_dev, slave_dev, slave_dev->addr_len);
 617         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 618         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 619         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 620 }
 621
 622 /* bond_do_fail_over_mac
 623  *
 624  * Perform special MAC address swapping for fail_over_mac settings
 625  *
 626  * Called with RTNL
 627  */
 628 static void bond_do_fail_over_mac(struct bonding *bond,
 629                                   struct slave *new_active,
 630                                   struct slave *old_active)
 631 {
 632         u8 tmp_mac[ETH_ALEN];
 633         struct sockaddr saddr;
 634         int rv;
 635
 636         switch (bond->params.fail_over_mac) {
 637         case BOND_FOM_ACTIVE:
 638                 if (new_active)
 639                         bond_set_dev_addr(bond->dev, new_active->dev);
 640                 break;
 641         case BOND_FOM_FOLLOW:
 642                 /* if new_active && old_active, swap them
 643                  * if just old_active, do nothing (going to no active slave)
 644                  * if just new_active, set new_active to bond's MAC
 645                  */
 646                 if (!new_active)
 647                         return;
 648
 649                 if (old_active) {
 650                         ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
 651                         ether_addr_copy(saddr.sa_data,
 652                                         old_active->dev->dev_addr);
 653                         saddr.sa_family = new_active->dev->type;
 654                 } else {
 655                         ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
 656                         saddr.sa_family = bond->dev->type;
 657                 }
 658
 659                 rv = dev_set_mac_address(new_active->dev, &saddr);
 660                 if (rv) {
 661                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 662                                    -rv, new_active->dev->name);
 663                         goto out;
 664                 }
 665
 666                 if (!old_active)
 667                         goto out;
 668
 669                 ether_addr_copy(saddr.sa_data, tmp_mac);
 670                 saddr.sa_family = old_active->dev->type;
 671
 672                 rv = dev_set_mac_address(old_active->dev, &saddr);
 673                 if (rv)
 674                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 675                                    -rv, new_active->dev->name);
 676 out:
 677                 break;
 678         default:
 679                 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
 680                            bond->params.fail_over_mac);
 681                 break;
 682         }
 683
 684 }
 685
 686 static bool bond_should_change_active(struct bonding *bond)
 687 {
 688         struct slave *prim = rtnl_dereference(bond->primary_slave);
 689         struct slave *curr = rtnl_dereference(bond->curr_active_slave);
 690
 691         if (!prim || !curr || curr->link != BOND_LINK_UP)
 692                 return true;
 693         if (bond->force_primary) {
 694                 bond->force_primary = false;
 695                 return true;
 696         }
 697         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 698             (prim->speed < curr->speed ||
 699              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 700                 return false;
 701         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 702                 return false;
 703         return true;
 704 }
 705
 706 /**
 707  * find_best_interface - select the best available slave to be the active one
 708  * @bond: our bonding struct
 709  */
 710 static struct slave *bond_find_best_slave(struct bonding *bond)
 711 {
 712         struct slave *slave, *bestslave = NULL, *primary;
 713         struct list_head *iter;
 714         int mintime = bond->params.updelay;
 715
 716         primary = rtnl_dereference(bond->primary_slave);
 717         if (primary && primary->link == BOND_LINK_UP &&
 718             bond_should_change_active(bond))
 719                 return primary;
 720
 721         bond_for_each_slave(bond, slave, iter) {
 722                 if (slave->link == BOND_LINK_UP)
 723                         return slave;
 724                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
 725                     slave->delay < mintime) {
 726                         mintime = slave->delay;
 727                         bestslave = slave;
 728                 }
 729         }
 730
 731         return bestslave;
 732 }
 733
 734 static bool bond_should_notify_peers(struct bonding *bond)
 735 {
 736         struct slave *slave;
 737
 738         rcu_read_lock();
 739         slave = rcu_dereference(bond->curr_active_slave);
 740         rcu_read_unlock();
 741
 742         netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
 743                    slave ? slave->dev->name : "NULL");
 744
 745         if (!slave || !bond->send_peer_notif ||
 746             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 747                 return false;
 748
 749         return true;
 750 }
 751
 752 /**
 753  * change_active_interface - change the active slave into the specified one
 754  * @bond: our bonding struct
 755  * @new: the new slave to make the active one
 756  *
 757  * Set the new slave to the bond's settings and unset them on the old
 758  * curr_active_slave.
 759  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 760  *
 761  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 762  * because it is apparently the best available slave we have, even though its
 763  * updelay hasn't timed out yet.
 764  *
 765  * Caller must hold RTNL.
 766  */
 767 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 768 {
 769         struct slave *old_active;
 770
 771         ASSERT_RTNL();
 772
 773         old_active = rtnl_dereference(bond->curr_active_slave);
 774
 775         if (old_active == new_active)
 776                 return;
 777
 778         if (new_active) {
 779                 new_active->last_link_up = jiffies;
 780
 781                 if (new_active->link == BOND_LINK_BACK) {
 782                         if (bond_uses_primary(bond)) {
 783                                 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
 784                                             new_active->dev->name,
 785                                             (bond->params.updelay - new_active->delay) * bond->params.miimon);
 786                         }
 787
 788                         new_active->delay = 0;
 789                         new_active->link = BOND_LINK_UP;
 790
 791                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 792                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 793
 794                         if (bond_is_lb(bond))
 795                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 796                 } else {
 797                         if (bond_uses_primary(bond)) {
 798                                 netdev_info(bond->dev, "making interface %s the new active one\n",
 799                                             new_active->dev->name);
 800                         }
 801                 }
 802         }
 803
 804         if (bond_uses_primary(bond))
 805                 bond_hw_addr_swap(bond, new_active, old_active);
 806
 807         if (bond_is_lb(bond)) {
 808                 bond_alb_handle_active_change(bond, new_active);
 809                 if (old_active)
 810                         bond_set_slave_inactive_flags(old_active,
 811                                                       BOND_SLAVE_NOTIFY_NOW);
 812                 if (new_active)
 813                         bond_set_slave_active_flags(new_active,
 814                                                     BOND_SLAVE_NOTIFY_NOW);
 815         } else {
 816                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 817         }
 818
 819         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
 820                 if (old_active)
 821                         bond_set_slave_inactive_flags(old_active,
 822                                                       BOND_SLAVE_NOTIFY_NOW);
 823
 824                 if (new_active) {
 825                         bool should_notify_peers = false;
 826
 827                         bond_set_slave_active_flags(new_active,
 828                                                     BOND_SLAVE_NOTIFY_NOW);
 829
 830                         if (bond->params.fail_over_mac)
 831                                 bond_do_fail_over_mac(bond, new_active,
 832                                                       old_active);
 833
 834                         if (netif_running(bond->dev)) {
 835                                 bond->send_peer_notif =
 836                                         bond->params.num_peer_notif;
 837                                 should_notify_peers =
 838                                         bond_should_notify_peers(bond);
 839                         }
 840
 841                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 842                         if (should_notify_peers)
 843                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 844                                                          bond->dev);
 845                 }
 846         }
 847
 848         /* resend IGMP joins since active slave has changed or
 849          * all were sent on curr_active_slave.
 850          * resend only if bond is brought up with the affected
 851          * bonding modes and the retransmission is enabled
 852          */
 853         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 854             ((bond_uses_primary(bond) && new_active) ||
 855              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
 856                 bond->igmp_retrans = bond->params.resend_igmp;
 857                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 858         }
 859 }
 860
 861 /**
 862  * bond_select_active_slave - select a new active slave, if needed
 863  * @bond: our bonding struct
 864  *
 865  * This functions should be called when one of the following occurs:
 866  * - The old curr_active_slave has been released or lost its link.
 867  * - The primary_slave has got its link back.
 868  * - A slave has got its link back and there's no old curr_active_slave.
 869  *
 870  * Caller must hold RTNL.
 871  */
 872 void bond_select_active_slave(struct bonding *bond)
 873 {
 874         struct slave *best_slave;
 875         int rv;
 876
 877         ASSERT_RTNL();
 878
 879         best_slave = bond_find_best_slave(bond);
 880         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
 881                 bond_change_active_slave(bond, best_slave);
 882                 rv = bond_set_carrier(bond);
 883                 if (!rv)
 884                         return;
 885
 886                 if (netif_carrier_ok(bond->dev)) {
 887                         netdev_info(bond->dev, "first active interface up!\n");
 888                 } else {
 889                         netdev_info(bond->dev, "now running without any active interface!\n");
 890                 }
 891         }
 892 }
 893
 894 #ifdef CONFIG_NET_POLL_CONTROLLER
 895 static inline int slave_enable_netpoll(struct slave *slave)
 896 {
 897         struct netpoll *np;
 898         int err = 0;
 899
 900         np = kzalloc(sizeof(*np), GFP_KERNEL);
 901         err = -ENOMEM;
 902         if (!np)
 903                 goto out;
 904
 905         err = __netpoll_setup(np, slave->dev);
 906         if (err) {
 907                 kfree(np);
 908                 goto out;
 909         }
 910         slave->np = np;
 911 out:
 912         return err;
 913 }
 914 static inline void slave_disable_netpoll(struct slave *slave)
 915 {
 916         struct netpoll *np = slave->np;
 917
 918         if (!np)
 919                 return;
 920
 921         slave->np = NULL;
 922         __netpoll_free_async(np);
 923 }
 924
 925 static void bond_poll_controller(struct net_device *bond_dev)
 926 {
 927 }
 928
 929 static void bond_netpoll_cleanup(struct net_device *bond_dev)
 930 {
 931         struct bonding *bond = netdev_priv(bond_dev);
 932         struct list_head *iter;
 933         struct slave *slave;
 934
 935         bond_for_each_slave(bond, slave, iter)
 936                 if (bond_slave_is_up(slave))
 937                         slave_disable_netpoll(slave);
 938 }
 939
 940 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
 941 {
 942         struct bonding *bond = netdev_priv(dev);
 943         struct list_head *iter;
 944         struct slave *slave;
 945         int err = 0;
 946
 947         bond_for_each_slave(bond, slave, iter) {
 948                 err = slave_enable_netpoll(slave);
 949                 if (err) {
 950                         bond_netpoll_cleanup(dev);
 951                         break;
 952                 }
 953         }
 954         return err;
 955 }
 956 #else
 957 static inline int slave_enable_netpoll(struct slave *slave)
 958 {
 959         return 0;
 960 }
 961 static inline void slave_disable_netpoll(struct slave *slave)
 962 {
 963 }
 964 static void bond_netpoll_cleanup(struct net_device *bond_dev)
 965 {
 966 }
 967 #endif
 968
 969 /*---------------------------------- IOCTL ----------------------------------*/
 970
 971 static netdev_features_t bond_fix_features(struct net_device *dev,
 972                                            netdev_features_t features)
 973 {
 974         struct bonding *bond = netdev_priv(dev);
 975         struct list_head *iter;
 976         netdev_features_t mask;
 977         struct slave *slave;
 978
 979         mask = features;
 980         features &= ~NETIF_F_ONE_FOR_ALL;
 981         features |= NETIF_F_ALL_FOR_ALL;
 982
 983         bond_for_each_slave(bond, slave, iter) {
 984                 features = netdev_increment_features(features,
 985                                                      slave->dev->features,
 986                                                      mask);
 987         }
 988         features = netdev_add_tso_features(features, mask);
 989
 990         return features;
 991 }
 992
 993 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
 994                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
 995                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
 996
 997 #define BOND_ENC_FEATURES       (NETIF_F_ALL_CSUM | NETIF_F_SG | NETIF_F_RXCSUM |\
 998                                  NETIF_F_TSO | NETIF_F_GSO_UDP_TUNNEL)
 999
1000 static void bond_compute_features(struct bonding *bond)
1001 {
1002         unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1003         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1004         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1005         struct net_device *bond_dev = bond->dev;
1006         struct list_head *iter;
1007         struct slave *slave;
1008         unsigned short max_hard_header_len = ETH_HLEN;
1009         unsigned int gso_max_size = GSO_MAX_SIZE;
1010         u16 gso_max_segs = GSO_MAX_SEGS;
1011
1012         if (!bond_has_slaves(bond))
1013                 goto done;
1014         vlan_features &= NETIF_F_ALL_FOR_ALL;
1015
1016         bond_for_each_slave(bond, slave, iter) {
1017                 vlan_features = netdev_increment_features(vlan_features,
1018                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1019
1020                 enc_features = netdev_increment_features(enc_features,
1021                                                          slave->dev->hw_enc_features,
1022                                                          BOND_ENC_FEATURES);
1023                 dst_release_flag &= slave->dev->priv_flags;
1024                 if (slave->dev->hard_header_len > max_hard_header_len)
1025                         max_hard_header_len = slave->dev->hard_header_len;
1026
1027                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1028                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1029         }
1030
1031 done:
1032         bond_dev->vlan_features = vlan_features;
1033         bond_dev->hw_enc_features = enc_features;
1034         bond_dev->hard_header_len = max_hard_header_len;
1035         bond_dev->gso_max_segs = gso_max_segs;
1036         netif_set_gso_max_size(bond_dev, gso_max_size);
1037
1038         flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1039         bond_dev->priv_flags = flags | dst_release_flag;
1040
1041         netdev_change_features(bond_dev);
1042 }
1043
1044 static void bond_setup_by_slave(struct net_device *bond_dev,
1045                                 struct net_device *slave_dev)
1046 {
1047         bond_dev->header_ops        = slave_dev->header_ops;
1048
1049         bond_dev->type              = slave_dev->type;
1050         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1051         bond_dev->addr_len          = slave_dev->addr_len;
1052
1053         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1054                 slave_dev->addr_len);
1055 }
1056
1057 /* On bonding slaves other than the currently active slave, suppress
1058  * duplicates except for alb non-mcast/bcast.
