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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  35
  36 #include <linux/kernel.h>
  37 #include <linux/module.h>
  38 #include <linux/types.h>
  39 #include <linux/fcntl.h>
  40 #include <linux/interrupt.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/ioport.h>
  43 #include <linux/in.h>
  44 #include <net/ip.h>
  45 #include <linux/ip.h>
  46 #include <linux/tcp.h>
  47 #include <linux/udp.h>
  48 #include <linux/slab.h>
  49 #include <linux/string.h>
  50 #include <linux/init.h>
  51 #include <linux/timer.h>
  52 #include <linux/socket.h>
  53 #include <linux/ctype.h>
  54 #include <linux/inet.h>
  55 #include <linux/bitops.h>
  56 #include <linux/io.h>
  57 #include <asm/dma.h>
  58 #include <linux/uaccess.h>
  59 #include <linux/errno.h>
  60 #include <linux/netdevice.h>
  61 #include <linux/inetdevice.h>
  62 #include <linux/igmp.h>
  63 #include <linux/etherdevice.h>
  64 #include <linux/skbuff.h>
  65 #include <net/sock.h>
  66 #include <linux/rtnetlink.h>
  67 #include <linux/smp.h>
  68 #include <linux/if_ether.h>
  69 #include <net/arp.h>
  70 #include <linux/mii.h>
  71 #include <linux/ethtool.h>
  72 #include <linux/if_vlan.h>
  73 #include <linux/if_bonding.h>
  74 #include <linux/jiffies.h>
  75 #include <linux/preempt.h>
  76 #include <net/route.h>
  77 #include <net/net_namespace.h>
  78 #include <net/netns/generic.h>
  79 #include <net/pkt_sched.h>
  80 #include <linux/rculist.h>
  81 #include <net/flow_keys.h>
  82 #include "bonding.h"
  83 #include "bond_3ad.h"
  84 #include "bond_alb.h"
  85
  86 /*---------------------------- Module parameters ----------------------------*/
  87
  88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  89 #define BOND_LINK_MON_INTERV    0
  90 #define BOND_LINK_ARP_INTERV    0
  91
  92 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  93 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  94 static int num_peer_notif = 1;
  95 static int miimon       = BOND_LINK_MON_INTERV;
  96 static int updelay;
  97 static int downdelay;
  98 static int use_carrier  = 1;
  99 static char *mode;
 100 static char *primary;
 101 static char *primary_reselect;
 102 static char *lacp_rate;
 103 static int min_links;
 104 static char *ad_select;
 105 static char *xmit_hash_policy;
 106 static int arp_interval = BOND_LINK_ARP_INTERV;
 107 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 108 static char *arp_validate;
 109 static char *arp_all_targets;
 110 static char *fail_over_mac;
 111 static int all_slaves_active;
 112 static struct bond_params bonding_defaults;
 113 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 114
 115 module_param(max_bonds, int, 0);
 116 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 117 module_param(tx_queues, int, 0);
 118 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 119 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 120 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 121                                "failover event (alias of num_unsol_na)");
 122 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 123 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 124                                "failover event (alias of num_grat_arp)");
 125 module_param(miimon, int, 0);
 126 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 127 module_param(updelay, int, 0);
 128 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 129 module_param(downdelay, int, 0);
 130 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 131                             "in milliseconds");
 132 module_param(use_carrier, int, 0);
 133 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 134                               "0 for off, 1 for on (default)");
 135 module_param(mode, charp, 0);
 136 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 137                        "1 for active-backup, 2 for balance-xor, "
 138                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 139                        "6 for balance-alb");
 140 module_param(primary, charp, 0);
 141 MODULE_PARM_DESC(primary, "Primary network device to use");
 142 module_param(primary_reselect, charp, 0);
 143 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 144                                    "once it comes up; "
 145                                    "0 for always (default), "
 146                                    "1 for only if speed of primary is "
 147                                    "better, "
 148                                    "2 for only on active slave "
 149                                    "failure");
 150 module_param(lacp_rate, charp, 0);
 151 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 152                             "0 for slow, 1 for fast");
 153 module_param(ad_select, charp, 0);
 154 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 155                             "0 for stable (default), 1 for bandwidth, "
 156                             "2 for count");
 157 module_param(min_links, int, 0);
 158 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 159
 160 module_param(xmit_hash_policy, charp, 0);
 161 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 162                                    "0 for layer 2 (default), 1 for layer 3+4, "
 163                                    "2 for layer 2+3, 3 for encap layer 2+3, "
 164                                    "4 for encap layer 3+4");
 165 module_param(arp_interval, int, 0);
 166 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 167 module_param_array(arp_ip_target, charp, NULL, 0);
 168 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 169 module_param(arp_validate, charp, 0);
 170 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 171                                "0 for none (default), 1 for active, "
 172                                "2 for backup, 3 for all");
 173 module_param(arp_all_targets, charp, 0);
 174 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 175 module_param(fail_over_mac, charp, 0);
 176 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 177                                 "the same MAC; 0 for none (default), "
 178                                 "1 for active, 2 for follow");
 179 module_param(all_slaves_active, int, 0);
 180 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 181                                      "by setting active flag for all slaves; "
 182                                      "0 for never (default), 1 for always.");
 183 module_param(resend_igmp, int, 0);
 184 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 185                               "link failure");
 186
 187 /*----------------------------- Global variables ----------------------------*/
 188
 189 #ifdef CONFIG_NET_POLL_CONTROLLER
 190 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 191 #endif
 192
 193 int bond_net_id __read_mostly;
 194
 195 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 196 static int arp_ip_count;
 197 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 198 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 199 static int lacp_fast;
 200
 201 const struct bond_parm_tbl bond_lacp_tbl[] = {
 202 {       "slow",         AD_LACP_SLOW},
 203 {       "fast",         AD_LACP_FAST},
 204 {       NULL,           -1},
 205 };
 206
 207 const struct bond_parm_tbl bond_mode_tbl[] = {
 208 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 209 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 210 {       "balance-xor",          BOND_MODE_XOR},
 211 {       "broadcast",            BOND_MODE_BROADCAST},
 212 {       "802.3ad",              BOND_MODE_8023AD},
 213 {       "balance-tlb",          BOND_MODE_TLB},
 214 {       "balance-alb",          BOND_MODE_ALB},
 215 {       NULL,                   -1},
 216 };
 217
 218 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 219 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 220 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 221 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 222 {       "encap2+3",             BOND_XMIT_POLICY_ENCAP23},
 223 {       "encap3+4",             BOND_XMIT_POLICY_ENCAP34},
 224 {       NULL,                   -1},
 225 };
 226
 227 const struct bond_parm_tbl arp_all_targets_tbl[] = {
 228 {       "any",                  BOND_ARP_TARGETS_ANY},
 229 {       "all",                  BOND_ARP_TARGETS_ALL},
 230 {       NULL,                   -1},
 231 };
 232
 233 const struct bond_parm_tbl arp_validate_tbl[] = {
 234 {       "none",                 BOND_ARP_VALIDATE_NONE},
 235 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 236 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 237 {       "all",                  BOND_ARP_VALIDATE_ALL},
 238 {       NULL,                   -1},
 239 };
 240
 241 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 242 {       "none",                 BOND_FOM_NONE},
 243 {       "active",               BOND_FOM_ACTIVE},
 244 {       "follow",               BOND_FOM_FOLLOW},
 245 {       NULL,                   -1},
 246 };
 247
 248 const struct bond_parm_tbl pri_reselect_tbl[] = {
 249 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 250 {       "better",               BOND_PRI_RESELECT_BETTER},
 251 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 252 {       NULL,                   -1},
 253 };
 254
 255 struct bond_parm_tbl ad_select_tbl[] = {
 256 {       "stable",       BOND_AD_STABLE},
 257 {       "bandwidth",    BOND_AD_BANDWIDTH},
 258 {       "count",        BOND_AD_COUNT},
 259 {       NULL,           -1},
 260 };
 261
 262 /*-------------------------- Forward declarations ---------------------------*/
 263
 264 static int bond_init(struct net_device *bond_dev);
 265 static void bond_uninit(struct net_device *bond_dev);
 266
 267 /*---------------------------- General routines -----------------------------*/
 268
 269 const char *bond_mode_name(int mode)
 270 {
 271         static const char *names[] = {
 272                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 273                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 274                 [BOND_MODE_XOR] = "load balancing (xor)",
 275                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 276                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 277                 [BOND_MODE_TLB] = "transmit load balancing",
 278                 [BOND_MODE_ALB] = "adaptive load balancing",
 279         };
 280
 281         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 282                 return "unknown";
 283
 284         return names[mode];
 285 }
 286
 287 /*---------------------------------- VLAN -----------------------------------*/
 288
 289 /**
 290  * bond_dev_queue_xmit - Prepare skb for xmit.
 291  *
 292  * @bond: bond device that got this skb for tx.
 293  * @skb: hw accel VLAN tagged skb to transmit
 294  * @slave_dev: slave that is supposed to xmit this skbuff
 295  */
 296 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 297                         struct net_device *slave_dev)
 298 {
 299         skb->dev = slave_dev;
 300
 301         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 302                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 303         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 304
 305         if (unlikely(netpoll_tx_running(bond->dev)))
 306                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 307         else
 308                 dev_queue_xmit(skb);
 309
 310         return 0;
 311 }
 312
 313 /*
 314  * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 315  * We don't protect the slave list iteration with a lock because:
 316  * a. This operation is performed in IOCTL context,
 317  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 318  * c. Holding a lock with BH disabled while directly calling a base driver
 319  *    entry point is generally a BAD idea.
 320  *
 321  * The design of synchronization/protection for this operation in the 8021q
 322  * module is good for one or more VLAN devices over a single physical device
 323  * and cannot be extended for a teaming solution like bonding, so there is a
 324  * potential race condition here where a net device from the vlan group might
 325  * be referenced (either by a base driver or the 8021q code) while it is being
 326  * removed from the system. However, it turns out we're not making matters
 327  * worse, and if it works for regular VLAN usage it will work here too.
 328 */
 329
 330 /**
 331  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 332  * @bond_dev: bonding net device that got called
 333  * @vid: vlan id being added
 334  */
 335 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 336                                 __be16 proto, u16 vid)
 337 {
 338         struct bonding *bond = netdev_priv(bond_dev);
 339         struct slave *slave, *rollback_slave;
 340         struct list_head *iter;
 341         int res;
 342
 343         bond_for_each_slave(bond, slave, iter) {
 344                 res = vlan_vid_add(slave->dev, proto, vid);
 345                 if (res)
 346                         goto unwind;
 347         }
 348
 349         return 0;
 350
 351 unwind:
 352         /* unwind to the slave that failed */
 353         bond_for_each_slave(bond, rollback_slave, iter) {
 354                 if (rollback_slave == slave)
 355                         break;
 356
 357                 vlan_vid_del(rollback_slave->dev, proto, vid);
 358         }
 359
 360         return res;
 361 }
 362
 363 /**
 364  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 365  * @bond_dev: bonding net device that got called
 366  * @vid: vlan id being removed
 367  */
 368 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 369                                  __be16 proto, u16 vid)
 370 {
 371         struct bonding *bond = netdev_priv(bond_dev);
 372         struct list_head *iter;
 373         struct slave *slave;
 374
 375         bond_for_each_slave(bond, slave, iter)
 376                 vlan_vid_del(slave->dev, proto, vid);
 377
 378         if (bond_is_lb(bond))
 379                 bond_alb_clear_vlan(bond, vid);
 380
 381         return 0;
 382 }
 383
 384 /*------------------------------- Link status -------------------------------*/
 385
 386 /*
 387  * Set the carrier state for the master according to the state of its
 388  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 389  * do special 802.3ad magic.
 390  *
 391  * Returns zero if carrier state does not change, nonzero if it does.
 392  */
 393 static int bond_set_carrier(struct bonding *bond)
 394 {
 395         struct list_head *iter;
 396         struct slave *slave;
 397
 398         if (!bond_has_slaves(bond))
 399                 goto down;
 400
 401         if (bond->params.mode == BOND_MODE_8023AD)
 402                 return bond_3ad_set_carrier(bond);
 403
 404         bond_for_each_slave(bond, slave, iter) {
 405                 if (slave->link == BOND_LINK_UP) {
 406                         if (!netif_carrier_ok(bond->dev)) {
 407                                 netif_carrier_on(bond->dev);
 408                                 return 1;
 409                         }
 410                         return 0;
 411                 }
 412         }
 413
 414 down:
 415         if (netif_carrier_ok(bond->dev)) {
 416                 netif_carrier_off(bond->dev);
 417                 return 1;
 418         }
 419         return 0;
 420 }
 421
 422 /*
 423  * Get link speed and duplex from the slave's base driver
 424  * using ethtool. If for some reason the call fails or the
 425  * values are invalid, set speed and duplex to -1,
 426  * and return.
 427  */
 428 static void bond_update_speed_duplex(struct slave *slave)
 429 {
 430         struct net_device *slave_dev = slave->dev;
 431         struct ethtool_cmd ecmd;
 432         u32 slave_speed;
 433         int res;
 434
 435         slave->speed = SPEED_UNKNOWN;
 436         slave->duplex = DUPLEX_UNKNOWN;
 437
 438         res = __ethtool_get_settings(slave_dev, &ecmd);
 439         if (res < 0)
 440                 return;
 441
 442         slave_speed = ethtool_cmd_speed(&ecmd);
 443         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 444                 return;
 445
 446         switch (ecmd.duplex) {
 447         case DUPLEX_FULL:
 448         case DUPLEX_HALF:
 449                 break;
 450         default:
 451                 return;
 452         }
 453
 454         slave->speed = slave_speed;
 455         slave->duplex = ecmd.duplex;
 456
 457         return;
 458 }
 459
 460 /*
 461  * if <dev> supports MII link status reporting, check its link status.
 462  *
 463  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 464  * depending upon the setting of the use_carrier parameter.
 465  *
 466  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 467  * can't tell and just pretend it is), or 0, meaning that the link is
 468  * down.
 469  *
 470  * If reporting is non-zero, instead of faking link up, return -1 if
 471  * both ETHTOOL and MII ioctls fail (meaning the device does not
 472  * support them).  If use_carrier is set, return whatever it says.
 473  * It'd be nice if there was a good way to tell if a driver supports
 474  * netif_carrier, but there really isn't.
 475  */
 476 static int bond_check_dev_link(struct bonding *bond,
 477                                struct net_device *slave_dev, int reporting)
 478 {
 479         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 480         int (*ioctl)(struct net_device *, struct ifreq *, int);
 481         struct ifreq ifr;
 482         struct mii_ioctl_data *mii;
 483
 484         if (!reporting && !netif_running(slave_dev))
 485                 return 0;
 486
 487         if (bond->params.use_carrier)
 488                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 489
 490         /* Try to get link status using Ethtool first. */
 491         if (slave_dev->ethtool_ops->get_link)
 492                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 493                         BMSR_LSTATUS : 0;
 494
 495         /* Ethtool can't be used, fallback to MII ioctls. */
 496         ioctl = slave_ops->ndo_do_ioctl;
 497         if (ioctl) {
 498                 /* TODO: set pointer to correct ioctl on a per team member */
 499                 /*       bases to make this more efficient. that is, once  */
 500                 /*       we determine the correct ioctl, we will always    */
 501                 /*       call it and not the others for that team          */
 502                 /*       member.                                           */
 503
 504                 /*
 505                  * We cannot assume that SIOCGMIIPHY will also read a
 506                  * register; not all network drivers (e.g., e100)
 507                  * support that.
 508                  */
 509
 510                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 511                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 512                 mii = if_mii(&ifr);
 513                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 514                         mii->reg_num = MII_BMSR;
 515                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 516                                 return mii->val_out & BMSR_LSTATUS;
 517                 }
 518         }
 519
 520         /*
 521          * If reporting, report that either there's no dev->do_ioctl,
 522          * or both SIOCGMIIREG and get_link failed (meaning that we
 523          * cannot report link status).  If not reporting, pretend
 524          * we're ok.
 525          */
 526         return reporting ? -1 : BMSR_LSTATUS;
 527 }
 528
 529 /*----------------------------- Multicast list ------------------------------*/
 530
 531 /*
 532  * Push the promiscuity flag down to appropriate slaves
 533  */
 534 static int bond_set_promiscuity(struct bonding *bond, int inc)
 535 {
 536         struct list_head *iter;
 537         int err = 0;
 538
 539         if (USES_PRIMARY(bond->params.mode)) {
 540                 /* write lock already acquired */
 541                 if (bond->curr_active_slave) {
 542                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 543                                                   inc);
 544                 }
 545         } else {
 546                 struct slave *slave;
 547
 548                 bond_for_each_slave(bond, slave, iter) {
 549                         err = dev_set_promiscuity(slave->dev, inc);
 550                         if (err)
 551                                 return err;
 552                 }
 553         }
 554         return err;
 555 }
 556
 557 /*
 558  * Push the allmulti flag down to all slaves
 559  */
 560 static int bond_set_allmulti(struct bonding *bond, int inc)
 561 {
 562         struct list_head *iter;
 563         int err = 0;
 564
 565         if (USES_PRIMARY(bond->params.mode)) {
 566                 /* write lock already acquired */
 567                 if (bond->curr_active_slave) {
 568                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 569                                                inc);
 570                 }
 571         } else {
 572                 struct slave *slave;
 573
 574                 bond_for_each_slave(bond, slave, iter) {
 575                         err = dev_set_allmulti(slave->dev, inc);
 576                         if (err)
 577                                 return err;
 578                 }
 579         }
 580         return err;
 581 }
 582
 583 /*
 584  * Retrieve the list of registered multicast addresses for the bonding
 585  * device and retransmit an IGMP JOIN request to the current active
 586  * slave.
 587  */
 588 static void bond_resend_igmp_join_requests(struct bonding *bond)
 589 {
 590         if (!rtnl_trylock()) {
 591                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 592                 return;
 593         }
 594         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 595         rtnl_unlock();
 596
 597         /* We use curr_slave_lock to protect against concurrent access to
 598          * igmp_retrans from multiple running instances of this function and
 599          * bond_change_active_slave
 600          */
 601         write_lock_bh(&bond->curr_slave_lock);
 602         if (bond->igmp_retrans > 1) {
 603                 bond->igmp_retrans--;
 604                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 605         }
 606         write_unlock_bh(&bond->curr_slave_lock);
 607 }
 608
 609 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 610 {
 611         struct bonding *bond = container_of(work, struct bonding,
 612                                             mcast_work.work);
 613
 614         bond_resend_igmp_join_requests(bond);
 615 }
 616
 617 /* Flush bond's hardware addresses from slave
 618  */
 619 static void bond_hw_addr_flush(struct net_device *bond_dev,
 620                                struct net_device *slave_dev)
 621 {
 622         struct bonding *bond = netdev_priv(bond_dev);
 623
 624         dev_uc_unsync(slave_dev, bond_dev);
 625         dev_mc_unsync(slave_dev, bond_dev);
 626
 627         if (bond->params.mode == BOND_MODE_8023AD) {
 628                 /* del lacpdu mc addr from mc list */
 629                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 630
 631                 dev_mc_del(slave_dev, lacpdu_multicast);
 632         }
 633 }
 634
 635 /*--------------------------- Active slave change ---------------------------*/
 636
 637 /* Update the hardware address list and promisc/allmulti for the new and
 638  * old active slaves (if any).  Modes that are !USES_PRIMARY keep all
 639  * slaves up date at all times; only the USES_PRIMARY modes need to call
 640  * this function to swap these settings during a failover.
 641  */
 642 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 643                               struct slave *old_active)
 644 {
 645         ASSERT_RTNL();
 646
 647         if (old_active) {
 648                 if (bond->dev->flags & IFF_PROMISC)
 649                         dev_set_promiscuity(old_active->dev, -1);
 650
 651                 if (bond->dev->flags & IFF_ALLMULTI)
 652                         dev_set_allmulti(old_active->dev, -1);
 653
 654                 bond_hw_addr_flush(bond->dev, old_active->dev);
 655         }
 656
 657         if (new_active) {
 658                 /* FIXME: Signal errors upstream. */
 659                 if (bond->dev->flags & IFF_PROMISC)
 660                         dev_set_promiscuity(new_active->dev, 1);
 661
 662                 if (bond->dev->flags & IFF_ALLMULTI)
 663                         dev_set_allmulti(new_active->dev, 1);
 664
 665                 netif_addr_lock_bh(bond->dev);
 666                 dev_uc_sync(new_active->dev, bond->dev);
 667                 dev_mc_sync(new_active->dev, bond->dev);
 668                 netif_addr_unlock_bh(bond->dev);
 669         }
 670 }
 671
 672 /**
 673  * bond_set_dev_addr - clone slave's address to bond
 674  * @bond_dev: bond net device
 675  * @slave_dev: slave net device
 676  *
 677  * Should be called with RTNL held.
 678  */
 679 static void bond_set_dev_addr(struct net_device *bond_dev,
 680                               struct net_device *slave_dev)
 681 {
 682         pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 683                  bond_dev, slave_dev, slave_dev->addr_len);
 684         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 685         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 686         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 687 }
 688
 689 /*
 690  * bond_do_fail_over_mac
 691  *
 692  * Perform special MAC address swapping for fail_over_mac settings
 693  *
 694  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 695  */
 696 static void bond_do_fail_over_mac(struct bonding *bond,
 697                                   struct slave *new_active,
 698                                   struct slave *old_active)
 699         __releases(&bond->curr_slave_lock)
 700         __releases(&bond->lock)
 701         __acquires(&bond->lock)
 702         __acquires(&bond->curr_slave_lock)
 703 {
 704         u8 tmp_mac[ETH_ALEN];
 705         struct sockaddr saddr;
 706         int rv;
 707
 708         switch (bond->params.fail_over_mac) {
 709         case BOND_FOM_ACTIVE:
 710                 if (new_active) {
 711                         write_unlock_bh(&bond->curr_slave_lock);
 712                         read_unlock(&bond->lock);
 713                         bond_set_dev_addr(bond->dev, new_active->dev);
 714                         read_lock(&bond->lock);
 715                         write_lock_bh(&bond->curr_slave_lock);
 716                 }
 717                 break;
 718         case BOND_FOM_FOLLOW:
 719                 /*
 720                  * if new_active && old_active, swap them
 721                  * if just old_active, do nothing (going to no active slave)
 722                  * if just new_active, set new_active to bond's MAC
 723                  */
 724                 if (!new_active)
 725                         return;
 726
 727                 write_unlock_bh(&bond->curr_slave_lock);
 728                 read_unlock(&bond->lock);
 729
 730                 if (old_active) {
 731                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 732                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 733                                ETH_ALEN);
 734                         saddr.sa_family = new_active->dev->type;
 735                 } else {
 736                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 737                         saddr.sa_family = bond->dev->type;
 738                 }
 739
 740                 rv = dev_set_mac_address(new_active->dev, &saddr);
 741                 if (rv) {
 742                         pr_err("%s: Error %d setting MAC of slave %s\n",
 743                                bond->dev->name, -rv, new_active->dev->name);
 744                         goto out;
 745                 }
 746
 747                 if (!old_active)
 748                         goto out;
 749
 750                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 751                 saddr.sa_family = old_active->dev->type;
 752
 753                 rv = dev_set_mac_address(old_active->dev, &saddr);
 754                 if (rv)
 755                         pr_err("%s: Error %d setting MAC of slave %s\n",
 756                                bond->dev->name, -rv, new_active->dev->name);
 757 out:
 758                 read_lock(&bond->lock);
 759                 write_lock_bh(&bond->curr_slave_lock);
 760                 break;
 761         default:
 762                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 763                        bond->dev->name, bond->params.fail_over_mac);
 764                 break;
 765         }
 766
 767 }
 768
 769 static bool bond_should_change_active(struct bonding *bond)
 770 {
 771         struct slave *prim = bond->primary_slave;
 772         struct slave *curr = bond->curr_active_slave;
 773
 774         if (!prim || !curr || curr->link != BOND_LINK_UP)
 775                 return true;
 776         if (bond->force_primary) {
 777                 bond->force_primary = false;
 778                 return true;
 779         }
 780         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 781             (prim->speed < curr->speed ||
 782              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 783                 return false;
 784         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 785                 return false;
 786         return true;
 787 }
 788
 789 /**
 790  * find_best_interface - select the best available slave to be the active one
 791  * @bond: our bonding struct
 792  */
 793 static struct slave *bond_find_best_slave(struct bonding *bond)
 794 {
 795         struct slave *slave, *bestslave = NULL;
 796         struct list_head *iter;
 797         int mintime = bond->params.updelay;
 798
 799         if (bond->primary_slave && bond->primary_slave->link == BOND_LINK_UP &&
 800             bond_should_change_active(bond))
 801                 return bond->primary_slave;
 802
 803         bond_for_each_slave(bond, slave, iter) {
 804                 if (slave->link == BOND_LINK_UP)
 805                         return slave;
 806                 if (slave->link == BOND_LINK_BACK && IS_UP(slave->dev) &&
 807                     slave->delay < mintime) {
 808                         mintime = slave->delay;
 809                         bestslave = slave;
 810                 }
 811         }
 812
 813         return bestslave;
 814 }
 815
 816 static bool bond_should_notify_peers(struct bonding *bond)
 817 {
 818         struct slave *slave = bond->curr_active_slave;
 819
 820         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
 821                  bond->dev->name, slave ? slave->dev->name : "NULL");
 822
 823         if (!slave || !bond->send_peer_notif ||
 824             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 825                 return false;
 826
 827         return true;
 828 }
 829
 830 /**
 831  * change_active_interface - change the active slave into the specified one
 832  * @bond: our bonding struct
 833  * @new: the new slave to make the active one
 834  *
 835  * Set the new slave to the bond's settings and unset them on the old
 836  * curr_active_slave.