1059  */
1060 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1061                                             struct slave *slave,
1062                                             struct bonding *bond)
1063 {
1064         if (bond_is_slave_inactive(slave)) {
1065                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1066                     skb->pkt_type != PACKET_BROADCAST &&
1067                     skb->pkt_type != PACKET_MULTICAST)
1068                         return false;
1069                 return true;
1070         }
1071         return false;
1072 }
1073
1074 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1075 {
1076         struct sk_buff *skb = *pskb;
1077         struct slave *slave;
1078         struct bonding *bond;
1079         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1080                           struct slave *);
1081         int ret = RX_HANDLER_ANOTHER;
1082
1083         skb = skb_share_check(skb, GFP_ATOMIC);
1084         if (unlikely(!skb))
1085                 return RX_HANDLER_CONSUMED;
1086
1087         *pskb = skb;
1088
1089         slave = bond_slave_get_rcu(skb->dev);
1090         bond = slave->bond;
1091
1092         recv_probe = ACCESS_ONCE(bond->recv_probe);
1093         if (recv_probe) {
1094                 ret = recv_probe(skb, bond, slave);
1095                 if (ret == RX_HANDLER_CONSUMED) {
1096                         consume_skb(skb);
1097                         return ret;
1098                 }
1099         }
1100
1101         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1102                 return RX_HANDLER_EXACT;
1103         }
1104
1105         skb->dev = bond->dev;
1106
1107         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1108             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1109             skb->pkt_type == PACKET_HOST) {
1110
1111                 if (unlikely(skb_cow_head(skb,
1112                                           skb->data - skb_mac_header(skb)))) {
1113                         kfree_skb(skb);
1114                         return RX_HANDLER_CONSUMED;
1115                 }
1116                 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1117         }
1118
1119         return ret;
1120 }
1121
1122 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1123                                       struct net_device *slave_dev,
1124                                       struct slave *slave)
1125 {
1126         int err;
1127
1128         err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1129         if (err)
1130                 return err;
1131         slave_dev->flags |= IFF_SLAVE;
1132         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1133         return 0;
1134 }
1135
1136 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1137                                   struct net_device *slave_dev)
1138 {
1139         netdev_upper_dev_unlink(slave_dev, bond_dev);
1140         slave_dev->flags &= ~IFF_SLAVE;
1141         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1142 }
1143
1144 static struct slave *bond_alloc_slave(struct bonding *bond)
1145 {
1146         struct slave *slave = NULL;
1147
1148         slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1149         if (!slave)
1150                 return NULL;
1151
1152         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1153                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1154                                                GFP_KERNEL);
1155                 if (!SLAVE_AD_INFO(slave)) {
1156                         kfree(slave);
1157                         return NULL;
1158                 }
1159         }
1160         return slave;
1161 }
1162
1163 static void bond_free_slave(struct slave *slave)
1164 {
1165         struct bonding *bond = bond_get_bond_by_slave(slave);
1166
1167         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1168                 kfree(SLAVE_AD_INFO(slave));
1169
1170         kfree(slave);
1171 }
1172
1173 /* enslave device <slave> to bond device <master> */
1174 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1175 {
1176         struct bonding *bond = netdev_priv(bond_dev);
1177         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1178         struct slave *new_slave = NULL, *prev_slave;
1179         struct sockaddr addr;
1180         int link_reporting;
1181         int res = 0, i;
1182
1183         if (!bond->params.use_carrier &&
1184             slave_dev->ethtool_ops->get_link == NULL &&
1185             slave_ops->ndo_do_ioctl == NULL) {
1186                 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1187                             slave_dev->name);
1188         }
1189
1190         /* already enslaved */
1191         if (slave_dev->flags & IFF_SLAVE) {
1192                 netdev_dbg(bond_dev, "Error: Device was already enslaved\n");
1193                 return -EBUSY;
1194         }
1195
1196         if (bond_dev == slave_dev) {
1197                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1198                 return -EPERM;
1199         }
1200
1201         /* vlan challenged mutual exclusion */
1202         /* no need to lock since we're protected by rtnl_lock */
1203         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1204                 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1205                            slave_dev->name);
1206                 if (vlan_uses_dev(bond_dev)) {
1207                         netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1208                                    slave_dev->name, bond_dev->name);
1209                         return -EPERM;
1210                 } else {
1211                         netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1212                                     slave_dev->name, slave_dev->name,
1213                                     bond_dev->name);
1214                 }
1215         } else {
1216                 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1217                            slave_dev->name);
1218         }
1219
1220         /* Old ifenslave binaries are no longer supported.  These can
1221          * be identified with moderate accuracy by the state of the slave:
1222          * the current ifenslave will set the interface down prior to
1223          * enslaving it; the old ifenslave will not.
1224          */
1225         if ((slave_dev->flags & IFF_UP)) {
1226                 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1227                            slave_dev->name);
1228                 res = -EPERM;
1229                 goto err_undo_flags;
1230         }
1231
1232         /* set bonding device ether type by slave - bonding netdevices are
1233          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1234          * there is a need to override some of the type dependent attribs/funcs.
1235          *
1236          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1237          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1238          */
1239         if (!bond_has_slaves(bond)) {
1240                 if (bond_dev->type != slave_dev->type) {
1241                         netdev_dbg(bond_dev, "change device type from %d to %d\n",
1242                                    bond_dev->type, slave_dev->type);
1243
1244                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1245                                                        bond_dev);
1246                         res = notifier_to_errno(res);
1247                         if (res) {
1248                                 netdev_err(bond_dev, "refused to change device type\n");
1249                                 res = -EBUSY;
1250                                 goto err_undo_flags;
1251                         }
1252
1253                         /* Flush unicast and multicast addresses */
1254                         dev_uc_flush(bond_dev);
1255                         dev_mc_flush(bond_dev);
1256
1257                         if (slave_dev->type != ARPHRD_ETHER)
1258                                 bond_setup_by_slave(bond_dev, slave_dev);
1259                         else {
1260                                 ether_setup(bond_dev);
1261                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1262                         }
1263
1264                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1265                                                  bond_dev);
1266                 }
1267         } else if (bond_dev->type != slave_dev->type) {
1268                 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1269                            slave_dev->name, slave_dev->type, bond_dev->type);
1270                 res = -EINVAL;
1271                 goto err_undo_flags;
1272         }
1273
1274         if (slave_ops->ndo_set_mac_address == NULL) {
1275                 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1276                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1277                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1278                         if (!bond_has_slaves(bond)) {
1279                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1280                                 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1281                         } else {
1282                                 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");
1283                                 res = -EOPNOTSUPP;
1284                                 goto err_undo_flags;
1285                         }
1286                 }
1287         }
1288
1289         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1290
1291         /* If this is the first slave, then we need to set the master's hardware
1292          * address to be the same as the slave's.
1293          */
1294         if (!bond_has_slaves(bond) &&
1295             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1296                 bond_set_dev_addr(bond->dev, slave_dev);
1297
1298         new_slave = bond_alloc_slave(bond);
1299         if (!new_slave) {
1300                 res = -ENOMEM;
1301                 goto err_undo_flags;
1302         }
1303
1304         new_slave->bond = bond;
1305         new_slave->dev = slave_dev;
1306         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1307          * is set via sysfs or module option if desired.
1308          */
1309         new_slave->queue_id = 0;
1310
1311         /* Save slave's original mtu and then set it to match the bond */
1312         new_slave->original_mtu = slave_dev->mtu;
1313         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1314         if (res) {
1315                 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1316                 goto err_free;
1317         }
1318
1319         /* Save slave's original ("permanent") mac address for modes
1320          * that need it, and for restoring it upon release, and then
1321          * set it to the master's address
1322          */
1323         ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1324
1325         if (!bond->params.fail_over_mac ||
1326             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1327                 /* Set slave to master's mac address.  The application already
1328                  * set the master's mac address to that of the first slave
1329                  */
1330                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1331                 addr.sa_family = slave_dev->type;
1332                 res = dev_set_mac_address(slave_dev, &addr);
1333                 if (res) {
1334                         netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1335                         goto err_restore_mtu;
1336                 }
1337         }
1338
1339         /* open the slave since the application closed it */
1340         res = dev_open(slave_dev);
1341         if (res) {
1342                 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1343                 goto err_restore_mac;
1344         }
1345
1346         slave_dev->priv_flags |= IFF_BONDING;
1347
1348         if (bond_is_lb(bond)) {
1349                 /* bond_alb_init_slave() must be called before all other stages since
1350                  * it might fail and we do not want to have to undo everything
1351                  */
1352                 res = bond_alb_init_slave(bond, new_slave);
1353                 if (res)
1354                         goto err_close;
1355         }
1356
1357         /* If the mode uses primary, then the following is handled by
1358          * bond_change_active_slave().
1359          */
1360         if (!bond_uses_primary(bond)) {
1361                 /* set promiscuity level to new slave */
1362                 if (bond_dev->flags & IFF_PROMISC) {
1363                         res = dev_set_promiscuity(slave_dev, 1);
1364                         if (res)
1365                                 goto err_close;
1366                 }
1367
1368                 /* set allmulti level to new slave */
1369                 if (bond_dev->flags & IFF_ALLMULTI) {
1370                         res = dev_set_allmulti(slave_dev, 1);
1371                         if (res)
1372                                 goto err_close;
1373                 }
1374
1375                 netif_addr_lock_bh(bond_dev);
1376
1377                 dev_mc_sync_multiple(slave_dev, bond_dev);
1378                 dev_uc_sync_multiple(slave_dev, bond_dev);
1379
1380                 netif_addr_unlock_bh(bond_dev);
1381         }
1382
1383         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1384                 /* add lacpdu mc addr to mc list */
1385                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1386
1387                 dev_mc_add(slave_dev, lacpdu_multicast);
1388         }
1389
1390         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1391         if (res) {
1392                 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1393                            slave_dev->name);
1394                 goto err_close;
1395         }
1396
1397         prev_slave = bond_last_slave(bond);
1398
1399         new_slave->delay = 0;
1400         new_slave->link_failure_count = 0;
1401
1402         bond_update_speed_duplex(new_slave);
1403
1404         new_slave->last_rx = jiffies -
1405                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1406         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1407                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1408
1409         if (bond->params.miimon && !bond->params.use_carrier) {
1410                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1411
1412                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1413                         /* miimon is set but a bonded network driver
1414                          * does not support ETHTOOL/MII and
1415                          * arp_interval is not set.  Note: if
1416                          * use_carrier is enabled, we will never go
1417                          * here (because netif_carrier is always
1418                          * supported); thus, we don't need to change
1419                          * the messages for netif_carrier.
1420                          */
1421                         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",
1422                                     slave_dev->name);
1423                 } else if (link_reporting == -1) {
1424                         /* unable get link status using mii/ethtool */
1425                         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",
1426                                     slave_dev->name);
1427                 }
1428         }
1429
1430         /* check for initial state */
1431         if (bond->params.miimon) {
1432                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1433                         if (bond->params.updelay) {
1434                                 new_slave->link = BOND_LINK_BACK;
1435                                 new_slave->delay = bond->params.updelay;
1436                         } else {
1437                                 new_slave->link = BOND_LINK_UP;
1438                         }
1439                 } else {
1440                         new_slave->link = BOND_LINK_DOWN;
1441                 }
1442         } else if (bond->params.arp_interval) {
1443                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1444                         BOND_LINK_UP : BOND_LINK_DOWN);
1445         } else {
1446                 new_slave->link = BOND_LINK_UP;
1447         }
1448
1449         if (new_slave->link != BOND_LINK_DOWN)
1450                 new_slave->last_link_up = jiffies;
1451         netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1452                    new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1453                    (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1454
1455         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1456                 /* if there is a primary slave, remember it */
1457                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1458                         rcu_assign_pointer(bond->primary_slave, new_slave);
1459                         bond->force_primary = true;
1460                 }
1461         }
1462
1463         switch (BOND_MODE(bond)) {
1464         case BOND_MODE_ACTIVEBACKUP:
1465                 bond_set_slave_inactive_flags(new_slave,
1466                                               BOND_SLAVE_NOTIFY_NOW);
1467                 break;
1468         case BOND_MODE_8023AD:
1469                 /* in 802.3ad mode, the internal mechanism
1470                  * will activate the slaves in the selected
1471                  * aggregator
1472                  */
1473                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1474                 /* if this is the first slave */
1475                 if (!prev_slave) {
1476                         SLAVE_AD_INFO(new_slave)->id = 1;
1477                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1478                          * can be called only after the mac address of the bond is set
1479                          */
1480                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1481                 } else {
1482                         SLAVE_AD_INFO(new_slave)->id =
1483                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1484                 }
1485
1486                 bond_3ad_bind_slave(new_slave);
1487                 break;
1488         case BOND_MODE_TLB:
1489         case BOND_MODE_ALB:
1490                 bond_set_active_slave(new_slave);
1491                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1492                 break;
1493         default:
1494                 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1495
1496                 /* always active in trunk mode */
1497                 bond_set_active_slave(new_slave);
1498
1499                 /* In trunking mode there is little meaning to curr_active_slave
1500                  * anyway (it holds no special properties of the bond device),
1501                  * so we can change it without calling change_active_interface()
1502                  */
1503                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1504                     new_slave->link == BOND_LINK_UP)
1505                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1506
1507                 break;
1508         } /* switch(bond_mode) */
1509
1510 #ifdef CONFIG_NET_POLL_CONTROLLER
1511         slave_dev->npinfo = bond->dev->npinfo;
1512         if (slave_dev->npinfo) {
1513                 if (slave_enable_netpoll(new_slave)) {
1514                         netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1515                         res = -EBUSY;
1516                         goto err_detach;
1517                 }
1518         }
1519 #endif
1520
1521         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1522                                          new_slave);
1523         if (res) {
1524                 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1525                 goto err_detach;
1526         }
1527
1528         res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1529         if (res) {
1530                 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1531                 goto err_unregister;
1532         }
1533
1534         res = bond_sysfs_slave_add(new_slave);
1535         if (res) {
1536                 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1537                 goto err_upper_unlink;
1538         }
1539
1540         bond->slave_cnt++;
1541         bond_compute_features(bond);
1542         bond_set_carrier(bond);
1543
1544         if (bond_uses_primary(bond)) {
1545                 block_netpoll_tx();
1546                 bond_select_active_slave(bond);
1547                 unblock_netpoll_tx();
1548         }
1549
1550         netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1551                     slave_dev->name,
1552                     bond_is_active_slave(new_slave) ? "an active" : "a backup",
1553                     new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1554
1555         /* enslave is successful */
1556         return 0;
1557
1558 /* Undo stages on error */
1559 err_upper_unlink:
1560         bond_upper_dev_unlink(bond_dev, slave_dev);
1561
1562 err_unregister:
1563         netdev_rx_handler_unregister(slave_dev);
1564
1565 err_detach:
1566         if (!bond_uses_primary(bond))
1567                 bond_hw_addr_flush(bond_dev, slave_dev);
1568
1569         vlan_vids_del_by_dev(slave_dev, bond_dev);
1570         if (rcu_access_pointer(bond->primary_slave) == new_slave)
1571                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1572         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1573                 block_netpoll_tx();
1574                 bond_change_active_slave(bond, NULL);
1575                 bond_select_active_slave(bond);
1576                 unblock_netpoll_tx();
1577         }
1578         /* either primary_slave or curr_active_slave might've changed */
1579         synchronize_rcu();
1580         slave_disable_netpoll(new_slave);
1581
1582 err_close:
1583         slave_dev->priv_flags &= ~IFF_BONDING;
1584         dev_close(slave_dev);
1585
1586 err_restore_mac:
1587         if (!bond->params.fail_over_mac ||
1588             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1589                 /* XXX TODO - fom follow mode needs to change master's
1590                  * MAC if this slave's MAC is in use by the bond, or at
1591                  * least print a warning.
1592                  */
1593                 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1594                 addr.sa_family = slave_dev->type;
1595                 dev_set_mac_address(slave_dev, &addr);
1596         }
1597
1598 err_restore_mtu:
1599         dev_set_mtu(slave_dev, new_slave->original_mtu);
1600
1601 err_free:
1602         bond_free_slave(new_slave);
1603
1604 err_undo_flags:
1605         /* Enslave of first slave has failed and we need to fix master's mac */
1606         if (!bond_has_slaves(bond) &&
1607             ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
1608                 eth_hw_addr_random(bond_dev);
1609
1610         return res;
1611 }
1612
1613 /* Try to release the slave device <slave> from the bond device <master>
1614  * It is legal to access curr_active_slave without a lock because all the function
1615  * is RTNL-locked. If "all" is true it means that the function is being called
1616  * while destroying a bond interface and all slaves are being released.
1617  *
1618  * The rules for slave state should be:
1619  *   for Active/Backup:
1620  *     Active stays on all backups go down
1621  *   for Bonded connections:
1622  *     The first up interface should be left on and all others downed.
1623  */
1624 static int __bond_release_one(struct net_device *bond_dev,
1625                               struct net_device *slave_dev,
1626                               bool all)
1627 {
1628         struct bonding *bond = netdev_priv(bond_dev);
1629         struct slave *slave, *oldcurrent;
1630         struct sockaddr addr;
1631         int old_flags = bond_dev->flags;
1632         netdev_features_t old_features = bond_dev->features;
1633
1634         /* slave is not a slave or master is not master of this slave */
1635         if (!(slave_dev->flags & IFF_SLAVE) ||
1636             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1637                 netdev_err(bond_dev, "cannot release %s\n",
1638                            slave_dev->name);
1639                 return -EINVAL;
1640         }
1641
1642         block_netpoll_tx();
1643
1644         slave = bond_get_slave_by_dev(bond, slave_dev);
1645         if (!slave) {
1646                 /* not a slave of this bond */
1647                 netdev_info(bond_dev, "%s not enslaved\n",
1648                             slave_dev->name);
1649                 unblock_netpoll_tx();
1650                 return -EINVAL;
1651         }
1652
1653         bond_sysfs_slave_del(slave);
1654
1655         bond_upper_dev_unlink(bond_dev, slave_dev);
1656         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1657          * for this slave anymore.
1658          */
1659         netdev_rx_handler_unregister(slave_dev);
1660
1661         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1662                 bond_3ad_unbind_slave(slave);
1663
1664         netdev_info(bond_dev, "Releasing %s interface %s\n",
1665                     bond_is_active_slave(slave) ? "active" : "backup",
1666                     slave_dev->name);
1667
1668         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1669
1670         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1671
1672         if (!all && (!bond->params.fail_over_mac ||
1673                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1674                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1675                     bond_has_slaves(bond))
1676                         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",
1677                                     slave_dev->name, slave->perm_hwaddr,
1678                                     bond_dev->name, slave_dev->name);
1679         }
1680
1681         if (rtnl_dereference(bond->primary_slave) == slave)
1682                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1683
1684         if (oldcurrent == slave)
1685                 bond_change_active_slave(bond, NULL);
1686
1687         if (bond_is_lb(bond)) {
1688                 /* Must be called only after the slave has been
1689                  * detached from the list and the curr_active_slave
1690                  * has been cleared (if our_slave == old_current),
1691                  * but before a new active slave is selected.
1692                  */
1693                 bond_alb_deinit_slave(bond, slave);
1694         }
1695
1696         if (all) {
1697                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1698         } else if (oldcurrent == slave) {
1699                 /* Note that we hold RTNL over this sequence, so there
1700                  * is no concern that another slave add/remove event
1701                  * will interfere.
1702                  */
1703                 bond_select_active_slave(bond);
1704         }
1705
1706         if (!bond_has_slaves(bond)) {
1707                 bond_set_carrier(bond);
1708                 eth_hw_addr_random(bond_dev);
1709         }
1710
1711         unblock_netpoll_tx();
1712         synchronize_rcu();
1713         bond->slave_cnt--;
1714
1715         if (!bond_has_slaves(bond)) {
1716                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1717                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1718         }
1719
1720         bond_compute_features(bond);
1721         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1722             (old_features & NETIF_F_VLAN_CHALLENGED))
1723                 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1724                             slave_dev->name, bond_dev->name);
1725
1726         vlan_vids_del_by_dev(slave_dev, bond_dev);
1727
1728         /* If the mode uses primary, then this case was handled above by
1729          * bond_change_active_slave(..., NULL)
1730          */
1731         if (!bond_uses_primary(bond)) {
1732                 /* unset promiscuity level from slave
1733                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1734                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1735                  * value of that flag before that change, as that was the value
1736                  * when this slave was attached, so we cache at the start of the
1737                  * function and use it here. Same goes for ALLMULTI below
1738                  */
1739                 if (old_flags & IFF_PROMISC)
1740                         dev_set_promiscuity(slave_dev, -1);
1741
1742                 /* unset allmulti level from slave */
1743                 if (old_flags & IFF_ALLMULTI)
1744                         dev_set_allmulti(slave_dev, -1);
1745
1746                 bond_hw_addr_flush(bond_dev, slave_dev);
1747         }
1748
1749         slave_disable_netpoll(slave);
1750
1751         /* close slave before restoring its mac address */
1752         dev_close(slave_dev);
1753
1754         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1755             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1756                 /* restore original ("permanent") mac address */
1757                 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1758                 addr.sa_family = slave_dev->type;
1759                 dev_set_mac_address(slave_dev, &addr);
1760         }
1761
1762         dev_set_mtu(slave_dev, slave->original_mtu);
1763
1764         slave_dev->priv_flags &= ~IFF_BONDING;
1765
1766         bond_free_slave(slave);
1767
1768         return 0;
1769 }
1770
1771 /* A wrapper used because of ndo_del_link */
1772 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1773 {
1774         return __bond_release_one(bond_dev, slave_dev, false);
1775 }
1776
1777 /* First release a slave and then destroy the bond if no more slaves are left.
1778  * Must be under rtnl_lock when this function is called.
1779  */
1780 static int  bond_release_and_destroy(struct net_device *bond_dev,
1781                                      struct net_device *slave_dev)
1782 {
1783         struct bonding *bond = netdev_priv(bond_dev);
1784         int ret;
1785
1786         ret = bond_release(bond_dev, slave_dev);
1787         if (ret == 0 && !bond_has_slaves(bond)) {
1788                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1789                 netdev_info(bond_dev, "Destroying bond %s\n",
1790                             bond_dev->name);
1791                 unregister_netdevice(bond_dev);
1792         }
1793         return ret;
1794 }
1795
1796 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1797 {
1798         struct bonding *bond = netdev_priv(bond_dev);
1799
1800         info->bond_mode = BOND_MODE(bond);
1801         info->miimon = bond->params.miimon;
1802         info->num_slaves = bond->slave_cnt;
1803
1804         return 0;
1805 }
1806
1807 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1808 {
1809         struct bonding *bond = netdev_priv(bond_dev);
1810         struct list_head *iter;
1811         int i = 0, res = -ENODEV;
1812         struct slave *slave;
1813
1814         bond_for_each_slave(bond, slave, iter) {
1815                 if (i++ == (int)info->slave_id) {
1816                         res = 0;
1817                         strcpy(info->slave_name, slave->dev->name);
1818                         info->link = slave->link;
1819                         info->state = bond_slave_state(slave);
1820                         info->link_failure_count = slave->link_failure_count;
1821                         break;
1822                 }
1823         }
1824
1825         return res;
1826 }
1827
1828 /*-------------------------------- Monitoring -------------------------------*/
1829
1830 /* called with rcu_read_lock() */
1831 static int bond_miimon_inspect(struct bonding *bond)
1832 {
1833         int link_state, commit = 0;
1834         struct list_head *iter;
1835         struct slave *slave;
1836         bool ignore_updelay;
1837
1838         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
1839
1840         bond_for_each_slave_rcu(bond, slave, iter) {
1841                 slave->new_link = BOND_LINK_NOCHANGE;
1842
1843                 link_state = bond_check_dev_link(bond, slave->dev, 0);
1844
1845                 switch (slave->link) {
1846                 case BOND_LINK_UP:
1847                         if (link_state)
1848                                 continue;
1849
1850                         slave->link = BOND_LINK_FAIL;
1851                         slave->delay = bond->params.downdelay;
1852                         if (slave->delay) {
1853                                 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
1854                                             (BOND_MODE(bond) ==
1855                                              BOND_MODE_ACTIVEBACKUP) ?
1856                                              (bond_is_active_slave(slave) ?
1857                                               "active " : "backup ") : "",
1858                                             slave->dev->name,
1859                                             bond->params.downdelay * bond->params.miimon);
1860                         }
1861                         /*FALLTHRU*/
1862                 case BOND_LINK_FAIL:
1863                         if (link_state) {
1864                                 /* recovered before downdelay expired */
1865                                 slave->link = BOND_LINK_UP;
1866                                 slave->last_link_up = jiffies;
1867                                 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
1868                                             (bond->params.downdelay - slave->delay) *
1869                                             bond->params.miimon,
1870                                             slave->dev->name);
1871                                 continue;
1872                         }
1873
1874                         if (slave->delay <= 0) {
1875                                 slave->new_link = BOND_LINK_DOWN;
1876                                 commit++;
1877                                 continue;
1878                         }
1879
1880                         slave->delay--;
1881                         break;
1882
1883                 case BOND_LINK_DOWN:
1884                         if (!link_state)
1885                                 continue;
1886
1887                         slave->link = BOND_LINK_BACK;
1888                         slave->delay = bond->params.updelay;
1889
1890                         if (slave->delay) {
1891                                 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
1892                                             slave->dev->name,
1893                                             ignore_updelay ? 0 :
1894                                             bond->params.updelay *
1895                                             bond->params.miimon);
1896                         }
1897                         /*FALLTHRU*/
1898                 case BOND_LINK_BACK:
1899                         if (!link_state) {
1900                                 slave->link = BOND_LINK_DOWN;
1901                                 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
1902                                             (bond->params.updelay - slave->delay) *
1903                                             bond->params.miimon,
1904                                             slave->dev->name);
1905
1906                                 continue;
1907                         }
1908
1909                         if (ignore_updelay)
1910                                 slave->delay = 0;
1911
1912                         if (slave->delay <= 0) {
1913                                 slave->new_link = BOND_LINK_UP;
1914                                 commit++;
1915                                 ignore_updelay = false;
1916                                 continue;
1917                         }
1918
1919                         slave->delay--;
1920                         break;
1921                 }
1922         }
1923
1924         return commit;
1925 }
1926
1927 static void bond_miimon_commit(struct bonding *bond)
1928 {
1929         struct list_head *iter;
1930         struct slave *slave, *primary;
1931
1932         bond_for_each_slave(bond, slave, iter) {
1933                 switch (slave->new_link) {
1934                 case BOND_LINK_NOCHANGE:
1935                         continue;
1936
1937                 case BOND_LINK_UP:
1938                         slave->link = BOND_LINK_UP;
1939                         slave->last_link_up = jiffies;
1940
1941                         primary = rtnl_dereference(bond->primary_slave);
1942                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1943                                 /* prevent it from being the active one */
1944                                 bond_set_backup_slave(slave);
1945                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1946                                 /* make it immediately active */
1947                                 bond_set_active_slave(slave);
1948                         } else if (slave != primary) {
1949                                 /* prevent it from being the active one */
1950                                 bond_set_backup_slave(slave);
1951                         }
1952
1953                         netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
1954                                     slave->dev->name,
1955                                     slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
1956                                     slave->duplex ? "full" : "half");
1957
1958                         /* notify ad that the link status has changed */
1959                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1960                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
1961
1962                         if (bond_is_lb(bond))
1963                                 bond_alb_handle_link_change(bond, slave,
1964                                                             BOND_LINK_UP);
1965
1966                         if (!bond->curr_active_slave || slave == primary)
1967                                 goto do_failover;
1968
1969                         continue;
1970
1971                 case BOND_LINK_DOWN:
1972                         if (slave->link_failure_count < UINT_MAX)
1973                                 slave->link_failure_count++;
1974
1975                         slave->link = BOND_LINK_DOWN;
1976
1977                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
1978                             BOND_MODE(bond) == BOND_MODE_8023AD)
1979                                 bond_set_slave_inactive_flags(slave,
1980                                                               BOND_SLAVE_NOTIFY_NOW);
1981
1982                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
1983                                     slave->dev->name);
1984
1985                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1986                                 bond_3ad_handle_link_change(slave,
1987                                                             BOND_LINK_DOWN);
1988
1989                         if (bond_is_lb(bond))
1990                                 bond_alb_handle_link_change(bond, slave,
1991                                                             BOND_LINK_DOWN);
1992
1993                         if (slave == rcu_access_pointer(bond->curr_active_slave))
1994                                 goto do_failover;
1995
1996                         continue;
1997
1998                 default:
1999                         netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2000                                    slave->new_link, slave->dev->name);
2001                         slave->new_link = BOND_LINK_NOCHANGE;
2002
2003                         continue;
2004                 }
2005
2006 do_failover:
2007                 block_netpoll_tx();
2008                 bond_select_active_slave(bond);
2009                 unblock_netpoll_tx();
2010         }
2011
2012         bond_set_carrier(bond);
2013 }
2014
2015 /* bond_mii_monitor
2016  *
2017  * Really a wrapper that splits the mii monitor into two phases: an
2018  * inspection, then (if inspection indicates something needs to be done)
2019  * an acquisition of appropriate locks followed by a commit phase to
2020  * implement whatever link state changes are indicated.
2021  */
2022 static void bond_mii_monitor(struct work_struct *work)
2023 {
2024         struct bonding *bond = container_of(work, struct bonding,
2025                                             mii_work.work);
2026         bool should_notify_peers = false;
2027         unsigned long delay;
2028
2029         delay = msecs_to_jiffies(bond->params.miimon);
2030
2031         if (!bond_has_slaves(bond))
2032                 goto re_arm;
2033
2034         rcu_read_lock();
2035
2036         should_notify_peers = bond_should_notify_peers(bond);
2037
2038         if (bond_miimon_inspect(bond)) {
2039                 rcu_read_unlock();
2040
2041                 /* Race avoidance with bond_close cancel of workqueue */
2042                 if (!rtnl_trylock()) {
2043                         delay = 1;
2044                         should_notify_peers = false;
2045                         goto re_arm;
2046                 }
2047
2048                 bond_miimon_commit(bond);
2049
2050                 rtnl_unlock();  /* might sleep, hold no other locks */
2051         } else
2052                 rcu_read_unlock();
2053
2054 re_arm:
2055         if (bond->params.miimon)
2056                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2057
2058         if (should_notify_peers) {
2059                 if (!rtnl_trylock())
2060                         return;
2061                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2062                 rtnl_unlock();
2063         }
2064 }
2065
2066 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2067 {
2068         struct net_device *upper;
2069         struct list_head *iter;
2070         bool ret = false;
2071
2072         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2073                 return true;
2074
2075         rcu_read_lock();
2076         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2077                 if (ip == bond_confirm_addr(upper, 0, ip)) {
2078                         ret = true;
2079                         break;
2080                 }
2081         }
2082         rcu_read_unlock();
2083
2084         return ret;
2085 }
2086
2087 /* We go to the (large) trouble of VLAN tagging ARP frames because
2088  * switches in VLAN mode (especially if ports are configured as
2089  * "native" to a VLAN) might not pass non-tagged frames.