 837  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 838  *
 839  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 840  * because it is apparently the best available slave we have, even though its
 841  * updelay hasn't timed out yet.
 842  *
 843  * If new_active is not NULL, caller must hold bond->lock for read and
 844  * curr_slave_lock for write_bh.
 845  */
 846 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 847 {
 848         struct slave *old_active = bond->curr_active_slave;
 849
 850         if (old_active == new_active)
 851                 return;
 852
 853         if (new_active) {
 854                 new_active->jiffies = jiffies;
 855
 856                 if (new_active->link == BOND_LINK_BACK) {
 857                         if (USES_PRIMARY(bond->params.mode)) {
 858                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
 859                                         bond->dev->name, new_active->dev->name,
 860                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
 861                         }
 862
 863                         new_active->delay = 0;
 864                         new_active->link = BOND_LINK_UP;
 865
 866                         if (bond->params.mode == BOND_MODE_8023AD)
 867                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 868
 869                         if (bond_is_lb(bond))
 870                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 871                 } else {
 872                         if (USES_PRIMARY(bond->params.mode)) {
 873                                 pr_info("%s: making interface %s the new active one.\n",
 874                                         bond->dev->name, new_active->dev->name);
 875                         }
 876                 }
 877         }
 878
 879         if (USES_PRIMARY(bond->params.mode))
 880                 bond_hw_addr_swap(bond, new_active, old_active);
 881
 882         if (bond_is_lb(bond)) {
 883                 bond_alb_handle_active_change(bond, new_active);
 884                 if (old_active)
 885                         bond_set_slave_inactive_flags(old_active);
 886                 if (new_active)
 887                         bond_set_slave_active_flags(new_active);
 888         } else {
 889                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 890         }
 891
 892         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
 893                 if (old_active)
 894                         bond_set_slave_inactive_flags(old_active);
 895
 896                 if (new_active) {
 897                         bool should_notify_peers = false;
 898
 899                         bond_set_slave_active_flags(new_active);
 900
 901                         if (bond->params.fail_over_mac)
 902                                 bond_do_fail_over_mac(bond, new_active,
 903                                                       old_active);
 904
 905                         if (netif_running(bond->dev)) {
 906                                 bond->send_peer_notif =
 907                                         bond->params.num_peer_notif;
 908                                 should_notify_peers =
 909                                         bond_should_notify_peers(bond);
 910                         }
 911
 912                         write_unlock_bh(&bond->curr_slave_lock);
 913                         read_unlock(&bond->lock);
 914
 915                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 916                         if (should_notify_peers)
 917                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 918                                                          bond->dev);
 919
 920                         read_lock(&bond->lock);
 921                         write_lock_bh(&bond->curr_slave_lock);
 922                 }
 923         }
 924
 925         /* resend IGMP joins since active slave has changed or
 926          * all were sent on curr_active_slave.
 927          * resend only if bond is brought up with the affected
 928          * bonding modes and the retransmission is enabled */
 929         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 930             ((USES_PRIMARY(bond->params.mode) && new_active) ||
 931              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
 932                 bond->igmp_retrans = bond->params.resend_igmp;
 933                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 934         }
 935 }
 936
 937 /**
 938  * bond_select_active_slave - select a new active slave, if needed
 939  * @bond: our bonding struct
 940  *
 941  * This functions should be called when one of the following occurs:
 942  * - The old curr_active_slave has been released or lost its link.
 943  * - The primary_slave has got its link back.
 944  * - A slave has got its link back and there's no old curr_active_slave.
 945  *
 946  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
 947  */
 948 void bond_select_active_slave(struct bonding *bond)
 949 {
 950         struct slave *best_slave;
 951         int rv;
 952
 953         best_slave = bond_find_best_slave(bond);
 954         if (best_slave != bond->curr_active_slave) {
 955                 bond_change_active_slave(bond, best_slave);
 956                 rv = bond_set_carrier(bond);
 957                 if (!rv)
 958                         return;
 959
 960                 if (netif_carrier_ok(bond->dev)) {
 961                         pr_info("%s: first active interface up!\n",
 962                                 bond->dev->name);
 963                 } else {
 964                         pr_info("%s: now running without any active interface !\n",
 965                                 bond->dev->name);
 966                 }
 967         }
 968 }
 969
 970 #ifdef CONFIG_NET_POLL_CONTROLLER
 971 static inline int slave_enable_netpoll(struct slave *slave)
 972 {
 973         struct netpoll *np;
 974         int err = 0;
 975
 976         np = kzalloc(sizeof(*np), GFP_ATOMIC);
 977         err = -ENOMEM;
 978         if (!np)
 979                 goto out;
 980
 981         err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
 982         if (err) {
 983                 kfree(np);
 984                 goto out;
 985         }
 986         slave->np = np;
 987 out:
 988         return err;
 989 }
 990 static inline void slave_disable_netpoll(struct slave *slave)
 991 {
 992         struct netpoll *np = slave->np;
 993
 994         if (!np)
 995                 return;
 996
 997         slave->np = NULL;
 998         __netpoll_free_async(np);
 999 }
1000 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1001 {
1002         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1003                 return false;
1004         if (!slave_dev->netdev_ops->ndo_poll_controller)
1005                 return false;
1006         return true;
1007 }
1008
1009 static void bond_poll_controller(struct net_device *bond_dev)
1010 {
1011 }
1012
1013 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1014 {
1015         struct bonding *bond = netdev_priv(bond_dev);
1016         struct list_head *iter;
1017         struct slave *slave;
1018
1019         bond_for_each_slave(bond, slave, iter)
1020                 if (IS_UP(slave->dev))
1021                         slave_disable_netpoll(slave);
1022 }
1023
1024 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1025 {
1026         struct bonding *bond = netdev_priv(dev);
1027         struct list_head *iter;
1028         struct slave *slave;
1029         int err = 0;
1030
1031         bond_for_each_slave(bond, slave, iter) {
1032                 err = slave_enable_netpoll(slave);
1033                 if (err) {
1034                         bond_netpoll_cleanup(dev);
1035                         break;
1036                 }
1037         }
1038         return err;
1039 }
1040 #else
1041 static inline int slave_enable_netpoll(struct slave *slave)
1042 {
1043         return 0;
1044 }
1045 static inline void slave_disable_netpoll(struct slave *slave)
1046 {
1047 }
1048 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1049 {
1050 }
1051 #endif
1052
1053 /*---------------------------------- IOCTL ----------------------------------*/
1054
1055 static netdev_features_t bond_fix_features(struct net_device *dev,
1056                                            netdev_features_t features)
1057 {
1058         struct bonding *bond = netdev_priv(dev);
1059         struct list_head *iter;
1060         netdev_features_t mask;
1061         struct slave *slave;
1062
1063         if (!bond_has_slaves(bond)) {
1064                 /* Disable adding VLANs to empty bond. But why? --mq */
1065                 features |= NETIF_F_VLAN_CHALLENGED;
1066                 return features;
1067         }
1068
1069         mask = features;
1070         features &= ~NETIF_F_ONE_FOR_ALL;
1071         features |= NETIF_F_ALL_FOR_ALL;
1072
1073         bond_for_each_slave(bond, slave, iter) {
1074                 features = netdev_increment_features(features,
1075                                                      slave->dev->features,
1076                                                      mask);
1077         }
1078         features = netdev_add_tso_features(features, mask);
1079
1080         return features;
1081 }
1082
1083 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1084                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1085                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1086
1087 static void bond_compute_features(struct bonding *bond)
1088 {
1089         unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1090         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1091         struct net_device *bond_dev = bond->dev;
1092         struct list_head *iter;
1093         struct slave *slave;
1094         unsigned short max_hard_header_len = ETH_HLEN;
1095         unsigned int gso_max_size = GSO_MAX_SIZE;
1096         u16 gso_max_segs = GSO_MAX_SEGS;
1097
1098         if (!bond_has_slaves(bond))
1099                 goto done;
1100
1101         bond_for_each_slave(bond, slave, iter) {
1102                 vlan_features = netdev_increment_features(vlan_features,
1103                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1104
1105                 dst_release_flag &= slave->dev->priv_flags;
1106                 if (slave->dev->hard_header_len > max_hard_header_len)
1107                         max_hard_header_len = slave->dev->hard_header_len;
1108
1109                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1110                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1111         }
1112
1113 done:
1114         bond_dev->vlan_features = vlan_features;
1115         bond_dev->hard_header_len = max_hard_header_len;
1116         bond_dev->gso_max_segs = gso_max_segs;
1117         netif_set_gso_max_size(bond_dev, gso_max_size);
1118
1119         flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1120         bond_dev->priv_flags = flags | dst_release_flag;
1121
1122         netdev_change_features(bond_dev);
1123 }
1124
1125 static void bond_setup_by_slave(struct net_device *bond_dev,
1126                                 struct net_device *slave_dev)
1127 {
1128         bond_dev->header_ops        = slave_dev->header_ops;
1129
1130         bond_dev->type              = slave_dev->type;
1131         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1132         bond_dev->addr_len          = slave_dev->addr_len;
1133
1134         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1135                 slave_dev->addr_len);
1136 }
1137
1138 /* On bonding slaves other than the currently active slave, suppress
1139  * duplicates except for alb non-mcast/bcast.
1140  */
1141 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1142                                             struct slave *slave,
1143                                             struct bonding *bond)
1144 {
1145         if (bond_is_slave_inactive(slave)) {
1146                 if (bond->params.mode == BOND_MODE_ALB &&
1147                     skb->pkt_type != PACKET_BROADCAST &&
1148                     skb->pkt_type != PACKET_MULTICAST)
1149                         return false;
1150                 return true;
1151         }
1152         return false;
1153 }
1154
1155 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1156 {
1157         struct sk_buff *skb = *pskb;
1158         struct slave *slave;
1159         struct bonding *bond;
1160         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1161                           struct slave *);
1162         int ret = RX_HANDLER_ANOTHER;
1163
1164         skb = skb_share_check(skb, GFP_ATOMIC);
1165         if (unlikely(!skb))
1166                 return RX_HANDLER_CONSUMED;
1167
1168         *pskb = skb;
1169
1170         slave = bond_slave_get_rcu(skb->dev);
1171         bond = slave->bond;
1172
1173         if (bond->params.arp_interval)
1174                 slave->dev->last_rx = jiffies;
1175
1176         recv_probe = ACCESS_ONCE(bond->recv_probe);
1177         if (recv_probe) {
1178                 ret = recv_probe(skb, bond, slave);
1179                 if (ret == RX_HANDLER_CONSUMED) {
1180                         consume_skb(skb);
1181                         return ret;
1182                 }
1183         }
1184
1185         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1186                 return RX_HANDLER_EXACT;
1187         }
1188
1189         skb->dev = bond->dev;
1190
1191         if (bond->params.mode == BOND_MODE_ALB &&
1192             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1193             skb->pkt_type == PACKET_HOST) {
1194
1195                 if (unlikely(skb_cow_head(skb,
1196                                           skb->data - skb_mac_header(skb)))) {
1197                         kfree_skb(skb);
1198                         return RX_HANDLER_CONSUMED;
1199                 }
1200                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1201         }
1202
1203         return ret;
1204 }
1205
1206 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1207                                       struct net_device *slave_dev,
1208                                       struct slave *slave)
1209 {
1210         int err;
1211
1212         err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1213         if (err)
1214                 return err;
1215         slave_dev->flags |= IFF_SLAVE;
1216         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1217         return 0;
1218 }
1219
1220 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1221                                   struct net_device *slave_dev)
1222 {
1223         netdev_upper_dev_unlink(slave_dev, bond_dev);
1224         slave_dev->flags &= ~IFF_SLAVE;
1225         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1226 }
1227
1228 /* enslave device <slave> to bond device <master> */
1229 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1230 {
1231         struct bonding *bond = netdev_priv(bond_dev);
1232         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1233         struct slave *new_slave = NULL, *prev_slave;
1234         struct sockaddr addr;
1235         int link_reporting;
1236         int res = 0, i;
1237
1238         if (!bond->params.use_carrier &&
1239             slave_dev->ethtool_ops->get_link == NULL &&
1240             slave_ops->ndo_do_ioctl == NULL) {
1241                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1242                            bond_dev->name, slave_dev->name);
1243         }
1244
1245         /* already enslaved */
1246         if (slave_dev->flags & IFF_SLAVE) {
1247                 pr_debug("Error, Device was already enslaved\n");
1248                 return -EBUSY;
1249         }
1250
1251         /* vlan challenged mutual exclusion */
1252         /* no need to lock since we're protected by rtnl_lock */
1253         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1254                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1255                 if (vlan_uses_dev(bond_dev)) {
1256                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1257                                bond_dev->name, slave_dev->name, bond_dev->name);
1258                         return -EPERM;
1259                 } else {
1260                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1261                                    bond_dev->name, slave_dev->name,
1262                                    slave_dev->name, bond_dev->name);
1263                 }
1264         } else {
1265                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1266         }
1267
1268         /*
1269          * Old ifenslave binaries are no longer supported.  These can
1270          * be identified with moderate accuracy by the state of the slave:
1271          * the current ifenslave will set the interface down prior to
1272          * enslaving it; the old ifenslave will not.
1273          */
1274         if ((slave_dev->flags & IFF_UP)) {
1275                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1276                        slave_dev->name);
1277                 res = -EPERM;
1278                 goto err_undo_flags;
1279         }
1280
1281         /* set bonding device ether type by slave - bonding netdevices are
1282          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1283          * there is a need to override some of the type dependent attribs/funcs.
1284          *
1285          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1286          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1287          */
1288         if (!bond_has_slaves(bond)) {
1289                 if (bond_dev->type != slave_dev->type) {
1290                         pr_debug("%s: change device type from %d to %d\n",
1291                                  bond_dev->name,
1292                                  bond_dev->type, slave_dev->type);
1293
1294                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1295                                                        bond_dev);
1296                         res = notifier_to_errno(res);
1297                         if (res) {
1298                                 pr_err("%s: refused to change device type\n",
1299                                        bond_dev->name);
1300                                 res = -EBUSY;
1301                                 goto err_undo_flags;
1302                         }
1303
1304                         /* Flush unicast and multicast addresses */
1305                         dev_uc_flush(bond_dev);
1306                         dev_mc_flush(bond_dev);
1307
1308                         if (slave_dev->type != ARPHRD_ETHER)
1309                                 bond_setup_by_slave(bond_dev, slave_dev);
1310                         else {
1311                                 ether_setup(bond_dev);
1312                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1313                         }
1314
1315                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1316                                                  bond_dev);
1317                 }
1318         } else if (bond_dev->type != slave_dev->type) {
1319                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1320                        slave_dev->name,
1321                        slave_dev->type, bond_dev->type);
1322                 res = -EINVAL;
1323                 goto err_undo_flags;
1324         }
1325
1326         if (slave_ops->ndo_set_mac_address == NULL) {
1327                 if (!bond_has_slaves(bond)) {
1328                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1329                                    bond_dev->name);
1330                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1331                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1332                         pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active.\n",
1333                                bond_dev->name);
1334                         res = -EOPNOTSUPP;
1335                         goto err_undo_flags;
1336                 }
1337         }
1338
1339         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1340
1341         /* If this is the first slave, then we need to set the master's hardware
1342          * address to be the same as the slave's. */
1343         if (!bond_has_slaves(bond) &&
1344             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1345                 bond_set_dev_addr(bond->dev, slave_dev);
1346
1347         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1348         if (!new_slave) {
1349                 res = -ENOMEM;
1350                 goto err_undo_flags;
1351         }
1352         /*
1353          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1354          * is set via sysfs or module option if desired.
1355          */
1356         new_slave->queue_id = 0;
1357
1358         /* Save slave's original mtu and then set it to match the bond */
1359         new_slave->original_mtu = slave_dev->mtu;
1360         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1361         if (res) {
1362                 pr_debug("Error %d calling dev_set_mtu\n", res);
1363                 goto err_free;
1364         }
1365
1366         /*
1367          * Save slave's original ("permanent") mac address for modes
1368          * that need it, and for restoring it upon release, and then
1369          * set it to the master's address
1370          */
1371         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1372
1373         if (!bond->params.fail_over_mac) {
1374                 /*
1375                  * Set slave to master's mac address.  The application already
1376                  * set the master's mac address to that of the first slave
1377                  */
1378                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1379                 addr.sa_family = slave_dev->type;
1380                 res = dev_set_mac_address(slave_dev, &addr);
1381                 if (res) {
1382                         pr_debug("Error %d calling set_mac_address\n", res);
1383                         goto err_restore_mtu;
1384                 }
1385         }
1386
1387         /* open the slave since the application closed it */
1388         res = dev_open(slave_dev);
1389         if (res) {
1390                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1391                 goto err_restore_mac;
1392         }
1393
1394         new_slave->bond = bond;
1395         new_slave->dev = slave_dev;
1396         slave_dev->priv_flags |= IFF_BONDING;
1397
1398         if (bond_is_lb(bond)) {
1399                 /* bond_alb_init_slave() must be called before all other stages since
1400                  * it might fail and we do not want to have to undo everything
1401                  */
1402                 res = bond_alb_init_slave(bond, new_slave);
1403                 if (res)
1404                         goto err_close;
1405         }
1406
1407         /* If the mode USES_PRIMARY, then the following is handled by
1408          * bond_change_active_slave().
1409          */
1410         if (!USES_PRIMARY(bond->params.mode)) {
1411                 /* set promiscuity level to new slave */
1412                 if (bond_dev->flags & IFF_PROMISC) {
1413                         res = dev_set_promiscuity(slave_dev, 1);
1414                         if (res)
1415                                 goto err_close;
1416                 }
1417
1418                 /* set allmulti level to new slave */
1419                 if (bond_dev->flags & IFF_ALLMULTI) {
1420                         res = dev_set_allmulti(slave_dev, 1);
1421                         if (res)
1422                                 goto err_close;
1423                 }
1424
1425                 netif_addr_lock_bh(bond_dev);
1426
1427                 dev_mc_sync_multiple(slave_dev, bond_dev);
1428                 dev_uc_sync_multiple(slave_dev, bond_dev);
1429
1430                 netif_addr_unlock_bh(bond_dev);
1431         }
1432
1433         if (bond->params.mode == BOND_MODE_8023AD) {
1434                 /* add lacpdu mc addr to mc list */
1435                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1436
1437                 dev_mc_add(slave_dev, lacpdu_multicast);
1438         }
1439
1440         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1441         if (res) {
1442                 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1443                        bond_dev->name, slave_dev->name);
1444                 goto err_close;
1445         }
1446
1447         prev_slave = bond_last_slave(bond);
1448
1449         new_slave->delay = 0;
1450         new_slave->link_failure_count = 0;
1451
1452         bond_update_speed_duplex(new_slave);
1453
1454         new_slave->last_arp_rx = jiffies -
1455                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1456         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1457                 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
1458
1459         if (bond->params.miimon && !bond->params.use_carrier) {
1460                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1461
1462                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1463                         /*
1464                          * miimon is set but a bonded network driver
1465                          * does not support ETHTOOL/MII and
1466                          * arp_interval is not set.  Note: if
1467                          * use_carrier is enabled, we will never go
1468                          * here (because netif_carrier is always
1469                          * supported); thus, we don't need to change
1470                          * the messages for netif_carrier.
1471                          */
1472                         pr_warning("%s: Warning: 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",
1473                                bond_dev->name, slave_dev->name);
1474                 } else if (link_reporting == -1) {
1475                         /* unable get link status using mii/ethtool */
1476                         pr_warning("%s: Warning: 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",
1477                                    bond_dev->name, slave_dev->name);
1478                 }
1479         }
1480
1481         /* check for initial state */
1482         if (bond->params.miimon) {
1483                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1484                         if (bond->params.updelay) {
1485                                 new_slave->link = BOND_LINK_BACK;
1486                                 new_slave->delay = bond->params.updelay;
1487                         } else {
1488                                 new_slave->link = BOND_LINK_UP;
1489                         }
1490                 } else {
1491                         new_slave->link = BOND_LINK_DOWN;
1492                 }
1493         } else if (bond->params.arp_interval) {
1494                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1495                         BOND_LINK_UP : BOND_LINK_DOWN);
1496         } else {
1497                 new_slave->link = BOND_LINK_UP;
1498         }
1499
1500         if (new_slave->link != BOND_LINK_DOWN)
1501                 new_slave->jiffies = jiffies;
1502         pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1503                 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1504                         (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1505
1506         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1507                 /* if there is a primary slave, remember it */
1508                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1509                         bond->primary_slave = new_slave;
1510                         bond->force_primary = true;
1511                 }
1512         }
1513
1514         switch (bond->params.mode) {
1515         case BOND_MODE_ACTIVEBACKUP:
1516                 bond_set_slave_inactive_flags(new_slave);
1517                 break;
1518         case BOND_MODE_8023AD:
1519                 /* in 802.3ad mode, the internal mechanism
1520                  * will activate the slaves in the selected
1521                  * aggregator
1522                  */
1523                 bond_set_slave_inactive_flags(new_slave);
1524                 /* if this is the first slave */
1525                 if (!prev_slave) {
1526                         SLAVE_AD_INFO(new_slave).id = 1;
1527                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1528                          * can be called only after the mac address of the bond is set
1529                          */
1530                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1531                 } else {
1532                         SLAVE_AD_INFO(new_slave).id =
1533                                 SLAVE_AD_INFO(prev_slave).id + 1;
1534                 }
1535
1536                 bond_3ad_bind_slave(new_slave);
1537                 break;
1538         case BOND_MODE_TLB:
1539         case BOND_MODE_ALB:
1540                 bond_set_active_slave(new_slave);
1541                 bond_set_slave_inactive_flags(new_slave);
1542                 break;
1543         default:
1544                 pr_debug("This slave is always active in trunk mode\n");
1545
1546                 /* always active in trunk mode */
1547                 bond_set_active_slave(new_slave);
1548
1549                 /* In trunking mode there is little meaning to curr_active_slave
1550                  * anyway (it holds no special properties of the bond device),
1551                  * so we can change it without calling change_active_interface()
1552                  */
1553                 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1554                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1555
1556                 break;
1557         } /* switch(bond_mode) */
1558
1559 #ifdef CONFIG_NET_POLL_CONTROLLER
1560         slave_dev->npinfo = bond->dev->npinfo;
1561         if (slave_dev->npinfo) {
1562                 if (slave_enable_netpoll(new_slave)) {
1563                         read_unlock(&bond->lock);
1564                         pr_info("Error, %s: master_dev is using netpoll, "
1565                                  "but new slave device does not support netpoll.\n",
1566                                  bond_dev->name);
1567                         res = -EBUSY;
1568                         goto err_detach;
1569                 }
1570         }
1571 #endif
1572
1573         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1574                                          new_slave);
1575         if (res) {
1576                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1577                 goto err_detach;
1578         }
1579
1580         res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1581         if (res) {
1582                 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1583                 goto err_unregister;
1584         }
1585
1586         bond->slave_cnt++;
1587         bond_compute_features(bond);
1588         bond_set_carrier(bond);
1589
1590         if (USES_PRIMARY(bond->params.mode)) {
1591                 read_lock(&bond->lock);
1592                 write_lock_bh(&bond->curr_slave_lock);
1593                 bond_select_active_slave(bond);
1594                 write_unlock_bh(&bond->curr_slave_lock);
1595                 read_unlock(&bond->lock);
1596         }
1597
1598         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1599                 bond_dev->name, slave_dev->name,
1600                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1601                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1602
1603         /* enslave is successful */
1604         return 0;
1605
1606 /* Undo stages on error */
1607 err_unregister:
1608         netdev_rx_handler_unregister(slave_dev);
1609
1610 err_detach:
1611         if (!USES_PRIMARY(bond->params.mode))
1612                 bond_hw_addr_flush(bond_dev, slave_dev);
1613
1614         vlan_vids_del_by_dev(slave_dev, bond_dev);
1615         write_lock_bh(&bond->lock);
1616         if (bond->primary_slave == new_slave)
1617                 bond->primary_slave = NULL;
1618         if (bond->curr_active_slave == new_slave) {
1619                 bond_change_active_slave(bond, NULL);
1620                 write_unlock_bh(&bond->lock);
1621                 read_lock(&bond->lock);
1622                 write_lock_bh(&bond->curr_slave_lock);
1623                 bond_select_active_slave(bond);
1624                 write_unlock_bh(&bond->curr_slave_lock);
1625                 read_unlock(&bond->lock);
1626         } else {
1627                 write_unlock_bh(&bond->lock);
1628         }
1629         slave_disable_netpoll(new_slave);
1630
1631 err_close:
1632         slave_dev->priv_flags &= ~IFF_BONDING;
1633         dev_close(slave_dev);
1634
1635 err_restore_mac:
1636         if (!bond->params.fail_over_mac) {
1637                 /* XXX TODO - fom follow mode needs to change master's
1638                  * MAC if this slave's MAC is in use by the bond, or at
1639                  * least print a warning.