2090  */
2091 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2092                           __be32 dest_ip, __be32 src_ip,
2093                           struct bond_vlan_tag *tags)
2094 {
2095         struct sk_buff *skb;
2096         struct bond_vlan_tag *outer_tag = tags;
2097
2098         netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2099                    arp_op, slave_dev->name, &dest_ip, &src_ip);
2100
2101         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2102                          NULL, slave_dev->dev_addr, NULL);
2103
2104         if (!skb) {
2105                 net_err_ratelimited("ARP packet allocation failed\n");
2106                 return;
2107         }
2108
2109         if (!tags || tags->vlan_proto == VLAN_N_VID)
2110                 goto xmit;
2111
2112         tags++;
2113
2114         /* Go through all the tags backwards and add them to the packet */
2115         while (tags->vlan_proto != VLAN_N_VID) {
2116                 if (!tags->vlan_id) {
2117                         tags++;
2118                         continue;
2119                 }
2120
2121                 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2122                            ntohs(outer_tag->vlan_proto), tags->vlan_id);
2123                 skb = __vlan_put_tag(skb, tags->vlan_proto,
2124                                      tags->vlan_id);
2125                 if (!skb) {
2126                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2127                         return;
2128                 }
2129
2130                 tags++;
2131         }
2132         /* Set the outer tag */
2133         if (outer_tag->vlan_id) {
2134                 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2135                            ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2136                 skb = vlan_put_tag(skb, outer_tag->vlan_proto,
2137                                    outer_tag->vlan_id);
2138                 if (!skb) {
2139                         net_err_ratelimited("failed to insert outer VLAN tag\n");
2140                         return;
2141                 }
2142         }
2143
2144 xmit:
2145         arp_xmit(skb);
2146 }
2147
2148 /* Validate the device path between the @start_dev and the @end_dev.
2149  * The path is valid if the @end_dev is reachable through device
2150  * stacking.
2151  * When the path is validated, collect any vlan information in the
2152  * path.
2153  */
2154 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2155                                               struct net_device *end_dev,
2156                                               int level)
2157 {
2158         struct bond_vlan_tag *tags;
2159         struct net_device *upper;
2160         struct list_head  *iter;
2161
2162         if (start_dev == end_dev) {
2163                 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2164                 if (!tags)
2165                         return ERR_PTR(-ENOMEM);
2166                 tags[level].vlan_proto = VLAN_N_VID;
2167                 return tags;
2168         }
2169
2170         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2171                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2172                 if (IS_ERR_OR_NULL(tags)) {
2173                         if (IS_ERR(tags))
2174                                 return tags;
2175                         continue;
2176                 }
2177                 if (is_vlan_dev(upper)) {
2178                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2179                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2180                 }
2181
2182                 return tags;
2183         }
2184
2185         return NULL;
2186 }
2187
2188 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2189 {
2190         struct rtable *rt;
2191         struct bond_vlan_tag *tags;
2192         __be32 *targets = bond->params.arp_targets, addr;
2193         int i;
2194
2195         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2196                 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2197                 tags = NULL;
2198
2199                 /* Find out through which dev should the packet go */
2200                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2201                                      RTO_ONLINK, 0);
2202                 if (IS_ERR(rt)) {
2203                         /* there's no route to target - try to send arp
2204                          * probe to generate any traffic (arp_validate=0)
2205                          */
2206                         if (bond->params.arp_validate)
2207                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2208                                                      bond->dev->name,
2209                                                      &targets[i]);
2210                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2211                                       0, tags);
2212                         continue;
2213                 }
2214
2215                 /* bond device itself */
2216                 if (rt->dst.dev == bond->dev)
2217                         goto found;
2218
2219                 rcu_read_lock();
2220                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2221                 rcu_read_unlock();
2222
2223                 if (!IS_ERR_OR_NULL(tags))
2224                         goto found;
2225
2226                 /* Not our device - skip */
2227                 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2228                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2229
2230                 ip_rt_put(rt);
2231                 continue;
2232
2233 found:
2234                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2235                 ip_rt_put(rt);
2236                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2237                               addr, tags);
2238                 kfree(tags);
2239         }
2240 }
2241
2242 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2243 {
2244         int i;
2245
2246         if (!sip || !bond_has_this_ip(bond, tip)) {
2247                 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2248                            &sip, &tip);
2249                 return;
2250         }
2251
2252         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2253         if (i == -1) {
2254                 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2255                            &sip);
2256                 return;
2257         }
2258         slave->last_rx = jiffies;
2259         slave->target_last_arp_rx[i] = jiffies;
2260 }
2261
2262 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2263                  struct slave *slave)
2264 {
2265         struct arphdr *arp = (struct arphdr *)skb->data;
2266         struct slave *curr_active_slave;
2267         unsigned char *arp_ptr;
2268         __be32 sip, tip;
2269         int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2270
2271         if (!slave_do_arp_validate(bond, slave)) {
2272                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2273                     !slave_do_arp_validate_only(bond))
2274                         slave->last_rx = jiffies;
2275                 return RX_HANDLER_ANOTHER;
2276         } else if (!is_arp) {
2277                 return RX_HANDLER_ANOTHER;
2278         }
2279
2280         alen = arp_hdr_len(bond->dev);
2281
2282         netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2283                    skb->dev->name);
2284
2285         if (alen > skb_headlen(skb)) {
2286                 arp = kmalloc(alen, GFP_ATOMIC);
2287                 if (!arp)
2288                         goto out_unlock;
2289                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2290                         goto out_unlock;
2291         }
2292
2293         if (arp->ar_hln != bond->dev->addr_len ||
2294             skb->pkt_type == PACKET_OTHERHOST ||
2295             skb->pkt_type == PACKET_LOOPBACK ||
2296             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2297             arp->ar_pro != htons(ETH_P_IP) ||
2298             arp->ar_pln != 4)
2299                 goto out_unlock;
2300
2301         arp_ptr = (unsigned char *)(arp + 1);
2302         arp_ptr += bond->dev->addr_len;
2303         memcpy(&sip, arp_ptr, 4);
2304         arp_ptr += 4 + bond->dev->addr_len;
2305         memcpy(&tip, arp_ptr, 4);
2306
2307         netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2308                    slave->dev->name, bond_slave_state(slave),
2309                      bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2310                      &sip, &tip);
2311
2312         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2313
2314         /* Backup slaves won't see the ARP reply, but do come through
2315          * here for each ARP probe (so we swap the sip/tip to validate
2316          * the probe).  In a "redundant switch, common router" type of
2317          * configuration, the ARP probe will (hopefully) travel from
2318          * the active, through one switch, the router, then the other
2319          * switch before reaching the backup.
2320          *
2321          * We 'trust' the arp requests if there is an active slave and
2322          * it received valid arp reply(s) after it became active. This
2323          * is done to avoid endless looping when we can't reach the
2324          * arp_ip_target and fool ourselves with our own arp requests.
2325          */
2326
2327         if (bond_is_active_slave(slave))
2328                 bond_validate_arp(bond, slave, sip, tip);
2329         else if (curr_active_slave &&
2330                  time_after(slave_last_rx(bond, curr_active_slave),
2331                             curr_active_slave->last_link_up))
2332                 bond_validate_arp(bond, slave, tip, sip);
2333
2334 out_unlock:
2335         if (arp != (struct arphdr *)skb->data)
2336                 kfree(arp);
2337         return RX_HANDLER_ANOTHER;
2338 }
2339
2340 /* function to verify if we're in the arp_interval timeslice, returns true if
2341  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2342  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2343  */
2344 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2345                                   int mod)
2346 {
2347         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2348
2349         return time_in_range(jiffies,
2350                              last_act - delta_in_ticks,
2351                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2352 }
2353
2354 /* This function is called regularly to monitor each slave's link
2355  * ensuring that traffic is being sent and received when arp monitoring
2356  * is used in load-balancing mode. if the adapter has been dormant, then an
2357  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2358  * arp monitoring in active backup mode.
2359  */
2360 static void bond_loadbalance_arp_mon(struct work_struct *work)
2361 {
2362         struct bonding *bond = container_of(work, struct bonding,
2363                                             arp_work.work);
2364         struct slave *slave, *oldcurrent;
2365         struct list_head *iter;
2366         int do_failover = 0, slave_state_changed = 0;
2367
2368         if (!bond_has_slaves(bond))
2369                 goto re_arm;
2370
2371         rcu_read_lock();
2372
2373         oldcurrent = rcu_dereference(bond->curr_active_slave);
2374         /* see if any of the previous devices are up now (i.e. they have
2375          * xmt and rcv traffic). the curr_active_slave does not come into
2376          * the picture unless it is null. also, slave->last_link_up is not
2377          * needed here because we send an arp on each slave and give a slave
2378          * as long as it needs to get the tx/rx within the delta.
2379          * TODO: what about up/down delay in arp mode? it wasn't here before
2380          *       so it can wait
2381          */
2382         bond_for_each_slave_rcu(bond, slave, iter) {
2383                 unsigned long trans_start = dev_trans_start(slave->dev);
2384
2385                 if (slave->link != BOND_LINK_UP) {
2386                         if (bond_time_in_interval(bond, trans_start, 1) &&
2387                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2388
2389                                 slave->link  = BOND_LINK_UP;
2390                                 slave_state_changed = 1;
2391
2392                                 /* primary_slave has no meaning in round-robin
2393                                  * mode. the window of a slave being up and
2394                                  * curr_active_slave being null after enslaving
2395                                  * is closed.
2396                                  */
2397                                 if (!oldcurrent) {
2398                                         netdev_info(bond->dev, "link status definitely up for interface %s\n",
2399                                                     slave->dev->name);
2400                                         do_failover = 1;
2401                                 } else {
2402                                         netdev_info(bond->dev, "interface %s is now up\n",
2403                                                     slave->dev->name);
2404                                 }
2405                         }
2406                 } else {
2407                         /* slave->link == BOND_LINK_UP */
2408
2409                         /* not all switches will respond to an arp request
2410                          * when the source ip is 0, so don't take the link down
2411                          * if we don't know our ip yet
2412                          */
2413                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2414                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2415
2416                                 slave->link  = BOND_LINK_DOWN;
2417                                 slave_state_changed = 1;
2418
2419                                 if (slave->link_failure_count < UINT_MAX)
2420                                         slave->link_failure_count++;
2421
2422                                 netdev_info(bond->dev, "interface %s is now down\n",
2423                                             slave->dev->name);
2424
2425                                 if (slave == oldcurrent)
2426                                         do_failover = 1;
2427                         }
2428                 }
2429
2430                 /* note: if switch is in round-robin mode, all links
2431                  * must tx arp to ensure all links rx an arp - otherwise
2432                  * links may oscillate or not come up at all; if switch is
2433                  * in something like xor mode, there is nothing we can
2434                  * do - all replies will be rx'ed on same link causing slaves
2435                  * to be unstable during low/no traffic periods
2436                  */
2437                 if (bond_slave_is_up(slave))
2438                         bond_arp_send_all(bond, slave);
2439         }
2440
2441         rcu_read_unlock();
2442
2443         if (do_failover || slave_state_changed) {
2444                 if (!rtnl_trylock())
2445                         goto re_arm;
2446
2447                 if (slave_state_changed) {
2448                         bond_slave_state_change(bond);
2449                 } else if (do_failover) {
2450                         block_netpoll_tx();
2451                         bond_select_active_slave(bond);
2452                         unblock_netpoll_tx();
2453                 }
2454                 rtnl_unlock();
2455         }
2456
2457 re_arm:
2458         if (bond->params.arp_interval)
2459                 queue_delayed_work(bond->wq, &bond->arp_work,
2460                                    msecs_to_jiffies(bond->params.arp_interval));
2461 }
2462
2463 /* Called to inspect slaves for active-backup mode ARP monitor link state
2464  * changes.  Sets new_link in slaves to specify what action should take
2465  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2466  * to link states must be committed.
2467  *
2468  * Called with rcu_read_lock held.
2469  */
2470 static int bond_ab_arp_inspect(struct bonding *bond)
2471 {
2472         unsigned long trans_start, last_rx;
2473         struct list_head *iter;
2474         struct slave *slave;
2475         int commit = 0;
2476
2477         bond_for_each_slave_rcu(bond, slave, iter) {
2478                 slave->new_link = BOND_LINK_NOCHANGE;
2479                 last_rx = slave_last_rx(bond, slave);
2480
2481                 if (slave->link != BOND_LINK_UP) {
2482                         if (bond_time_in_interval(bond, last_rx, 1)) {
2483                                 slave->new_link = BOND_LINK_UP;
2484                                 commit++;
2485                         }
2486                         continue;
2487                 }
2488
2489                 /* Give slaves 2*delta after being enslaved or made
2490                  * active.  This avoids bouncing, as the last receive
2491                  * times need a full ARP monitor cycle to be updated.
2492                  */
2493                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2494                         continue;
2495
2496                 /* Backup slave is down if:
2497                  * - No current_arp_slave AND
2498                  * - more than 3*delta since last receive AND
2499                  * - the bond has an IP address
2500                  *
2501                  * Note: a non-null current_arp_slave indicates
2502                  * the curr_active_slave went down and we are
2503                  * searching for a new one; under this condition
2504                  * we only take the curr_active_slave down - this
2505                  * gives each slave a chance to tx/rx traffic
2506                  * before being taken out
2507                  */
2508                 if (!bond_is_active_slave(slave) &&
2509                     !rcu_access_pointer(bond->current_arp_slave) &&
2510                     !bond_time_in_interval(bond, last_rx, 3)) {
2511                         slave->new_link = BOND_LINK_DOWN;
2512                         commit++;
2513                 }
2514
2515                 /* Active slave is down if:
2516                  * - more than 2*delta since transmitting OR
2517                  * - (more than 2*delta since receive AND
2518                  *    the bond has an IP address)
2519                  */
2520                 trans_start = dev_trans_start(slave->dev);
2521                 if (bond_is_active_slave(slave) &&
2522                     (!bond_time_in_interval(bond, trans_start, 2) ||
2523                      !bond_time_in_interval(bond, last_rx, 2))) {
2524                         slave->new_link = BOND_LINK_DOWN;
2525                         commit++;
2526                 }
2527         }
2528
2529         return commit;
2530 }
2531
2532 /* Called to commit link state changes noted by inspection step of
2533  * active-backup mode ARP monitor.