1640                  */
1641                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1642                 addr.sa_family = slave_dev->type;
1643                 dev_set_mac_address(slave_dev, &addr);
1644         }
1645
1646 err_restore_mtu:
1647         dev_set_mtu(slave_dev, new_slave->original_mtu);
1648
1649 err_free:
1650         kfree(new_slave);
1651
1652 err_undo_flags:
1653         /* Enslave of first slave has failed and we need to fix master's mac */
1654         if (!bond_has_slaves(bond) &&
1655             ether_addr_equal(bond_dev->dev_addr, slave_dev->dev_addr))
1656                 eth_hw_addr_random(bond_dev);
1657
1658         return res;
1659 }
1660
1661 /*
1662  * Try to release the slave device <slave> from the bond device <master>
1663  * It is legal to access curr_active_slave without a lock because all the function
1664  * is write-locked. If "all" is true it means that the function is being called
1665  * while destroying a bond interface and all slaves are being released.
1666  *
1667  * The rules for slave state should be:
1668  *   for Active/Backup:
1669  *     Active stays on all backups go down
1670  *   for Bonded connections:
1671  *     The first up interface should be left on and all others downed.
1672  */
1673 static int __bond_release_one(struct net_device *bond_dev,
1674                               struct net_device *slave_dev,
1675                               bool all)
1676 {
1677         struct bonding *bond = netdev_priv(bond_dev);
1678         struct slave *slave, *oldcurrent;
1679         struct sockaddr addr;
1680         int old_flags = bond_dev->flags;
1681         netdev_features_t old_features = bond_dev->features;
1682
1683         /* slave is not a slave or master is not master of this slave */
1684         if (!(slave_dev->flags & IFF_SLAVE) ||
1685             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1686                 pr_err("%s: Error: cannot release %s.\n",
1687                        bond_dev->name, slave_dev->name);
1688                 return -EINVAL;
1689         }
1690
1691         block_netpoll_tx();
1692         write_lock_bh(&bond->lock);
1693
1694         slave = bond_get_slave_by_dev(bond, slave_dev);
1695         if (!slave) {
1696                 /* not a slave of this bond */
1697                 pr_info("%s: %s not enslaved\n",
1698                         bond_dev->name, slave_dev->name);
1699                 write_unlock_bh(&bond->lock);
1700                 unblock_netpoll_tx();
1701                 return -EINVAL;
1702         }
1703
1704         write_unlock_bh(&bond->lock);
1705
1706         /* release the slave from its bond */
1707         bond->slave_cnt--;
1708
1709         bond_upper_dev_unlink(bond_dev, slave_dev);
1710         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1711          * for this slave anymore.
1712          */
1713         netdev_rx_handler_unregister(slave_dev);
1714         write_lock_bh(&bond->lock);
1715
1716         /* Inform AD package of unbinding of slave. */
1717         if (bond->params.mode == BOND_MODE_8023AD) {
1718                 /* must be called before the slave is
1719                  * detached from the list
1720                  */
1721                 bond_3ad_unbind_slave(slave);
1722         }
1723
1724         pr_info("%s: releasing %s interface %s\n",
1725                 bond_dev->name,
1726                 bond_is_active_slave(slave) ? "active" : "backup",
1727                 slave_dev->name);
1728
1729         oldcurrent = bond->curr_active_slave;
1730
1731         bond->current_arp_slave = NULL;
1732
1733         if (!all && !bond->params.fail_over_mac) {
1734                 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1735                     bond_has_slaves(bond))
1736                         pr_warn("%s: Warning: 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",
1737                                    bond_dev->name, slave_dev->name,
1738                                    slave->perm_hwaddr,
1739                                    bond_dev->name, slave_dev->name);
1740         }
1741
1742         if (bond->primary_slave == slave)
1743                 bond->primary_slave = NULL;
1744
1745         if (oldcurrent == slave)
1746                 bond_change_active_slave(bond, NULL);
1747
1748         if (bond_is_lb(bond)) {
1749                 /* Must be called only after the slave has been
1750                  * detached from the list and the curr_active_slave
1751                  * has been cleared (if our_slave == old_current),
1752                  * but before a new active slave is selected.
1753                  */
1754                 write_unlock_bh(&bond->lock);
1755                 bond_alb_deinit_slave(bond, slave);
1756                 write_lock_bh(&bond->lock);
1757         }
1758
1759         if (all) {
1760                 rcu_assign_pointer(bond->curr_active_slave, NULL);
1761         } else if (oldcurrent == slave) {
1762                 /*
1763                  * Note that we hold RTNL over this sequence, so there
1764                  * is no concern that another slave add/remove event
1765                  * will interfere.
1766                  */
1767                 write_unlock_bh(&bond->lock);
1768                 read_lock(&bond->lock);
1769                 write_lock_bh(&bond->curr_slave_lock);
1770
1771                 bond_select_active_slave(bond);
1772
1773                 write_unlock_bh(&bond->curr_slave_lock);
1774                 read_unlock(&bond->lock);
1775                 write_lock_bh(&bond->lock);
1776         }
1777
1778         if (!bond_has_slaves(bond)) {
1779                 bond_set_carrier(bond);
1780                 eth_hw_addr_random(bond_dev);
1781
1782                 if (vlan_uses_dev(bond_dev)) {
1783                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1784                                    bond_dev->name, bond_dev->name);
1785                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1786                                    bond_dev->name);
1787                 }
1788         }
1789
1790         write_unlock_bh(&bond->lock);
1791         unblock_netpoll_tx();
1792         synchronize_rcu();
1793
1794         if (!bond_has_slaves(bond)) {
1795                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1796                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1797         }
1798
1799         bond_compute_features(bond);
1800         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1801             (old_features & NETIF_F_VLAN_CHALLENGED))
1802                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1803                         bond_dev->name, slave_dev->name, bond_dev->name);
1804
1805         /* must do this from outside any spinlocks */
1806         vlan_vids_del_by_dev(slave_dev, bond_dev);
1807
1808         /* If the mode USES_PRIMARY, then this cases was handled above by
1809          * bond_change_active_slave(..., NULL)
1810          */
1811         if (!USES_PRIMARY(bond->params.mode)) {
1812                 /* unset promiscuity level from slave
1813                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1814                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1815                  * value of that flag before that change, as that was the value
1816                  * when this slave was attached, so we cache at the start of the
1817                  * function and use it here. Same goes for ALLMULTI below
1818                  */
1819                 if (old_flags & IFF_PROMISC)
1820                         dev_set_promiscuity(slave_dev, -1);
1821
1822                 /* unset allmulti level from slave */
1823                 if (old_flags & IFF_ALLMULTI)
1824                         dev_set_allmulti(slave_dev, -1);
1825
1826                 bond_hw_addr_flush(bond_dev, slave_dev);
1827         }
1828
1829         slave_disable_netpoll(slave);
1830
1831         /* close slave before restoring its mac address */
1832         dev_close(slave_dev);
1833
1834         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1835                 /* restore original ("permanent") mac address */
1836                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1837                 addr.sa_family = slave_dev->type;
1838                 dev_set_mac_address(slave_dev, &addr);
1839         }
1840
1841         dev_set_mtu(slave_dev, slave->original_mtu);
1842
1843         slave_dev->priv_flags &= ~IFF_BONDING;
1844
1845         kfree(slave);
1846
1847         return 0;  /* deletion OK */
1848 }
1849
1850 /* A wrapper used because of ndo_del_link */
1851 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1852 {
1853         return __bond_release_one(bond_dev, slave_dev, false);
1854 }
1855
1856 /*
1857 * First release a slave and then destroy the bond if no more slaves are left.
1858 * Must be under rtnl_lock when this function is called.
1859 */
1860 static int  bond_release_and_destroy(struct net_device *bond_dev,
1861                                      struct net_device *slave_dev)
1862 {
1863         struct bonding *bond = netdev_priv(bond_dev);
1864         int ret;
1865
1866         ret = bond_release(bond_dev, slave_dev);
1867         if (ret == 0 && !bond_has_slaves(bond)) {
1868                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1869                 pr_info("%s: destroying bond %s.\n",
1870                         bond_dev->name, bond_dev->name);
1871                 unregister_netdevice(bond_dev);
1872         }
1873         return ret;
1874 }
1875
1876 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1877 {
1878         struct bonding *bond = netdev_priv(bond_dev);
1879
1880         info->bond_mode = bond->params.mode;
1881         info->miimon = bond->params.miimon;
1882
1883         read_lock(&bond->lock);
1884         info->num_slaves = bond->slave_cnt;
1885         read_unlock(&bond->lock);
1886
1887         return 0;
1888 }
1889
1890 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1891 {
1892         struct bonding *bond = netdev_priv(bond_dev);
1893         struct list_head *iter;
1894         int i = 0, res = -ENODEV;
1895         struct slave *slave;
1896
1897         read_lock(&bond->lock);
1898         bond_for_each_slave(bond, slave, iter) {
1899                 if (i++ == (int)info->slave_id) {
1900                         res = 0;
1901                         strcpy(info->slave_name, slave->dev->name);
1902                         info->link = slave->link;
1903                         info->state = bond_slave_state(slave);
1904                         info->link_failure_count = slave->link_failure_count;
1905                         break;
1906                 }
1907         }
1908         read_unlock(&bond->lock);
1909
1910         return res;
1911 }
1912
1913 /*-------------------------------- Monitoring -------------------------------*/
1914
1915
1916 static int bond_miimon_inspect(struct bonding *bond)
1917 {
1918         int link_state, commit = 0;
1919         struct list_head *iter;
1920         struct slave *slave;
1921         bool ignore_updelay;
1922
1923         ignore_updelay = !bond->curr_active_slave ? true : false;
1924
1925         bond_for_each_slave(bond, slave, iter) {
1926                 slave->new_link = BOND_LINK_NOCHANGE;
1927
1928                 link_state = bond_check_dev_link(bond, slave->dev, 0);
1929
1930                 switch (slave->link) {
1931                 case BOND_LINK_UP:
1932                         if (link_state)
1933                                 continue;
1934
1935                         slave->link = BOND_LINK_FAIL;
1936                         slave->delay = bond->params.downdelay;
1937                         if (slave->delay) {
1938                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
1939                                         bond->dev->name,
1940                                         (bond->params.mode ==
1941                                          BOND_MODE_ACTIVEBACKUP) ?
1942                                         (bond_is_active_slave(slave) ?
1943                                          "active " : "backup ") : "",
1944                                         slave->dev->name,
1945                                         bond->params.downdelay * bond->params.miimon);
1946                         }
1947                         /*FALLTHRU*/
1948                 case BOND_LINK_FAIL:
1949                         if (link_state) {
1950                                 /*
1951                                  * recovered before downdelay expired
1952                                  */
1953                                 slave->link = BOND_LINK_UP;
1954                                 slave->jiffies = jiffies;
1955                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
1956                                         bond->dev->name,
1957                                         (bond->params.downdelay - slave->delay) *
1958                                         bond->params.miimon,
1959                                         slave->dev->name);
1960                                 continue;
1961                         }
1962
1963                         if (slave->delay <= 0) {
1964                                 slave->new_link = BOND_LINK_DOWN;
1965                                 commit++;
1966                                 continue;
1967                         }
1968
1969                         slave->delay--;
1970                         break;
1971
1972                 case BOND_LINK_DOWN:
1973                         if (!link_state)
1974                                 continue;
1975
1976                         slave->link = BOND_LINK_BACK;
1977                         slave->delay = bond->params.updelay;
1978
1979                         if (slave->delay) {
1980                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
1981                                         bond->dev->name, slave->dev->name,
1982                                         ignore_updelay ? 0 :
1983                                         bond->params.updelay *
1984                                         bond->params.miimon);
1985                         }
1986                         /*FALLTHRU*/
1987                 case BOND_LINK_BACK:
1988                         if (!link_state) {
1989                                 slave->link = BOND_LINK_DOWN;
1990                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
1991                                         bond->dev->name,
1992                                         (bond->params.updelay - slave->delay) *
1993                                         bond->params.miimon,
1994                                         slave->dev->name);
1995
1996                                 continue;
1997                         }
1998
1999                         if (ignore_updelay)
2000                                 slave->delay = 0;
2001
2002                         if (slave->delay <= 0) {
2003                                 slave->new_link = BOND_LINK_UP;
2004                                 commit++;
2005                                 ignore_updelay = false;
2006                                 continue;
2007                         }
2008
2009                         slave->delay--;
2010                         break;
2011                 }
2012         }
2013
2014         return commit;
2015 }
2016
2017 static void bond_miimon_commit(struct bonding *bond)
2018 {
2019         struct list_head *iter;
2020         struct slave *slave;
2021
2022         bond_for_each_slave(bond, slave, iter) {
2023                 switch (slave->new_link) {
2024                 case BOND_LINK_NOCHANGE:
2025                         continue;
2026
2027                 case BOND_LINK_UP:
2028                         slave->link = BOND_LINK_UP;
2029                         slave->jiffies = jiffies;
2030
2031                         if (bond->params.mode == BOND_MODE_8023AD) {
2032                                 /* prevent it from being the active one */
2033                                 bond_set_backup_slave(slave);
2034                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2035                                 /* make it immediately active */
2036                                 bond_set_active_slave(slave);
2037                         } else if (slave != bond->primary_slave) {
2038                                 /* prevent it from being the active one */
2039                                 bond_set_backup_slave(slave);
2040                         }
2041
2042                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2043                                 bond->dev->name, slave->dev->name,
2044                                 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2045                                 slave->duplex ? "full" : "half");
2046
2047                         /* notify ad that the link status has changed */
2048                         if (bond->params.mode == BOND_MODE_8023AD)
2049                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2050
2051                         if (bond_is_lb(bond))
2052                                 bond_alb_handle_link_change(bond, slave,
2053                                                             BOND_LINK_UP);
2054
2055                         if (!bond->curr_active_slave ||
2056                             (slave == bond->primary_slave))
2057                                 goto do_failover;
2058
2059                         continue;
2060
2061                 case BOND_LINK_DOWN:
2062                         if (slave->link_failure_count < UINT_MAX)
2063                                 slave->link_failure_count++;
2064
2065                         slave->link = BOND_LINK_DOWN;
2066
2067                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2068                             bond->params.mode == BOND_MODE_8023AD)
2069                                 bond_set_slave_inactive_flags(slave);
2070
2071                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2072                                 bond->dev->name, slave->dev->name);
2073
2074                         if (bond->params.mode == BOND_MODE_8023AD)
2075                                 bond_3ad_handle_link_change(slave,
2076                                                             BOND_LINK_DOWN);
2077
2078                         if (bond_is_lb(bond))
2079                                 bond_alb_handle_link_change(bond, slave,
2080                                                             BOND_LINK_DOWN);
2081
2082                         if (slave == bond->curr_active_slave)
2083                                 goto do_failover;
2084
2085                         continue;
2086
2087                 default:
2088                         pr_err("%s: invalid new link %d on slave %s\n",
2089                                bond->dev->name, slave->new_link,
2090                                slave->dev->name);
2091                         slave->new_link = BOND_LINK_NOCHANGE;
2092
2093                         continue;
2094                 }
2095
2096 do_failover:
2097                 ASSERT_RTNL();
2098                 block_netpoll_tx();
2099                 write_lock_bh(&bond->curr_slave_lock);
2100                 bond_select_active_slave(bond);
2101                 write_unlock_bh(&bond->curr_slave_lock);
2102                 unblock_netpoll_tx();
2103         }
2104
2105         bond_set_carrier(bond);
2106 }
2107
2108 /*
2109  * bond_mii_monitor
2110  *
2111  * Really a wrapper that splits the mii monitor into two phases: an
2112  * inspection, then (if inspection indicates something needs to be done)
2113  * an acquisition of appropriate locks followed by a commit phase to
2114  * implement whatever link state changes are indicated.
2115  */
2116 void bond_mii_monitor(struct work_struct *work)
2117 {
2118         struct bonding *bond = container_of(work, struct bonding,
2119                                             mii_work.work);
2120         bool should_notify_peers = false;
2121         unsigned long delay;
2122
2123         read_lock(&bond->lock);
2124
2125         delay = msecs_to_jiffies(bond->params.miimon);
2126
2127         if (!bond_has_slaves(bond))
2128                 goto re_arm;
2129
2130         should_notify_peers = bond_should_notify_peers(bond);
2131
2132         if (bond_miimon_inspect(bond)) {
2133                 read_unlock(&bond->lock);
2134
2135                 /* Race avoidance with bond_close cancel of workqueue */
2136                 if (!rtnl_trylock()) {
2137                         read_lock(&bond->lock);
2138                         delay = 1;
2139                         should_notify_peers = false;
2140                         goto re_arm;
2141                 }
2142
2143                 read_lock(&bond->lock);
2144
2145                 bond_miimon_commit(bond);
2146
2147                 read_unlock(&bond->lock);
2148                 rtnl_unlock();  /* might sleep, hold no other locks */
2149                 read_lock(&bond->lock);
2150         }
2151
2152 re_arm:
2153         if (bond->params.miimon)
2154                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2155
2156         read_unlock(&bond->lock);
2157
2158         if (should_notify_peers) {
2159                 if (!rtnl_trylock())
2160                         return;
2161                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2162                 rtnl_unlock();
2163         }
2164 }
2165
2166 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2167 {
2168         struct net_device *upper;
2169         struct list_head *iter;
2170         bool ret = false;
2171
2172         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2173                 return true;
2174
2175         rcu_read_lock();
2176         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2177                 if (ip == bond_confirm_addr(upper, 0, ip)) {
2178                         ret = true;
2179                         break;
2180                 }
2181         }
2182         rcu_read_unlock();
2183
2184         return ret;
2185 }
2186
2187 /*
2188  * We go to the (large) trouble of VLAN tagging ARP frames because
2189  * switches in VLAN mode (especially if ports are configured as
2190  * "native" to a VLAN) might not pass non-tagged frames.
2191  */
2192 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2193 {
2194         struct sk_buff *skb;
2195
2196         pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2197                  slave_dev->name, &dest_ip, &src_ip, vlan_id);
2198
2199         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2200                          NULL, slave_dev->dev_addr, NULL);
2201
2202         if (!skb) {
2203                 pr_err("ARP packet allocation failed\n");
2204                 return;
2205         }
2206         if (vlan_id) {
2207                 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2208                 if (!skb) {
2209                         pr_err("failed to insert VLAN tag\n");
2210                         return;
2211                 }
2212         }
2213         arp_xmit(skb);
2214 }
2215
2216
2217 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2218 {
2219         struct net_device *upper, *vlan_upper;
2220         struct list_head *iter, *vlan_iter;
2221         struct rtable *rt;
2222         __be32 *targets = bond->params.arp_targets, addr;
2223         int i, vlan_id;
2224
2225         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2226                 pr_debug("basa: target %pI4\n", &targets[i]);
2227
2228                 /* Find out through which dev should the packet go */
2229                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2230                                      RTO_ONLINK, 0);
2231                 if (IS_ERR(rt)) {
2232                         pr_debug("%s: no route to arp_ip_target %pI4\n",
2233                                  bond->dev->name, &targets[i]);
2234                         continue;
2235                 }
2236
2237                 vlan_id = 0;
2238
2239                 /* bond device itself */
2240                 if (rt->dst.dev == bond->dev)
2241                         goto found;
2242
2243                 rcu_read_lock();
2244                 /* first we search only for vlan devices. for every vlan
2245                  * found we verify its upper dev list, searching for the
2246                  * rt->dst.dev. If found we save the tag of the vlan and
2247                  * proceed to send the packet.
2248                  *
2249                  * TODO: QinQ?