2534  *
2535  * Called with RTNL hold.
2536  */
2537 static void bond_ab_arp_commit(struct bonding *bond)
2538 {
2539         unsigned long trans_start;
2540         struct list_head *iter;
2541         struct slave *slave;
2542
2543         bond_for_each_slave(bond, slave, iter) {
2544                 switch (slave->new_link) {
2545                 case BOND_LINK_NOCHANGE:
2546                         continue;
2547
2548                 case BOND_LINK_UP:
2549                         trans_start = dev_trans_start(slave->dev);
2550                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
2551                             (!rtnl_dereference(bond->curr_active_slave) &&
2552                              bond_time_in_interval(bond, trans_start, 1))) {
2553                                 struct slave *current_arp_slave;
2554
2555                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2556                                 slave->link = BOND_LINK_UP;
2557                                 if (current_arp_slave) {
2558                                         bond_set_slave_inactive_flags(
2559                                                 current_arp_slave,
2560                                                 BOND_SLAVE_NOTIFY_NOW);
2561                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2562                                 }
2563
2564                                 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2565                                             slave->dev->name);
2566
2567                                 if (!rtnl_dereference(bond->curr_active_slave) ||
2568                                     slave == rtnl_dereference(bond->primary_slave))
2569                                         goto do_failover;
2570
2571                         }
2572
2573                         continue;
2574
2575                 case BOND_LINK_DOWN:
2576                         if (slave->link_failure_count < UINT_MAX)
2577                                 slave->link_failure_count++;
2578
2579                         slave->link = BOND_LINK_DOWN;
2580                         bond_set_slave_inactive_flags(slave,
2581                                                       BOND_SLAVE_NOTIFY_NOW);
2582
2583                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2584                                     slave->dev->name);
2585
2586                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
2587                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2588                                 goto do_failover;
2589                         }
2590
2591                         continue;
2592
2593                 default:
2594                         netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2595                                    slave->new_link, slave->dev->name);
2596                         continue;
2597                 }
2598
2599 do_failover:
2600                 block_netpoll_tx();
2601                 bond_select_active_slave(bond);
2602                 unblock_netpoll_tx();
2603         }
2604
2605         bond_set_carrier(bond);
2606 }
2607
2608 /* Send ARP probes for active-backup mode ARP monitor.
2609  *
2610  * Called with rcu_read_lock held.
2611  */
2612 static bool bond_ab_arp_probe(struct bonding *bond)
2613 {
2614         struct slave *slave, *before = NULL, *new_slave = NULL,
2615                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2616                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2617         struct list_head *iter;
2618         bool found = false;
2619         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2620
2621         if (curr_arp_slave && curr_active_slave)
2622                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2623                             curr_arp_slave->dev->name,
2624                             curr_active_slave->dev->name);
2625
2626         if (curr_active_slave) {
2627                 bond_arp_send_all(bond, curr_active_slave);
2628                 return should_notify_rtnl;
2629         }
2630
2631         /* if we don't have a curr_active_slave, search for the next available
2632          * backup slave from the current_arp_slave and make it the candidate
2633          * for becoming the curr_active_slave
2634          */
2635
2636         if (!curr_arp_slave) {
2637                 curr_arp_slave = bond_first_slave_rcu(bond);
2638                 if (!curr_arp_slave)
2639                         return should_notify_rtnl;
2640         }
2641
2642         bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2643
2644         bond_for_each_slave_rcu(bond, slave, iter) {
2645                 if (!found && !before && bond_slave_is_up(slave))
2646                         before = slave;
2647
2648                 if (found && !new_slave && bond_slave_is_up(slave))
2649                         new_slave = slave;
2650                 /* if the link state is up at this point, we
2651                  * mark it down - this can happen if we have
2652                  * simultaneous link failures and
2653                  * reselect_active_interface doesn't make this
2654                  * one the current slave so it is still marked
2655                  * up when it is actually down
2656                  */
2657                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2658                         slave->link = BOND_LINK_DOWN;
2659                         if (slave->link_failure_count < UINT_MAX)
2660                                 slave->link_failure_count++;
2661
2662                         bond_set_slave_inactive_flags(slave,
2663                                                       BOND_SLAVE_NOTIFY_LATER);
2664
2665                         netdev_info(bond->dev, "backup interface %s is now down\n",
2666                                     slave->dev->name);
2667                 }
2668                 if (slave == curr_arp_slave)
2669                         found = true;
2670         }
2671
2672         if (!new_slave && before)
2673                 new_slave = before;
2674
2675         if (!new_slave)
2676                 goto check_state;
2677
2678         new_slave->link = BOND_LINK_BACK;
2679         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2680         bond_arp_send_all(bond, new_slave);
2681         new_slave->last_link_up = jiffies;
2682         rcu_assign_pointer(bond->current_arp_slave, new_slave);
2683
2684 check_state:
2685         bond_for_each_slave_rcu(bond, slave, iter) {
2686                 if (slave->should_notify) {
2687                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2688                         break;
2689                 }
2690         }
2691         return should_notify_rtnl;
2692 }
2693
2694 static void bond_activebackup_arp_mon(struct work_struct *work)
2695 {
2696         struct bonding *bond = container_of(work, struct bonding,
2697                                             arp_work.work);
2698         bool should_notify_peers = false;
2699         bool should_notify_rtnl = false;
2700         int delta_in_ticks;
2701
2702         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2703
2704         if (!bond_has_slaves(bond))
2705                 goto re_arm;
2706
2707         rcu_read_lock();
2708
2709         should_notify_peers = bond_should_notify_peers(bond);
2710
2711         if (bond_ab_arp_inspect(bond)) {
2712                 rcu_read_unlock();
2713
2714                 /* Race avoidance with bond_close flush of workqueue */
2715                 if (!rtnl_trylock()) {
2716                         delta_in_ticks = 1;
2717                         should_notify_peers = false;
2718                         goto re_arm;
2719                 }
2720
2721                 bond_ab_arp_commit(bond);
2722
2723                 rtnl_unlock();
2724                 rcu_read_lock();
2725         }
2726
2727         should_notify_rtnl = bond_ab_arp_probe(bond);
2728         rcu_read_unlock();
2729
2730 re_arm:
2731         if (bond->params.arp_interval)
2732                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2733
2734         if (should_notify_peers || should_notify_rtnl) {
2735                 if (!rtnl_trylock())
2736                         return;
2737
2738                 if (should_notify_peers)
2739                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2740                                                  bond->dev);
2741                 if (should_notify_rtnl)
2742                         bond_slave_state_notify(bond);
2743
2744                 rtnl_unlock();
2745         }
2746 }
2747
2748 /*-------------------------- netdev event handling --------------------------*/
2749
2750 /* Change device name */
2751 static int bond_event_changename(struct bonding *bond)
2752 {
2753         bond_remove_proc_entry(bond);
2754         bond_create_proc_entry(bond);
2755
2756         bond_debug_reregister(bond);
2757
2758         return NOTIFY_DONE;
2759 }
2760
2761 static int bond_master_netdev_event(unsigned long event,
2762                                     struct net_device *bond_dev)
2763 {
2764         struct bonding *event_bond = netdev_priv(bond_dev);
2765
2766         switch (event) {
2767         case NETDEV_CHANGENAME:
2768                 return bond_event_changename(event_bond);
2769         case NETDEV_UNREGISTER:
2770                 bond_remove_proc_entry(event_bond);
2771                 break;
2772         case NETDEV_REGISTER:
2773                 bond_create_proc_entry(event_bond);
2774                 break;
2775         case NETDEV_NOTIFY_PEERS:
2776                 if (event_bond->send_peer_notif)
2777                         event_bond->send_peer_notif--;
2778                 break;
2779         default:
2780                 break;
2781         }
2782
2783         return NOTIFY_DONE;
2784 }
2785
2786 static int bond_slave_netdev_event(unsigned long event,
2787                                    struct net_device *slave_dev)
2788 {
2789         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
2790         struct bonding *bond;
2791         struct net_device *bond_dev;
2792         u32 old_speed;
2793         u8 old_duplex;
2794
2795         /* A netdev event can be generated while enslaving a device
2796          * before netdev_rx_handler_register is called in which case
2797          * slave will be NULL
2798          */
2799         if (!slave)
2800                 return NOTIFY_DONE;
2801         bond_dev = slave->bond->dev;
2802         bond = slave->bond;
2803         primary = rtnl_dereference(bond->primary_slave);
2804
2805         switch (event) {
2806         case NETDEV_UNREGISTER:
2807                 if (bond_dev->type != ARPHRD_ETHER)
2808                         bond_release_and_destroy(bond_dev, slave_dev);
2809                 else
2810                         bond_release(bond_dev, slave_dev);
2811                 break;
2812         case NETDEV_UP:
2813         case NETDEV_CHANGE:
2814                 old_speed = slave->speed;
2815                 old_duplex = slave->duplex;
2816
2817                 bond_update_speed_duplex(slave);
2818
2819                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2820                         if (old_speed != slave->speed)
2821                                 bond_3ad_adapter_speed_changed(slave);
2822                         if (old_duplex != slave->duplex)
2823                                 bond_3ad_adapter_duplex_changed(slave);
2824                 }
2825                 break;
2826         case NETDEV_DOWN:
2827                 break;
2828         case NETDEV_CHANGEMTU:
2829                 /* TODO: Should slaves be allowed to
2830                  * independently alter their MTU?  For
2831                  * an active-backup bond, slaves need
2832                  * not be the same type of device, so
2833                  * MTUs may vary.  For other modes,
2834                  * slaves arguably should have the
2835                  * same MTUs. To do this, we'd need to
2836                  * take over the slave's change_mtu
2837                  * function for the duration of their
2838                  * servitude.
2839                  */
2840                 break;
2841         case NETDEV_CHANGENAME:
2842                 /* we don't care if we don't have primary set */
2843                 if (!bond_uses_primary(bond) ||
2844                     !bond->params.primary[0])
2845                         break;
2846
2847                 if (slave == primary) {
2848                         /* slave's name changed - he's no longer primary */
2849                         RCU_INIT_POINTER(bond->primary_slave, NULL);
2850                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2851                         /* we have a new primary slave */
2852                         rcu_assign_pointer(bond->primary_slave, slave);
2853                 } else { /* we didn't change primary - exit */
2854                         break;
2855                 }
2856
2857                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
2858                             primary ? slave_dev->name : "none");
2859
2860                 block_netpoll_tx();
2861                 bond_select_active_slave(bond);
2862                 unblock_netpoll_tx();
2863                 break;
2864         case NETDEV_FEAT_CHANGE:
2865                 bond_compute_features(bond);
2866                 break;
2867         case NETDEV_RESEND_IGMP:
2868                 /* Propagate to master device */
2869                 call_netdevice_notifiers(event, slave->bond->dev);
2870                 break;
2871         default:
2872                 break;
2873         }
2874
2875         return NOTIFY_DONE;
2876 }
2877
2878 /* bond_netdev_event: handle netdev notifier chain events.
2879  *
2880  * This function receives events for the netdev chain.  The caller (an
2881  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2882  * locks for us to safely manipulate the slave devices (RTNL lock,
2883  * dev_probe_lock).
2884  */
2885 static int bond_netdev_event(struct notifier_block *this,
2886                              unsigned long event, void *ptr)
2887 {
2888         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2889
2890         netdev_dbg(event_dev, "event: %lx\n", event);
2891
2892         if (!(event_dev->priv_flags & IFF_BONDING))
2893                 return NOTIFY_DONE;
2894
2895         if (event_dev->flags & IFF_MASTER) {
2896                 netdev_dbg(event_dev, "IFF_MASTER\n");
2897                 return bond_master_netdev_event(event, event_dev);
2898         }
2899
2900         if (event_dev->flags & IFF_SLAVE) {
2901                 netdev_dbg(event_dev, "IFF_SLAVE\n");
2902                 return bond_slave_netdev_event(event, event_dev);
2903         }
2904
2905         return NOTIFY_DONE;
2906 }
2907
2908 static struct notifier_block bond_netdev_notifier = {
2909         .notifier_call = bond_netdev_event,
2910 };
2911
2912 /*---------------------------- Hashing Policies -----------------------------*/
2913
2914 /* L2 hash helper */
2915 static inline u32 bond_eth_hash(struct sk_buff *skb)
2916 {
2917         struct ethhdr *ep, hdr_tmp;
2918
2919         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
2920         if (ep)
2921                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
2922         return 0;
2923 }
2924
2925 /* Extract the appropriate headers based on bond's xmit policy */
2926 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
2927                               struct flow_keys *fk)
2928 {
2929         const struct ipv6hdr *iph6;
2930         const struct iphdr *iph;
2931         int noff, proto = -1;
2932
2933         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
2934                 return skb_flow_dissect(skb, fk);
2935
2936         fk->ports = 0;
2937         noff = skb_network_offset(skb);
2938         if (skb->protocol == htons(ETH_P_IP)) {
2939                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
2940                         return false;
2941                 iph = ip_hdr(skb);
2942                 fk->src = iph->saddr;
2943                 fk->dst = iph->daddr;
2944                 noff += iph->ihl << 2;
2945                 if (!ip_is_fragment(iph))
2946                         proto = iph->protocol;
2947         } else if (skb->protocol == htons(ETH_P_IPV6)) {
2948                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
2949                         return false;
2950                 iph6 = ipv6_hdr(skb);
2951                 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
2952                 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
2953                 noff += sizeof(*iph6);
2954                 proto = iph6->nexthdr;
2955         } else {
2956                 return false;
2957         }
2958         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
2959                 fk->ports = skb_flow_get_ports(skb, noff, proto);
2960
2961         return true;
2962 }
2963
2964 /**
2965  * bond_xmit_hash - generate a hash value based on the xmit policy
2966  * @bond: bonding device
2967  * @skb: buffer to use for headers
2968  *
2969  * This function will extract the necessary headers from the skb buffer and use
2970  * them to generate a hash based on the xmit_policy set in the bonding device
2971  */
2972 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
2973 {
2974         struct flow_keys flow;
2975         u32 hash;
2976
2977         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
2978             !bond_flow_dissect(bond, skb, &flow))
2979                 return bond_eth_hash(skb);
2980
2981         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
2982             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
2983                 hash = bond_eth_hash(skb);
2984         else
2985                 hash = (__force u32)flow.ports;
2986         hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
2987         hash ^= (hash >> 16);
2988         hash ^= (hash >> 8);
2989
2990         return hash;
2991 }
2992
2993 /*-------------------------- Device entry points ----------------------------*/
2994
2995 static void bond_work_init_all(struct bonding *bond)
2996 {
2997         INIT_DELAYED_WORK(&bond->mcast_work,
2998                           bond_resend_igmp_join_requests_delayed);
2999         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3000         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3001         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3002                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3003         else
3004                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3005         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3006 }
3007
3008 static void bond_work_cancel_all(struct bonding *bond)
3009 {
3010         cancel_delayed_work_sync(&bond->mii_work);
3011         cancel_delayed_work_sync(&bond->arp_work);
3012         cancel_delayed_work_sync(&bond->alb_work);
3013         cancel_delayed_work_sync(&bond->ad_work);
3014         cancel_delayed_work_sync(&bond->mcast_work);
3015 }
3016
3017 static int bond_open(struct net_device *bond_dev)
3018 {
3019         struct bonding *bond = netdev_priv(bond_dev);
3020         struct list_head *iter;
3021         struct slave *slave;
3022
3023         /* reset slave->backup and slave->inactive */
3024         if (bond_has_slaves(bond)) {
3025                 bond_for_each_slave(bond, slave, iter) {
3026                         if (bond_uses_primary(bond) &&
3027                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3028                                 bond_set_slave_inactive_flags(slave,
3029                                                               BOND_SLAVE_NOTIFY_NOW);
3030                         } else {
3031                                 bond_set_slave_active_flags(slave,
3032                                                             BOND_SLAVE_NOTIFY_NOW);
3033                         }
3034                 }
3035         }
3036
3037         bond_work_init_all(bond);
3038
3039         if (bond_is_lb(bond)) {
3040                 /* bond_alb_initialize must be called before the timer
3041                  * is started.