2250                  */
2251                 netdev_for_each_all_upper_dev_rcu(bond->dev, vlan_upper,
2252                                                   vlan_iter) {
2253                         if (!is_vlan_dev(vlan_upper))
2254                                 continue;
2255                         netdev_for_each_all_upper_dev_rcu(vlan_upper, upper,
2256                                                           iter) {
2257                                 if (upper == rt->dst.dev) {
2258                                         vlan_id = vlan_dev_vlan_id(vlan_upper);
2259                                         rcu_read_unlock();
2260                                         goto found;
2261                                 }
2262                         }
2263                 }
2264
2265                 /* if the device we're looking for is not on top of any of
2266                  * our upper vlans, then just search for any dev that
2267                  * matches, and in case it's a vlan - save the id
2268                  */
2269                 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2270                         if (upper == rt->dst.dev) {
2271                                 /* if it's a vlan - get its VID */
2272                                 if (is_vlan_dev(upper))
2273                                         vlan_id = vlan_dev_vlan_id(upper);
2274
2275                                 rcu_read_unlock();
2276                                 goto found;
2277                         }
2278                 }
2279                 rcu_read_unlock();
2280
2281                 /* Not our device - skip */
2282                 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2283                          bond->dev->name, &targets[i],
2284                          rt->dst.dev ? rt->dst.dev->name : "NULL");
2285
2286                 ip_rt_put(rt);
2287                 continue;
2288
2289 found:
2290                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2291                 ip_rt_put(rt);
2292                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2293                               addr, vlan_id);
2294         }
2295 }
2296
2297 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2298 {
2299         int i;
2300
2301         if (!sip || !bond_has_this_ip(bond, tip)) {
2302                 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2303                 return;
2304         }
2305
2306         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2307         if (i == -1) {
2308                 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2309                 return;
2310         }
2311         slave->last_arp_rx = jiffies;
2312         slave->target_last_arp_rx[i] = jiffies;
2313 }
2314
2315 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2316                  struct slave *slave)
2317 {
2318         struct arphdr *arp = (struct arphdr *)skb->data;
2319         unsigned char *arp_ptr;
2320         __be32 sip, tip;
2321         int alen;
2322
2323         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2324                 return RX_HANDLER_ANOTHER;
2325
2326         read_lock(&bond->lock);
2327
2328         if (!slave_do_arp_validate(bond, slave))
2329                 goto out_unlock;
2330
2331         alen = arp_hdr_len(bond->dev);
2332
2333         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2334                  bond->dev->name, skb->dev->name);
2335
2336         if (alen > skb_headlen(skb)) {
2337                 arp = kmalloc(alen, GFP_ATOMIC);
2338                 if (!arp)
2339                         goto out_unlock;
2340                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2341                         goto out_unlock;
2342         }
2343
2344         if (arp->ar_hln != bond->dev->addr_len ||
2345             skb->pkt_type == PACKET_OTHERHOST ||
2346             skb->pkt_type == PACKET_LOOPBACK ||
2347             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2348             arp->ar_pro != htons(ETH_P_IP) ||
2349             arp->ar_pln != 4)
2350                 goto out_unlock;
2351
2352         arp_ptr = (unsigned char *)(arp + 1);
2353         arp_ptr += bond->dev->addr_len;
2354         memcpy(&sip, arp_ptr, 4);
2355         arp_ptr += 4 + bond->dev->addr_len;
2356         memcpy(&tip, arp_ptr, 4);
2357
2358         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2359                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2360                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2361                  &sip, &tip);
2362
2363         /*
2364          * Backup slaves won't see the ARP reply, but do come through
2365          * here for each ARP probe (so we swap the sip/tip to validate
2366          * the probe).  In a "redundant switch, common router" type of
2367          * configuration, the ARP probe will (hopefully) travel from
2368          * the active, through one switch, the router, then the other
2369          * switch before reaching the backup.
2370          *
2371          * We 'trust' the arp requests if there is an active slave and
2372          * it received valid arp reply(s) after it became active. This
2373          * is done to avoid endless looping when we can't reach the
2374          * arp_ip_target and fool ourselves with our own arp requests.
2375          */
2376         if (bond_is_active_slave(slave))
2377                 bond_validate_arp(bond, slave, sip, tip);
2378         else if (bond->curr_active_slave &&
2379                  time_after(slave_last_rx(bond, bond->curr_active_slave),
2380                             bond->curr_active_slave->jiffies))
2381                 bond_validate_arp(bond, slave, tip, sip);
2382
2383 out_unlock:
2384         read_unlock(&bond->lock);
2385         if (arp != (struct arphdr *)skb->data)
2386                 kfree(arp);
2387         return RX_HANDLER_ANOTHER;
2388 }
2389
2390 /* function to verify if we're in the arp_interval timeslice, returns true if
2391  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2392  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2393  */
2394 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2395                                   int mod)
2396 {
2397         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2398
2399         return time_in_range(jiffies,
2400                              last_act - delta_in_ticks,
2401                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2402 }
2403
2404 /*
2405  * this function is called regularly to monitor each slave's link
2406  * ensuring that traffic is being sent and received when arp monitoring
2407  * is used in load-balancing mode. if the adapter has been dormant, then an
2408  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2409  * arp monitoring in active backup mode.
2410  */
2411 void bond_loadbalance_arp_mon(struct work_struct *work)
2412 {
2413         struct bonding *bond = container_of(work, struct bonding,
2414                                             arp_work.work);
2415         struct slave *slave, *oldcurrent;
2416         struct list_head *iter;
2417         int do_failover = 0;
2418
2419         read_lock(&bond->lock);
2420
2421         if (!bond_has_slaves(bond))
2422                 goto re_arm;
2423
2424         oldcurrent = bond->curr_active_slave;
2425         /* see if any of the previous devices are up now (i.e. they have
2426          * xmt and rcv traffic). the curr_active_slave does not come into
2427          * the picture unless it is null. also, slave->jiffies is not needed
2428          * here because we send an arp on each slave and give a slave as
2429          * long as it needs to get the tx/rx within the delta.
2430          * TODO: what about up/down delay in arp mode? it wasn't here before
2431          *       so it can wait
2432          */
2433         bond_for_each_slave(bond, slave, iter) {
2434                 unsigned long trans_start = dev_trans_start(slave->dev);
2435
2436                 if (slave->link != BOND_LINK_UP) {
2437                         if (bond_time_in_interval(bond, trans_start, 1) &&
2438                             bond_time_in_interval(bond, slave->dev->last_rx, 1)) {
2439
2440                                 slave->link  = BOND_LINK_UP;
2441                                 bond_set_active_slave(slave);
2442
2443                                 /* primary_slave has no meaning in round-robin
2444                                  * mode. the window of a slave being up and
2445                                  * curr_active_slave being null after enslaving
2446                                  * is closed.
2447                                  */
2448                                 if (!oldcurrent) {
2449                                         pr_info("%s: link status definitely up for interface %s, ",
2450                                                 bond->dev->name,
2451                                                 slave->dev->name);
2452                                         do_failover = 1;
2453                                 } else {
2454                                         pr_info("%s: interface %s is now up\n",
2455                                                 bond->dev->name,
2456                                                 slave->dev->name);
2457                                 }
2458                         }
2459                 } else {
2460                         /* slave->link == BOND_LINK_UP */
2461
2462                         /* not all switches will respond to an arp request
2463                          * when the source ip is 0, so don't take the link down
2464                          * if we don't know our ip yet
2465                          */
2466                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2467                             !bond_time_in_interval(bond, slave->dev->last_rx, 2)) {
2468
2469                                 slave->link  = BOND_LINK_DOWN;
2470                                 bond_set_backup_slave(slave);
2471
2472                                 if (slave->link_failure_count < UINT_MAX)
2473                                         slave->link_failure_count++;
2474
2475                                 pr_info("%s: interface %s is now down.\n",
2476                                         bond->dev->name,
2477                                         slave->dev->name);
2478
2479                                 if (slave == oldcurrent)
2480                                         do_failover = 1;
2481                         }
2482                 }
2483
2484                 /* note: if switch is in round-robin mode, all links
2485                  * must tx arp to ensure all links rx an arp - otherwise
2486                  * links may oscillate or not come up at all; if switch is
2487                  * in something like xor mode, there is nothing we can
2488                  * do - all replies will be rx'ed on same link causing slaves
2489                  * to be unstable during low/no traffic periods
2490                  */
2491                 if (IS_UP(slave->dev))
2492                         bond_arp_send_all(bond, slave);
2493         }
2494
2495         if (do_failover) {
2496                 block_netpoll_tx();
2497                 write_lock_bh(&bond->curr_slave_lock);
2498
2499                 bond_select_active_slave(bond);
2500
2501                 write_unlock_bh(&bond->curr_slave_lock);
2502                 unblock_netpoll_tx();
2503         }
2504
2505 re_arm:
2506         if (bond->params.arp_interval)
2507                 queue_delayed_work(bond->wq, &bond->arp_work,
2508                                    msecs_to_jiffies(bond->params.arp_interval));
2509
2510         read_unlock(&bond->lock);
2511 }
2512
2513 /*
2514  * Called to inspect slaves for active-backup mode ARP monitor link state
2515  * changes.  Sets new_link in slaves to specify what action should take
2516  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2517  * to link states must be committed.
2518  *
2519  * Called with bond->lock held for read.
2520  */
2521 static int bond_ab_arp_inspect(struct bonding *bond)
2522 {
2523         unsigned long trans_start, last_rx;
2524         struct list_head *iter;
2525         struct slave *slave;
2526         int commit = 0;
2527
2528         bond_for_each_slave(bond, slave, iter) {
2529                 slave->new_link = BOND_LINK_NOCHANGE;
2530                 last_rx = slave_last_rx(bond, slave);
2531
2532                 if (slave->link != BOND_LINK_UP) {
2533                         if (bond_time_in_interval(bond, last_rx, 1)) {
2534                                 slave->new_link = BOND_LINK_UP;
2535                                 commit++;
2536                         }
2537                         continue;
2538                 }
2539
2540                 /*
2541                  * Give slaves 2*delta after being enslaved or made
2542                  * active.  This avoids bouncing, as the last receive
2543                  * times need a full ARP monitor cycle to be updated.
2544                  */
2545                 if (bond_time_in_interval(bond, slave->jiffies, 2))
2546                         continue;
2547
2548                 /*
2549                  * Backup slave is down if:
2550                  * - No current_arp_slave AND
2551                  * - more than 3*delta since last receive AND
2552                  * - the bond has an IP address
2553                  *
2554                  * Note: a non-null current_arp_slave indicates
2555                  * the curr_active_slave went down and we are
2556                  * searching for a new one; under this condition
2557                  * we only take the curr_active_slave down - this
2558                  * gives each slave a chance to tx/rx traffic
2559                  * before being taken out
2560                  */
2561                 if (!bond_is_active_slave(slave) &&
2562                     !bond->current_arp_slave &&
2563                     !bond_time_in_interval(bond, last_rx, 3)) {
2564                         slave->new_link = BOND_LINK_DOWN;
2565                         commit++;
2566                 }
2567
2568                 /*
2569                  * Active slave is down if:
2570                  * - more than 2*delta since transmitting OR
2571                  * - (more than 2*delta since receive AND
2572                  *    the bond has an IP address)
2573                  */
2574                 trans_start = dev_trans_start(slave->dev);
2575                 if (bond_is_active_slave(slave) &&
2576                     (!bond_time_in_interval(bond, trans_start, 2) ||
2577                      !bond_time_in_interval(bond, last_rx, 2))) {
2578                         slave->new_link = BOND_LINK_DOWN;
2579                         commit++;
2580                 }
2581         }
2582
2583         return commit;
2584 }
2585
2586 /*
2587  * Called to commit link state changes noted by inspection step of
2588  * active-backup mode ARP monitor.
2589  *
2590  * Called with RTNL and bond->lock for read.
2591  */
2592 static void bond_ab_arp_commit(struct bonding *bond)
2593 {
2594         unsigned long trans_start;
2595         struct list_head *iter;
2596         struct slave *slave;
2597
2598         bond_for_each_slave(bond, slave, iter) {
2599                 switch (slave->new_link) {
2600                 case BOND_LINK_NOCHANGE:
2601                         continue;
2602
2603                 case BOND_LINK_UP:
2604                         trans_start = dev_trans_start(slave->dev);
2605                         if (bond->curr_active_slave != slave ||
2606                             (!bond->curr_active_slave &&
2607                              bond_time_in_interval(bond, trans_start, 1))) {
2608                                 slave->link = BOND_LINK_UP;
2609                                 if (bond->current_arp_slave) {
2610                                         bond_set_slave_inactive_flags(
2611                                                 bond->current_arp_slave);
2612                                         bond->current_arp_slave = NULL;
2613                                 }
2614
2615                                 pr_info("%s: link status definitely up for interface %s.\n",
2616                                         bond->dev->name, slave->dev->name);
2617
2618                                 if (!bond->curr_active_slave ||
2619                                     (slave == bond->primary_slave))
2620                                         goto do_failover;
2621
2622                         }
2623
2624                         continue;
2625
2626                 case BOND_LINK_DOWN:
2627                         if (slave->link_failure_count < UINT_MAX)
2628                                 slave->link_failure_count++;
2629
2630                         slave->link = BOND_LINK_DOWN;
2631                         bond_set_slave_inactive_flags(slave);
2632
2633                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2634                                 bond->dev->name, slave->dev->name);
2635
2636                         if (slave == bond->curr_active_slave) {
2637                                 bond->current_arp_slave = NULL;
2638                                 goto do_failover;
2639                         }
2640
2641                         continue;
2642
2643                 default:
2644                         pr_err("%s: impossible: new_link %d on slave %s\n",
2645                                bond->dev->name, slave->new_link,
2646                                slave->dev->name);
2647                         continue;
2648                 }
2649
2650 do_failover:
2651                 ASSERT_RTNL();
2652                 block_netpoll_tx();
2653                 write_lock_bh(&bond->curr_slave_lock);
2654                 bond_select_active_slave(bond);
2655                 write_unlock_bh(&bond->curr_slave_lock);
2656                 unblock_netpoll_tx();
2657         }
2658
2659         bond_set_carrier(bond);
2660 }
2661
2662 /*
2663  * Send ARP probes for active-backup mode ARP monitor.