3042                  */
3043                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3044                         return -ENOMEM;
3045                 if (bond->params.tlb_dynamic_lb)
3046                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3047         }
3048
3049         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3050                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3051
3052         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3053                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3054                 bond->recv_probe = bond_arp_rcv;
3055         }
3056
3057         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3058                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3059                 /* register to receive LACPDUs */
3060                 bond->recv_probe = bond_3ad_lacpdu_recv;
3061                 bond_3ad_initiate_agg_selection(bond, 1);
3062         }
3063
3064         return 0;
3065 }
3066
3067 static int bond_close(struct net_device *bond_dev)
3068 {
3069         struct bonding *bond = netdev_priv(bond_dev);
3070
3071         bond_work_cancel_all(bond);
3072         bond->send_peer_notif = 0;
3073         if (bond_is_lb(bond))
3074                 bond_alb_deinitialize(bond);
3075         bond->recv_probe = NULL;
3076
3077         return 0;
3078 }
3079
3080 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3081                                                 struct rtnl_link_stats64 *stats)
3082 {
3083         struct bonding *bond = netdev_priv(bond_dev);
3084         struct rtnl_link_stats64 temp;
3085         struct list_head *iter;
3086         struct slave *slave;
3087
3088         memset(stats, 0, sizeof(*stats));
3089
3090         bond_for_each_slave(bond, slave, iter) {
3091                 const struct rtnl_link_stats64 *sstats =
3092                         dev_get_stats(slave->dev, &temp);
3093
3094                 stats->rx_packets += sstats->rx_packets;
3095                 stats->rx_bytes += sstats->rx_bytes;
3096                 stats->rx_errors += sstats->rx_errors;
3097                 stats->rx_dropped += sstats->rx_dropped;
3098
3099                 stats->tx_packets += sstats->tx_packets;
3100                 stats->tx_bytes += sstats->tx_bytes;
3101                 stats->tx_errors += sstats->tx_errors;
3102                 stats->tx_dropped += sstats->tx_dropped;
3103
3104                 stats->multicast += sstats->multicast;
3105                 stats->collisions += sstats->collisions;
3106
3107                 stats->rx_length_errors += sstats->rx_length_errors;
3108                 stats->rx_over_errors += sstats->rx_over_errors;
3109                 stats->rx_crc_errors += sstats->rx_crc_errors;
3110                 stats->rx_frame_errors += sstats->rx_frame_errors;
3111                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3112                 stats->rx_missed_errors += sstats->rx_missed_errors;
3113
3114                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3115                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3116                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3117                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3118                 stats->tx_window_errors += sstats->tx_window_errors;
3119         }
3120
3121         return stats;
3122 }
3123
3124 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3125 {
3126         struct bonding *bond = netdev_priv(bond_dev);
3127         struct net_device *slave_dev = NULL;
3128         struct ifbond k_binfo;
3129         struct ifbond __user *u_binfo = NULL;
3130         struct ifslave k_sinfo;
3131         struct ifslave __user *u_sinfo = NULL;
3132         struct mii_ioctl_data *mii = NULL;
3133         struct bond_opt_value newval;
3134         struct net *net;
3135         int res = 0;
3136
3137         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3138
3139         switch (cmd) {
3140         case SIOCGMIIPHY:
3141                 mii = if_mii(ifr);
3142                 if (!mii)
3143                         return -EINVAL;
3144
3145                 mii->phy_id = 0;
3146                 /* Fall Through */
3147         case SIOCGMIIREG:
3148                 /* We do this again just in case we were called by SIOCGMIIREG
3149                  * instead of SIOCGMIIPHY.
3150                  */
3151                 mii = if_mii(ifr);
3152                 if (!mii)
3153                         return -EINVAL;
3154
3155                 if (mii->reg_num == 1) {
3156                         mii->val_out = 0;
3157                         if (netif_carrier_ok(bond->dev))
3158                                 mii->val_out = BMSR_LSTATUS;
3159                 }
3160
3161                 return 0;
3162         case BOND_INFO_QUERY_OLD:
3163         case SIOCBONDINFOQUERY:
3164                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3165
3166                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3167                         return -EFAULT;
3168
3169                 res = bond_info_query(bond_dev, &k_binfo);
3170                 if (res == 0 &&
3171                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3172                         return -EFAULT;
3173
3174                 return res;
3175         case BOND_SLAVE_INFO_QUERY_OLD:
3176         case SIOCBONDSLAVEINFOQUERY:
3177                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3178
3179                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3180                         return -EFAULT;
3181
3182                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3183                 if (res == 0 &&
3184                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3185                         return -EFAULT;
3186
3187                 return res;
3188         default:
3189                 break;
3190         }
3191
3192         net = dev_net(bond_dev);
3193
3194         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3195                 return -EPERM;
3196
3197         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3198
3199         netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3200
3201         if (!slave_dev)
3202                 return -ENODEV;
3203
3204         netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3205         switch (cmd) {
3206         case BOND_ENSLAVE_OLD:
3207         case SIOCBONDENSLAVE:
3208                 res = bond_enslave(bond_dev, slave_dev);
3209                 break;
3210         case BOND_RELEASE_OLD:
3211         case SIOCBONDRELEASE:
3212                 res = bond_release(bond_dev, slave_dev);
3213                 break;
3214         case BOND_SETHWADDR_OLD:
3215         case SIOCBONDSETHWADDR:
3216                 bond_set_dev_addr(bond_dev, slave_dev);
3217                 res = 0;
3218                 break;
3219         case BOND_CHANGE_ACTIVE_OLD:
3220         case SIOCBONDCHANGEACTIVE:
3221                 bond_opt_initstr(&newval, slave_dev->name);
3222                 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3223                 break;
3224         default:
3225                 res = -EOPNOTSUPP;
3226         }
3227
3228         return res;
3229 }
3230
3231 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3232 {
3233         struct bonding *bond = netdev_priv(bond_dev);
3234
3235         if (change & IFF_PROMISC)
3236                 bond_set_promiscuity(bond,
3237                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3238
3239         if (change & IFF_ALLMULTI)
3240                 bond_set_allmulti(bond,
3241                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3242 }
3243
3244 static void bond_set_rx_mode(struct net_device *bond_dev)
3245 {
3246         struct bonding *bond = netdev_priv(bond_dev);
3247         struct list_head *iter;
3248         struct slave *slave;
3249
3250         rcu_read_lock();
3251         if (bond_uses_primary(bond)) {
3252                 slave = rcu_dereference(bond->curr_active_slave);
3253                 if (slave) {
3254                         dev_uc_sync(slave->dev, bond_dev);
3255                         dev_mc_sync(slave->dev, bond_dev);
3256                 }
3257         } else {
3258                 bond_for_each_slave_rcu(bond, slave, iter) {
3259                         dev_uc_sync_multiple(slave->dev, bond_dev);
3260                         dev_mc_sync_multiple(slave->dev, bond_dev);
3261                 }
3262         }
3263         rcu_read_unlock();
3264 }
3265
3266 static int bond_neigh_init(struct neighbour *n)
3267 {
3268         struct bonding *bond = netdev_priv(n->dev);
3269         const struct net_device_ops *slave_ops;
3270         struct neigh_parms parms;
3271         struct slave *slave;
3272         int ret;
3273
3274         slave = bond_first_slave(bond);
3275         if (!slave)
3276                 return 0;
3277         slave_ops = slave->dev->netdev_ops;
3278         if (!slave_ops->ndo_neigh_setup)
3279                 return 0;
3280
3281         parms.neigh_setup = NULL;
3282         parms.neigh_cleanup = NULL;
3283         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3284         if (ret)
3285                 return ret;
3286
3287         /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3288          * after the last slave has been detached.  Assumes that all slaves
3289          * utilize the same neigh_cleanup (true at this writing as only user
3290          * is ipoib).
3291          */
3292         n->parms->neigh_cleanup = parms.neigh_cleanup;
3293
3294         if (!parms.neigh_setup)
3295                 return 0;
3296
3297         return parms.neigh_setup(n);
3298 }
3299
3300 /* The bonding ndo_neigh_setup is called at init time beofre any
3301  * slave exists. So we must declare proxy setup function which will
3302  * be used at run time to resolve the actual slave neigh param setup.
3303  *
3304  * It's also called by master devices (such as vlans) to setup their
3305  * underlying devices. In that case - do nothing, we're already set up from
3306  * our init.
3307  */
3308 static int bond_neigh_setup(struct net_device *dev,
3309                             struct neigh_parms *parms)
3310 {
3311         /* modify only our neigh_parms */
3312         if (parms->dev == dev)
3313                 parms->neigh_setup = bond_neigh_init;
3314
3315         return 0;
3316 }
3317
3318 /* Change the MTU of all of a master's slaves to match the master */
3319 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3320 {
3321         struct bonding *bond = netdev_priv(bond_dev);
3322         struct slave *slave, *rollback_slave;
3323         struct list_head *iter;
3324         int res = 0;
3325
3326         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3327
3328         bond_for_each_slave(bond, slave, iter) {
3329                 netdev_dbg(bond_dev, "s %p c_m %p\n",
3330                            slave, slave->dev->netdev_ops->ndo_change_mtu);
3331
3332                 res = dev_set_mtu(slave->dev, new_mtu);
3333
3334                 if (res) {
3335                         /* If we failed to set the slave's mtu to the new value
3336                          * we must abort the operation even in ACTIVE_BACKUP
3337                          * mode, because if we allow the backup slaves to have
3338                          * different mtu values than the active slave we'll
3339                          * need to change their mtu when doing a failover. That
3340                          * means changing their mtu from timer context, which
3341                          * is probably not a good idea.
3342                          */
3343                         netdev_dbg(bond_dev, "err %d %s\n", res,
3344                                    slave->dev->name);
3345                         goto unwind;
3346                 }
3347         }
3348
3349         bond_dev->mtu = new_mtu;
3350
3351         return 0;
3352
3353 unwind:
3354         /* unwind from head to the slave that failed */
3355         bond_for_each_slave(bond, rollback_slave, iter) {
3356                 int tmp_res;
3357
3358                 if (rollback_slave == slave)
3359                         break;
3360
3361                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3362                 if (tmp_res) {
3363                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3364                                    tmp_res, rollback_slave->dev->name);
3365                 }
3366         }
3367
3368         return res;
3369 }
3370
3371 /* Change HW address
3372  *
3373  * Note that many devices must be down to change the HW address, and
3374  * downing the master releases all slaves.  We can make bonds full of
3375  * bonding devices to test this, however.
3376  */
3377 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3378 {
3379         struct bonding *bond = netdev_priv(bond_dev);
3380         struct slave *slave, *rollback_slave;
3381         struct sockaddr *sa = addr, tmp_sa;
3382         struct list_head *iter;
3383         int res = 0;
3384
3385         if (BOND_MODE(bond) == BOND_MODE_ALB)
3386                 return bond_alb_set_mac_address(bond_dev, addr);
3387
3388
3389         netdev_dbg(bond_dev, "bond=%p\n", bond);
3390
3391         /* If fail_over_mac is enabled, do nothing and return success.
3392          * Returning an error causes ifenslave to fail.
3393          */
3394         if (bond->params.fail_over_mac &&
3395             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3396                 return 0;
3397
3398         if (!is_valid_ether_addr(sa->sa_data))
3399                 return -EADDRNOTAVAIL;
3400
3401         bond_for_each_slave(bond, slave, iter) {
3402                 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3403                 res = dev_set_mac_address(slave->dev, addr);
3404                 if (res) {
3405                         /* TODO: consider downing the slave
3406                          * and retry ?
3407                          * User should expect communications
3408                          * breakage anyway until ARP finish
3409                          * updating, so...
3410                          */
3411                         netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3412                         goto unwind;
3413                 }
3414         }
3415
3416         /* success */
3417         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3418         return 0;
3419
3420 unwind:
3421         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3422         tmp_sa.sa_family = bond_dev->type;
3423
3424         /* unwind from head to the slave that failed */
3425         bond_for_each_slave(bond, rollback_slave, iter) {
3426                 int tmp_res;
3427
3428                 if (rollback_slave == slave)
3429                         break;
3430
3431                 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3432                 if (tmp_res) {
3433                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3434                                    tmp_res, rollback_slave->dev->name);
3435                 }
3436         }
3437
3438         return res;
3439 }
3440
3441 /**
3442  * bond_xmit_slave_id - transmit skb through slave with slave_id
3443  * @bond: bonding device that is transmitting
3444  * @skb: buffer to transmit
3445  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3446  *
3447  * This function tries to transmit through slave with slave_id but in case
3448  * it fails, it tries to find the first available slave for transmission.
3449  * The skb is consumed in all cases, thus the function is void.
3450  */
3451 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3452 {
3453         struct list_head *iter;
3454         struct slave *slave;
3455         int i = slave_id;
3456
3457         /* Here we start from the slave with slave_id */
3458         bond_for_each_slave_rcu(bond, slave, iter) {
3459                 if (--i < 0) {
3460                         if (bond_slave_can_tx(slave)) {
3461                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3462                                 return;
3463                         }
3464                 }
3465         }
3466
3467         /* Here we start from the first slave up to slave_id */
3468         i = slave_id;
3469         bond_for_each_slave_rcu(bond, slave, iter) {
3470                 if (--i < 0)
3471                         break;
3472                 if (bond_slave_can_tx(slave)) {
3473                         bond_dev_queue_xmit(bond, skb, slave->dev);
3474                         return;
3475                 }
3476         }
3477         /* no slave that can tx has been found */
3478         dev_kfree_skb_any(skb);
3479 }
3480
3481 /**
3482  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3483  * @bond: bonding device to use
3484  *
3485  * Based on the value of the bonding device's packets_per_slave parameter
3486  * this function generates a slave id, which is usually used as the next
3487  * slave to transmit through.
3488  */
3489 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3490 {
3491         u32 slave_id;
3492         struct reciprocal_value reciprocal_packets_per_slave;
3493         int packets_per_slave = bond->params.packets_per_slave;
3494
3495         switch (packets_per_slave) {
3496         case 0:
3497                 slave_id = prandom_u32();
3498                 break;
3499         case 1:
3500                 slave_id = bond->rr_tx_counter;
3501                 break;
3502         default:
3503                 reciprocal_packets_per_slave =
3504                         bond->params.reciprocal_packets_per_slave;
3505                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3506                                              reciprocal_packets_per_slave);
3507                 break;
3508         }
3509         bond->rr_tx_counter++;
3510
3511         return slave_id;
3512 }
3513
3514 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3515 {
3516         struct bonding *bond = netdev_priv(bond_dev);
3517         struct iphdr *iph = ip_hdr(skb);
3518         struct slave *slave;
3519         u32 slave_id;
3520
3521         /* Start with the curr_active_slave that joined the bond as the
3522          * default for sending IGMP traffic.  For failover purposes one
3523          * needs to maintain some consistency for the interface that will
3524          * send the join/membership reports.  The curr_active_slave found
3525          * will send all of this type of traffic.
3526          */
3527         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3528                 slave = rcu_dereference(bond->curr_active_slave);
3529                 if (slave)
3530                         bond_dev_queue_xmit(bond, skb, slave->dev);
3531                 else
3532                         bond_xmit_slave_id(bond, skb, 0);
3533         } else {
3534                 int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3535
3536                 if (likely(slave_cnt)) {
3537                         slave_id = bond_rr_gen_slave_id(bond);
3538                         bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3539                 } else {
3540                         dev_kfree_skb_any(skb);
3541                 }
3542         }
3543
3544         return NETDEV_TX_OK;
3545 }
3546
3547 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3548  * the bond has a usable interface.
3549  */
3550 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3551 {
3552         struct bonding *bond = netdev_priv(bond_dev);
3553         struct slave *slave;
3554
3555         slave = rcu_dereference(bond->curr_active_slave);
3556         if (slave)
3557                 bond_dev_queue_xmit(bond, skb, slave->dev);
3558         else
3559                 dev_kfree_skb_any(skb);
3560
3561         return NETDEV_TX_OK;
3562 }
3563
3564 /* In bond_xmit_xor() , we determine the output device by using a pre-
3565  * determined xmit_hash_policy(), If the selected device is not enabled,
3566  * find the next active slave.
3567  */
3568 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3569 {
3570         struct bonding *bond = netdev_priv(bond_dev);
3571         int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3572
3573         if (likely(slave_cnt))
3574                 bond_xmit_slave_id(bond, skb,
3575                                    bond_xmit_hash(bond, skb) % slave_cnt);
3576         else
3577                 dev_kfree_skb_any(skb);
3578
3579         return NETDEV_TX_OK;
3580 }
3581
3582 /* in broadcast mode, we send everything to all usable interfaces. */
3583 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3584 {
3585         struct bonding *bond = netdev_priv(bond_dev);
3586         struct slave *slave = NULL;
3587         struct list_head *iter;
3588
3589         bond_for_each_slave_rcu(bond, slave, iter) {
3590                 if (bond_is_last_slave(bond, slave))
3591                         break;
3592                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3593                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3594
3595                         if (!skb2) {
3596                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3597                                                     bond_dev->name, __func__);
3598                                 continue;
3599                         }
3600                         bond_dev_queue_xmit(bond, skb2, slave->dev);
3601                 }
3602         }
3603         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3604                 bond_dev_queue_xmit(bond, skb, slave->dev);
3605         else
3606                 dev_kfree_skb_any(skb);
3607
3608         return NETDEV_TX_OK;
3609 }
3610
3611 /*------------------------- Device initialization ---------------------------*/
3612
3613 /* Lookup the slave that corresponds to a qid */
3614 static inline int bond_slave_override(struct bonding *bond,
3615                                       struct sk_buff *skb)
3616 {
3617         struct slave *slave = NULL;
3618         struct list_head *iter;
3619
3620         if (!skb->queue_mapping)
3621                 return 1;
3622
3623         /* Find out if any slaves have the same mapping as this skb. */
3624         bond_for_each_slave_rcu(bond, slave, iter) {
3625                 if (slave->queue_id == skb->queue_mapping) {
3626                         if (bond_slave_can_tx(slave)) {
3627                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3628                                 return 0;
3629                         }
3630                         /* If the slave isn't UP, use default transmit policy. */
3631                         break;
3632                 }
3633         }
3634
3635         return 1;
3636 }
3637
3638
3639 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3640                              void *accel_priv, select_queue_fallback_t fallback)
3641 {
3642         /* This helper function exists to help dev_pick_tx get the correct
3643          * destination queue.  Using a helper function skips a call to
3644          * skb_tx_hash and will put the skbs in the queue we expect on their
3645          * way down to the bonding driver.
3646          */
3647         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3648
3649         /* Save the original txq to restore before passing to the driver */
3650         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3651
3652         if (unlikely(txq >= dev->real_num_tx_queues)) {
3653                 do {
3654                         txq -= dev->real_num_tx_queues;
3655                 } while (txq >= dev->real_num_tx_queues);
3656         }
3657         return txq;
3658 }
3659
3660 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3661 {
3662         struct bonding *bond = netdev_priv(dev);
3663
3664         if (bond_should_override_tx_queue(bond) &&
3665             !bond_slave_override(bond, skb))
3666                 return NETDEV_TX_OK;
3667
3668         switch (BOND_MODE(bond)) {
3669         case BOND_MODE_ROUNDROBIN:
3670                 return bond_xmit_roundrobin(skb, dev);
3671         case BOND_MODE_ACTIVEBACKUP:
3672                 return bond_xmit_activebackup(skb, dev);
3673         case BOND_MODE_XOR:
3674                 return bond_xmit_xor(skb, dev);
3675         case BOND_MODE_BROADCAST:
3676                 return bond_xmit_broadcast(skb, dev);
3677         case BOND_MODE_8023AD:
3678                 return bond_3ad_xmit_xor(skb, dev);
3679         case BOND_MODE_ALB:
3680                 return bond_alb_xmit(skb, dev);
3681         case BOND_MODE_TLB:
3682                 return bond_tlb_xmit(skb, dev);
3683         default:
3684                 /* Should never happen, mode already checked */
3685                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
3686                 WARN_ON_ONCE(1);
3687                 dev_kfree_skb_any(skb);
3688                 return NETDEV_TX_OK;
3689         }
3690 }
3691
3692 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3693 {
3694         struct bonding *bond = netdev_priv(dev);
3695         netdev_tx_t ret = NETDEV_TX_OK;
3696
3697         /* If we risk deadlock from transmitting this in the
3698          * netpoll path, tell netpoll to queue the frame for later tx
3699          */
3700         if (unlikely(is_netpoll_tx_blocked(dev)))
3701                 return NETDEV_TX_BUSY;
3702
3703         rcu_read_lock();
3704         if (bond_has_slaves(bond))
3705                 ret = __bond_start_xmit(skb, dev);
3706         else
3707                 dev_kfree_skb_any(skb);
3708         rcu_read_unlock();
3709
3710         return ret;
3711 }
3712
3713 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3714                                      struct ethtool_cmd *ecmd)
3715 {
3716         struct bonding *bond = netdev_priv(bond_dev);
3717         unsigned long speed = 0;
3718         struct list_head *iter;
3719         struct slave *slave;
3720
3721         ecmd->duplex = DUPLEX_UNKNOWN;
3722         ecmd->port = PORT_OTHER;
3723
3724         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
3725          * do not need to check mode.  Though link speed might not represent
3726          * the true receive or transmit bandwidth (not all modes are symmetric)
3727          * this is an accurate maximum.
3728          */
3729         bond_for_each_slave(bond, slave, iter) {
3730                 if (bond_slave_can_tx(slave)) {
3731                         if (slave->speed != SPEED_UNKNOWN)
3732                                 speed += slave->speed;
3733                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
3734                             slave->duplex != DUPLEX_UNKNOWN)
3735                                 ecmd->duplex = slave->duplex;
3736                 }
3737         }
3738         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3739
3740         return 0;
3741 }
3742
3743 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3744                                      struct ethtool_drvinfo *drvinfo)
3745 {
3746         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3747         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3748         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3749                  BOND_ABI_VERSION);
3750 }
3751
3752 static const struct ethtool_ops bond_ethtool_ops = {
3753         .get_drvinfo            = bond_ethtool_get_drvinfo,
3754         .get_settings           = bond_ethtool_get_settings,
3755         .get_link               = ethtool_op_get_link,
3756 };
3757
3758 static const struct net_device_ops bond_netdev_ops = {
3759         .ndo_init               = bond_init,
3760         .ndo_uninit             = bond_uninit,
3761         .ndo_open               = bond_open,
3762         .ndo_stop               = bond_close,
3763         .ndo_start_xmit         = bond_start_xmit,
3764         .ndo_select_queue       = bond_select_queue,
3765         .ndo_get_stats64        = bond_get_stats,
3766         .ndo_do_ioctl           = bond_do_ioctl,
3767         .ndo_change_rx_flags    = bond_change_rx_flags,
3768         .ndo_set_rx_mode        = bond_set_rx_mode,
3769         .ndo_change_mtu         = bond_change_mtu,
3770         .ndo_set_mac_address    = bond_set_mac_address,
3771         .ndo_neigh_setup        = bond_neigh_setup,
3772         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
3773         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
3774 #ifdef CONFIG_NET_POLL_CONTROLLER
3775         .ndo_netpoll_setup      = bond_netpoll_setup,
3776         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
3777         .ndo_poll_controller    = bond_poll_controller,
3778 #endif
3779         .ndo_add_slave          = bond_enslave,
3780         .ndo_del_slave          = bond_release,
3781         .ndo_fix_features       = bond_fix_features,
3782 };
3783
3784 static const struct device_type bond_type = {
3785         .name = "bond",
3786 };
3787
3788 static void bond_destructor(struct net_device *bond_dev)
3789 {
3790         struct bonding *bond = netdev_priv(bond_dev);
3791         if (bond->wq)
3792                 destroy_workqueue(bond->wq);
3793         free_netdev(bond_dev);
3794 }
3795
3796 void bond_setup(struct net_device *bond_dev)
3797 {
3798         struct bonding *bond = netdev_priv(bond_dev);
3799
3800         spin_lock_init(&bond->mode_lock);
3801         bond->params = bonding_defaults;
3802
3803         /* Initialize pointers */
3804         bond->dev = bond_dev;
3805
3806         /* Initialize the device entry points */
3807         ether_setup(bond_dev);
3808         bond_dev->netdev_ops = &bond_netdev_ops;
3809         bond_dev->ethtool_ops = &bond_ethtool_ops;
3810
3811         bond_dev->destructor = bond_destructor;
3812
3813         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
3814
3815         /* Initialize the device options */
3816         bond_dev->tx_queue_len = 0;
3817         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
3818         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT;
3819         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
3820
3821         /* don't acquire bond device's netif_tx_lock when transmitting */
3822         bond_dev->features |= NETIF_F_LLTX;
3823
3824         /* By default, we declare the bond to be fully
3825          * VLAN hardware accelerated capable. Special
3826          * care is taken in the various xmit functions
3827          * when there are slaves that are not hw accel
3828          * capable
3829          */
3830
3831         /* Don't allow bond devices to change network namespaces. */
3832         bond_dev->features |= NETIF_F_NETNS_LOCAL;
3833
3834         bond_dev->hw_features = BOND_VLAN_FEATURES |
3835                                 NETIF_F_HW_VLAN_CTAG_TX |
3836                                 NETIF_F_HW_VLAN_CTAG_RX |
3837                                 NETIF_F_HW_VLAN_CTAG_FILTER;
3838
3839         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
3840         bond_dev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
3841         bond_dev->features |= bond_dev->hw_features;
3842 }
3843
3844 /* Destroy a bonding device.
3845  * Must be under rtnl_lock when this function is called.
3846  */
3847 static void bond_uninit(struct net_device *bond_dev)
3848 {
3849         struct bonding *bond = netdev_priv(bond_dev);
3850         struct list_head *iter;
3851         struct slave *slave;
3852
3853         bond_netpoll_cleanup(bond_dev);
3854
3855         /* Release the bonded slaves */
3856         bond_for_each_slave(bond, slave, iter)
3857                 __bond_release_one(bond_dev, slave->dev, true);
3858         netdev_info(bond_dev, "Released all slaves\n");
3859
3860         list_del(&bond->bond_list);
3861
3862         bond_debug_unregister(bond);
3863 }
3864
3865 /*------------------------- Module initialization ---------------------------*/
3866
3867 static int bond_check_params(struct bond_params *params)
3868 {
3869         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
3870         struct bond_opt_value newval;
3871         const struct bond_opt_value *valptr;
3872         int arp_all_targets_value;
3873
3874         /* Convert string parameters. */
3875         if (mode) {
3876                 bond_opt_initstr(&newval, mode);
3877                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
3878                 if (!valptr) {
3879                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
3880                         return -EINVAL;
3881                 }
3882                 bond_mode = valptr->value;
3883         }
3884
3885         if (xmit_hash_policy) {
3886                 if ((bond_mode != BOND_MODE_XOR) &&
3887                     (bond_mode != BOND_MODE_8023AD) &&
3888                     (bond_mode != BOND_MODE_TLB)) {
3889                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
3890                                 bond_mode_name(bond_mode));
3891                 } else {
3892                         bond_opt_initstr(&newval, xmit_hash_policy);
3893                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
3894                                                 &newval);
3895                         if (!valptr) {
3896                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
3897                                        xmit_hash_policy);
3898                                 return -EINVAL;
3899                         }
3900                         xmit_hashtype = valptr->value;
3901                 }
3902         }
3903
3904         if (lacp_rate) {
3905                 if (bond_mode != BOND_MODE_8023AD) {
3906                         pr_info("lacp_rate param is irrelevant in mode %s\n",
3907                                 bond_mode_name(bond_mode));
3908                 } else {
3909                         bond_opt_initstr(&newval, lacp_rate);
3910                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
3911                                                 &newval);
3912                         if (!valptr) {
3913                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
3914                                        lacp_rate);
3915                                 return -EINVAL;
3916                         }
3917                         lacp_fast = valptr->value;
3918                 }
3919         }
3920
3921         if (ad_select) {
3922                 bond_opt_initstr(&newval, ad_select);
3923                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
3924                                         &newval);
3925                 if (!valptr) {
3926                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
3927                         return -EINVAL;
3928                 }
3929                 params->ad_select = valptr->value;
3930                 if (bond_mode != BOND_MODE_8023AD)
3931                         pr_warn("ad_select param only affects 802.3ad mode\n");
3932         } else {
3933                 params->ad_select = BOND_AD_STABLE;
3934         }
3935
3936         if (max_bonds < 0) {
3937                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
3938                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
3939                 max_bonds = BOND_DEFAULT_MAX_BONDS;
3940         }
3941
3942         if (miimon < 0) {
3943                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
3944                         miimon, INT_MAX);
3945                 miimon = 0;
3946         }
3947
3948         if (updelay < 0) {
3949                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
3950                         updelay, INT_MAX);
3951                 updelay = 0;
3952         }
3953
3954         if (downdelay < 0) {
3955                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
3956                         downdelay, INT_MAX);
3957                 downdelay = 0;
3958         }
3959
3960         if ((use_carrier != 0) && (use_carrier != 1)) {
3961                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
3962                         use_carrier);
3963                 use_carrier = 1;
3964         }
3965
3966         if (num_peer_notif < 0 || num_peer_notif > 255) {
3967                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
3968                         num_peer_notif);
3969                 num_peer_notif = 1;
3970         }
3971
3972         /* reset values for 802.3ad/TLB/ALB */
3973         if (!bond_mode_uses_arp(bond_mode)) {
3974                 if (!miimon) {
3975                         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");
3976                         pr_warn("Forcing miimon to 100msec\n");
3977                         miimon = BOND_DEFAULT_MIIMON;
3978                 }
3979         }
3980
3981         if (tx_queues < 1 || tx_queues > 255) {
3982                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
3983                         tx_queues, BOND_DEFAULT_TX_QUEUES);
3984                 tx_queues = BOND_DEFAULT_TX_QUEUES;
3985         }
3986
3987         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
3988                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
3989                         all_slaves_active);
3990                 all_slaves_active = 0;
3991         }
3992
3993         if (resend_igmp < 0 || resend_igmp > 255) {
3994                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
3995                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
3996                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
3997         }
3998
3999         bond_opt_initval(&newval, packets_per_slave);
4000         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4001                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4002                         packets_per_slave, USHRT_MAX);
4003                 packets_per_slave = 1;
4004         }
4005
4006         if (bond_mode == BOND_MODE_ALB) {
4007                 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",
4008                           updelay);
4009         }
4010
4011         if (!miimon) {
4012                 if (updelay || downdelay) {
4013                         /* just warn the user the up/down delay will have
4014                          * no effect since miimon is zero...
4015                          */
4016                         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",
4017                                 updelay, downdelay);
4018                 }
4019         } else {
4020                 /* don't allow arp monitoring */
4021                 if (arp_interval) {
4022                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4023                                 miimon, arp_interval);
4024                         arp_interval = 0;
4025                 }
4026
4027                 if ((updelay % miimon) != 0) {
4028                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4029                                 updelay, miimon, (updelay / miimon) * miimon);
4030                 }
4031
4032                 updelay /= miimon;
4033
4034                 if ((downdelay % miimon) != 0) {
4035                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4036                                 downdelay, miimon,
4037                                 (downdelay / miimon) * miimon);
4038                 }
4039
4040                 downdelay /= miimon;
4041         }
4042
4043         if (arp_interval < 0) {
4044                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4045                         arp_interval, INT_MAX);
4046                 arp_interval = 0;
4047         }
4048
4049         for (arp_ip_count = 0, i = 0;
4050              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4051                 __be32 ip;
4052
4053                 /* not a complete check, but good enough to catch mistakes */
4054                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4055                     !bond_is_ip_target_ok(ip)) {
4056                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4057                                 arp_ip_target[i]);
4058                         arp_interval = 0;
4059                 } else {
4060                         if (bond_get_targets_ip(arp_target, ip) == -1)
4061                                 arp_target[arp_ip_count++] = ip;
4062                         else
4063                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4064                                         &ip);
4065                 }
4066         }
4067
4068         if (arp_interval && !arp_ip_count) {
4069                 /* don't allow arping if no arp_ip_target given... */
4070                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4071                         arp_interval);
4072                 arp_interval = 0;
4073         }
4074
4075         if (arp_validate) {
4076                 if (!arp_interval) {
4077                         pr_err("arp_validate requires arp_interval\n");
4078                         return -EINVAL;
4079                 }
4080
4081                 bond_opt_initstr(&newval, arp_validate);
4082                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4083                                         &newval);
4084                 if (!valptr) {
4085                         pr_err("Error: invalid arp_validate \"%s\"\n",
4086                                arp_validate);
4087                         return -EINVAL;
4088                 }
4089                 arp_validate_value = valptr->value;
4090         } else {
4091                 arp_validate_value = 0;
4092         }
4093
4094         arp_all_targets_value = 0;
4095         if (arp_all_targets) {
4096                 bond_opt_initstr(&newval, arp_all_targets);
4097                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4098                                         &newval);
4099                 if (!valptr) {
4100                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4101                                arp_all_targets);
4102                         arp_all_targets_value = 0;
4103                 } else {
4104                         arp_all_targets_value = valptr->value;
4105                 }
4106         }
4107
4108         if (miimon) {
4109                 pr_info("MII link monitoring set to %d ms\n", miimon);
4110         } else if (arp_interval) {
4111                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4112                                           arp_validate_value);
4113                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4114                         arp_interval, valptr->string, arp_ip_count);
4115
4116                 for (i = 0; i < arp_ip_count; i++)
4117                         pr_cont(" %s", arp_ip_target[i]);
4118
4119                 pr_cont("\n");
4120
4121         } else if (max_bonds) {
4122                 /* miimon and arp_interval not set, we need one so things
4123                  * work as expected, see bonding.txt for details
4124                  */
4125                 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");
4126         }
4127
4128         if (primary && !bond_mode_uses_primary(bond_mode)) {
4129                 /* currently, using a primary only makes sense
4130                  * in active backup, TLB or ALB modes
4131                  */
4132                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4133                         primary, bond_mode_name(bond_mode));
4134                 primary = NULL;
4135         }
4136
4137         if (primary && primary_reselect) {
4138                 bond_opt_initstr(&newval, primary_reselect);
4139                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4140                                         &newval);
4141                 if (!valptr) {
4142                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4143                                primary_reselect);
4144                         return -EINVAL;
4145                 }
4146                 primary_reselect_value = valptr->value;
4147         } else {
4148                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4149         }
4150
4151         if (fail_over_mac) {
4152                 bond_opt_initstr(&newval, fail_over_mac);
4153                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4154                                         &newval);
4155                 if (!valptr) {
4156                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4157                                fail_over_mac);
4158                         return -EINVAL;
4159                 }
4160                 fail_over_mac_value = valptr->value;
4161                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4162                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4163         } else {
4164                 fail_over_mac_value = BOND_FOM_NONE;
4165         }
4166
4167         if (lp_interval == 0) {
4168                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4169                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4170                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4171         }
4172
4173         /* fill params struct with the proper values */
4174         params->mode = bond_mode;
4175         params->xmit_policy = xmit_hashtype;
4176         params->miimon = miimon;
4177         params->num_peer_notif = num_peer_notif;
4178         params->arp_interval = arp_interval;
4179         params->arp_validate = arp_validate_value;
4180         params->arp_all_targets = arp_all_targets_value;
4181         params->updelay = updelay;
4182         params->downdelay = downdelay;
4183         params->use_carrier = use_carrier;
4184         params->lacp_fast = lacp_fast;
4185         params->primary[0] = 0;
4186         params->primary_reselect = primary_reselect_value;
4187         params->fail_over_mac = fail_over_mac_value;
4188         params->tx_queues = tx_queues;
4189         params->all_slaves_active = all_slaves_active;
4190         params->resend_igmp = resend_igmp;
4191         params->min_links = min_links;
4192         params->lp_interval = lp_interval;
4193         params->packets_per_slave = packets_per_slave;
4194         params->tlb_dynamic_lb = 1; /* Default value */
4195         if (packets_per_slave > 0) {
4196                 params->reciprocal_packets_per_slave =
4197                         reciprocal_value(packets_per_slave);
4198         } else {
4199                 /* reciprocal_packets_per_slave is unused if
4200                  * packets_per_slave is 0 or 1, just initialize it
4201                  */
4202                 params->reciprocal_packets_per_slave =
4203                         (struct reciprocal_value) { 0 };
4204         }
4205
4206         if (primary) {
4207                 strncpy(params->primary, primary, IFNAMSIZ);
4208                 params->primary[IFNAMSIZ - 1] = 0;
4209         }
4210
4211         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4212
4213         return 0;
4214 }
4215
4216 static struct lock_class_key bonding_netdev_xmit_lock_key;
4217 static struct lock_class_key bonding_netdev_addr_lock_key;
4218 static struct lock_class_key bonding_tx_busylock_key;
4219
4220 static void bond_set_lockdep_class_one(struct net_device *dev,
4221                                        struct netdev_queue *txq,
4222                                        void *_unused)
4223 {
4224         lockdep_set_class(&txq->_xmit_lock,
4225                           &bonding_netdev_xmit_lock_key);
4226 }
4227
4228 static void bond_set_lockdep_class(struct net_device *dev)
4229 {
4230         lockdep_set_class(&dev->addr_list_lock,
4231                           &bonding_netdev_addr_lock_key);
4232         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4233         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4234 }
4235
4236 /* Called from registration process */
4237 static int bond_init(struct net_device *bond_dev)
4238 {
4239         struct bonding *bond = netdev_priv(bond_dev);
4240         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4241
4242         netdev_dbg(bond_dev, "Begin bond_init\n");
4243
4244         bond->wq = create_singlethread_workqueue(bond_dev->name);
4245         if (!bond->wq)
4246                 return -ENOMEM;
4247
4248         bond_set_lockdep_class(bond_dev);
4249
4250         list_add_tail(&bond->bond_list, &bn->dev_list);
4251
4252         bond_prepare_sysfs_group(bond);
4253
4254         bond_debug_register(bond);
4255
4256         /* Ensure valid dev_addr */
4257         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4258             bond_dev->addr_assign_type == NET_ADDR_PERM)
4259                 eth_hw_addr_random(bond_dev);
4260
4261         return 0;
4262 }
4263
4264 unsigned int bond_get_num_tx_queues(void)
4265 {
4266         return tx_queues;
4267 }
4268
4269 /* Create a new bond based on the specified name and bonding parameters.
4270  * If name is NULL, obtain a suitable "bond%d" name for us.
4271  * Caller must NOT hold rtnl_lock; we need to release it here before we
4272  * set up our sysfs entries.
4273  */
4274 int bond_create(struct net *net, const char *name)
4275 {
4276         struct net_device *bond_dev;
4277         int res;
4278
4279         rtnl_lock();
4280
4281         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4282                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
4283                                    bond_setup, tx_queues);
4284         if (!bond_dev) {
4285                 pr_err("%s: eek! can't alloc netdev!\n", name);
4286                 rtnl_unlock();
4287                 return -ENOMEM;
4288         }
4289
4290         dev_net_set(bond_dev, net);
4291         bond_dev->rtnl_link_ops = &bond_link_ops;
4292
4293         res = register_netdevice(bond_dev);
4294
4295         netif_carrier_off(bond_dev);
4296
4297         rtnl_unlock();
4298         if (res < 0)
4299                 bond_destructor(bond_dev);
4300         return res;
4301 }
4302
4303 static int __net_init bond_net_init(struct net *net)
4304 {
4305         struct bond_net *bn = net_generic(net, bond_net_id);
4306
4307         bn->net = net;
4308         INIT_LIST_HEAD(&bn->dev_list);
4309
4310         bond_create_proc_dir(bn);
4311         bond_create_sysfs(bn);
4312
4313         return 0;
4314 }
4315
4316 static void __net_exit bond_net_exit(struct net *net)
4317 {
4318         struct bond_net *bn = net_generic(net, bond_net_id);
4319         struct bonding *bond, *tmp_bond;
4320         LIST_HEAD(list);
4321
4322         bond_destroy_sysfs(bn);
4323
4324         /* Kill off any bonds created after unregistering bond rtnl ops */
4325         rtnl_lock();
4326         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4327                 unregister_netdevice_queue(bond->dev, &list);
4328         unregister_netdevice_many(&list);
4329         rtnl_unlock();
4330
4331         bond_destroy_proc_dir(bn);
4332 }
4333
4334 static struct pernet_operations bond_net_ops = {
4335         .init = bond_net_init,
4336         .exit = bond_net_exit,
4337         .id   = &bond_net_id,
4338         .size = sizeof(struct bond_net),
4339 };
4340
4341 static int __init bonding_init(void)
4342 {
4343         int i;
4344         int res;
4345
4346         pr_info("%s", bond_version);
4347
4348         res = bond_check_params(&bonding_defaults);
4349         if (res)
4350                 goto out;
4351
4352         res = register_pernet_subsys(&bond_net_ops);
4353         if (res)
4354                 goto out;
4355
4356         res = bond_netlink_init();
4357         if (res)
4358                 goto err_link;
4359
4360         bond_create_debugfs();
4361
4362         for (i = 0; i < max_bonds; i++) {
4363                 res = bond_create(&init_net, NULL);
4364                 if (res)
4365                         goto err;
4366         }
4367
4368         register_netdevice_notifier(&bond_netdev_notifier);
4369 out:
4370         return res;
4371 err:
4372         bond_destroy_debugfs();
4373         bond_netlink_fini();
4374 err_link:
4375         unregister_pernet_subsys(&bond_net_ops);
4376         goto out;
4377
4378 }
4379
4380 static void __exit bonding_exit(void)
4381 {
4382         unregister_netdevice_notifier(&bond_netdev_notifier);
4383
4384         bond_destroy_debugfs();
4385
4386         bond_netlink_fini();
4387         unregister_pernet_subsys(&bond_net_ops);
4388
4389 #ifdef CONFIG_NET_POLL_CONTROLLER
4390         /* Make sure we don't have an imbalance on our netpoll blocking */
4391         WARN_ON(atomic_read(&netpoll_block_tx));
4392 #endif
4393 }
4394
4395 module_init(bonding_init);
4396 module_exit(bonding_exit);
4397 MODULE_LICENSE("GPL");
4398 MODULE_VERSION(DRV_VERSION);
4399 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4400 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");