2664  *
2665  * Called with bond->lock held for read.
2666  */
2667 static void bond_ab_arp_probe(struct bonding *bond)
2668 {
2669         struct slave *slave, *before = NULL, *new_slave = NULL;
2670         struct list_head *iter;
2671         bool found = false;
2672
2673         read_lock(&bond->curr_slave_lock);
2674
2675         if (bond->current_arp_slave && bond->curr_active_slave)
2676                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2677                         bond->current_arp_slave->dev->name,
2678                         bond->curr_active_slave->dev->name);
2679
2680         if (bond->curr_active_slave) {
2681                 bond_arp_send_all(bond, bond->curr_active_slave);
2682                 read_unlock(&bond->curr_slave_lock);
2683                 return;
2684         }
2685
2686         read_unlock(&bond->curr_slave_lock);
2687
2688         /* if we don't have a curr_active_slave, search for the next available
2689          * backup slave from the current_arp_slave and make it the candidate
2690          * for becoming the curr_active_slave
2691          */
2692
2693         if (!bond->current_arp_slave) {
2694                 bond->current_arp_slave = bond_first_slave(bond);
2695                 if (!bond->current_arp_slave)
2696                         return;
2697         }
2698
2699         bond_set_slave_inactive_flags(bond->current_arp_slave);
2700
2701         bond_for_each_slave(bond, slave, iter) {
2702                 if (!found && !before && IS_UP(slave->dev))
2703                         before = slave;
2704
2705                 if (found && !new_slave && IS_UP(slave->dev))
2706                         new_slave = slave;
2707                 /* if the link state is up at this point, we
2708                  * mark it down - this can happen if we have
2709                  * simultaneous link failures and
2710                  * reselect_active_interface doesn't make this
2711                  * one the current slave so it is still marked
2712                  * up when it is actually down
2713                  */
2714                 if (!IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
2715                         slave->link = BOND_LINK_DOWN;
2716                         if (slave->link_failure_count < UINT_MAX)
2717                                 slave->link_failure_count++;
2718
2719                         bond_set_slave_inactive_flags(slave);
2720
2721                         pr_info("%s: backup interface %s is now down.\n",
2722                                 bond->dev->name, slave->dev->name);
2723                 }
2724                 if (slave == bond->current_arp_slave)
2725                         found = true;
2726         }
2727
2728         if (!new_slave && before)
2729                 new_slave = before;
2730
2731         if (!new_slave)
2732                 return;
2733
2734         new_slave->link = BOND_LINK_BACK;
2735         bond_set_slave_active_flags(new_slave);
2736         bond_arp_send_all(bond, new_slave);
2737         new_slave->jiffies = jiffies;
2738         bond->current_arp_slave = new_slave;
2739
2740 }
2741
2742 void bond_activebackup_arp_mon(struct work_struct *work)
2743 {
2744         struct bonding *bond = container_of(work, struct bonding,
2745                                             arp_work.work);
2746         bool should_notify_peers = false;
2747         int delta_in_ticks;
2748
2749         read_lock(&bond->lock);
2750
2751         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2752
2753         if (!bond_has_slaves(bond))
2754                 goto re_arm;
2755
2756         should_notify_peers = bond_should_notify_peers(bond);
2757
2758         if (bond_ab_arp_inspect(bond)) {
2759                 read_unlock(&bond->lock);
2760
2761                 /* Race avoidance with bond_close flush of workqueue */
2762                 if (!rtnl_trylock()) {
2763                         read_lock(&bond->lock);
2764                         delta_in_ticks = 1;
2765                         should_notify_peers = false;
2766                         goto re_arm;
2767                 }
2768
2769                 read_lock(&bond->lock);
2770
2771                 bond_ab_arp_commit(bond);
2772
2773                 read_unlock(&bond->lock);
2774                 rtnl_unlock();
2775                 read_lock(&bond->lock);
2776         }
2777
2778         bond_ab_arp_probe(bond);
2779
2780 re_arm:
2781         if (bond->params.arp_interval)
2782                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2783
2784         read_unlock(&bond->lock);
2785
2786         if (should_notify_peers) {
2787                 if (!rtnl_trylock())
2788                         return;
2789                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2790                 rtnl_unlock();
2791         }
2792 }
2793
2794 /*-------------------------- netdev event handling --------------------------*/
2795
2796 /*
2797  * Change device name
2798  */
2799 static int bond_event_changename(struct bonding *bond)
2800 {
2801         bond_remove_proc_entry(bond);
2802         bond_create_proc_entry(bond);
2803
2804         bond_debug_reregister(bond);
2805
2806         return NOTIFY_DONE;
2807 }
2808
2809 static int bond_master_netdev_event(unsigned long event,
2810                                     struct net_device *bond_dev)
2811 {
2812         struct bonding *event_bond = netdev_priv(bond_dev);
2813
2814         switch (event) {
2815         case NETDEV_CHANGENAME:
2816                 return bond_event_changename(event_bond);
2817         case NETDEV_UNREGISTER:
2818                 bond_remove_proc_entry(event_bond);
2819                 break;
2820         case NETDEV_REGISTER:
2821                 bond_create_proc_entry(event_bond);
2822                 break;
2823         case NETDEV_NOTIFY_PEERS:
2824                 if (event_bond->send_peer_notif)
2825                         event_bond->send_peer_notif--;
2826                 break;
2827         default:
2828                 break;
2829         }
2830
2831         return NOTIFY_DONE;
2832 }
2833
2834 static int bond_slave_netdev_event(unsigned long event,
2835                                    struct net_device *slave_dev)
2836 {
2837         struct slave *slave = bond_slave_get_rtnl(slave_dev);
2838         struct bonding *bond;
2839         struct net_device *bond_dev;
2840         u32 old_speed;
2841         u8 old_duplex;
2842
2843         /* A netdev event can be generated while enslaving a device
2844          * before netdev_rx_handler_register is called in which case
2845          * slave will be NULL
2846          */
2847         if (!slave)
2848                 return NOTIFY_DONE;
2849         bond_dev = slave->bond->dev;
2850         bond = slave->bond;
2851
2852         switch (event) {
2853         case NETDEV_UNREGISTER:
2854                 if (bond_dev->type != ARPHRD_ETHER)
2855                         bond_release_and_destroy(bond_dev, slave_dev);
2856                 else
2857                         bond_release(bond_dev, slave_dev);
2858                 break;
2859         case NETDEV_UP:
2860         case NETDEV_CHANGE:
2861                 old_speed = slave->speed;
2862                 old_duplex = slave->duplex;
2863
2864                 bond_update_speed_duplex(slave);
2865
2866                 if (bond->params.mode == BOND_MODE_8023AD) {
2867                         if (old_speed != slave->speed)
2868                                 bond_3ad_adapter_speed_changed(slave);
2869                         if (old_duplex != slave->duplex)
2870                                 bond_3ad_adapter_duplex_changed(slave);
2871                 }
2872                 break;
2873         case NETDEV_DOWN:
2874                 /*
2875                  * ... Or is it this?
2876                  */
2877                 break;
2878         case NETDEV_CHANGEMTU:
2879                 /*
2880                  * TODO: Should slaves be allowed to
2881                  * independently alter their MTU?  For
2882                  * an active-backup bond, slaves need
2883                  * not be the same type of device, so
2884                  * MTUs may vary.  For other modes,
2885                  * slaves arguably should have the
2886                  * same MTUs. To do this, we'd need to
2887                  * take over the slave's change_mtu
2888                  * function for the duration of their
2889                  * servitude.
2890                  */
2891                 break;
2892         case NETDEV_CHANGENAME:
2893                 /*
2894                  * TODO: handle changing the primary's name
2895                  */
2896                 break;
2897         case NETDEV_FEAT_CHANGE:
2898                 bond_compute_features(bond);
2899                 break;
2900         case NETDEV_RESEND_IGMP:
2901                 /* Propagate to master device */
2902                 call_netdevice_notifiers(event, slave->bond->dev);
2903                 break;
2904         default:
2905                 break;
2906         }
2907
2908         return NOTIFY_DONE;
2909 }
2910
2911 /*
2912  * bond_netdev_event: handle netdev notifier chain events.
2913  *
2914  * This function receives events for the netdev chain.  The caller (an
2915  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2916  * locks for us to safely manipulate the slave devices (RTNL lock,
2917  * dev_probe_lock).
2918  */
2919 static int bond_netdev_event(struct notifier_block *this,
2920                              unsigned long event, void *ptr)
2921 {
2922         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2923
2924         pr_debug("event_dev: %s, event: %lx\n",
2925                  event_dev ? event_dev->name : "None",
2926                  event);
2927
2928         if (!(event_dev->priv_flags & IFF_BONDING))
2929                 return NOTIFY_DONE;
2930
2931         if (event_dev->flags & IFF_MASTER) {
2932                 pr_debug("IFF_MASTER\n");
2933                 return bond_master_netdev_event(event, event_dev);
2934         }
2935
2936         if (event_dev->flags & IFF_SLAVE) {
2937                 pr_debug("IFF_SLAVE\n");
2938                 return bond_slave_netdev_event(event, event_dev);
2939         }
2940
2941         return NOTIFY_DONE;
2942 }
2943
2944 static struct notifier_block bond_netdev_notifier = {
2945         .notifier_call = bond_netdev_event,
2946 };
2947
2948 /*---------------------------- Hashing Policies -----------------------------*/
2949
2950 /* L2 hash helper */
2951 static inline u32 bond_eth_hash(struct sk_buff *skb)
2952 {
2953         struct ethhdr *data = (struct ethhdr *)skb->data;
2954
2955         if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
2956                 return data->h_dest[5] ^ data->h_source[5];
2957
2958         return 0;
2959 }
2960
2961 /* Extract the appropriate headers based on bond's xmit policy */
2962 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
2963                               struct flow_keys *fk)
2964 {
2965         const struct ipv6hdr *iph6;
2966         const struct iphdr *iph;
2967         int noff, proto = -1;
2968
2969         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
2970                 return skb_flow_dissect(skb, fk);
2971
2972         fk->ports = 0;
2973         noff = skb_network_offset(skb);
2974         if (skb->protocol == htons(ETH_P_IP)) {
2975                 if (!pskb_may_pull(skb, noff + sizeof(*iph)))
2976                         return false;
2977                 iph = ip_hdr(skb);
2978                 fk->src = iph->saddr;
2979                 fk->dst = iph->daddr;
2980                 noff += iph->ihl << 2;
2981                 if (!ip_is_fragment(iph))
2982                         proto = iph->protocol;
2983         } else if (skb->protocol == htons(ETH_P_IPV6)) {
2984                 if (!pskb_may_pull(skb, noff + sizeof(*iph6)))
2985                         return false;
2986                 iph6 = ipv6_hdr(skb);
2987                 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
2988                 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
2989                 noff += sizeof(*iph6);
2990                 proto = iph6->nexthdr;
2991         } else {
2992                 return false;
2993         }
2994         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
2995                 fk->ports = skb_flow_get_ports(skb, noff, proto);
2996
2997         return true;
2998 }
2999
3000 /**
3001  * bond_xmit_hash - generate a hash value based on the xmit policy
3002  * @bond: bonding device
3003  * @skb: buffer to use for headers
3004  * @count: modulo value
3005  *
3006  * This function will extract the necessary headers from the skb buffer and use
3007  * them to generate a hash based on the xmit_policy set in the bonding device
3008  * which will be reduced modulo count before returning.
3009  */
3010 int bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, int count)
3011 {
3012         struct flow_keys flow;
3013         u32 hash;
3014
3015         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3016             !bond_flow_dissect(bond, skb, &flow))
3017                 return bond_eth_hash(skb) % count;
3018
3019         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3020             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3021                 hash = bond_eth_hash(skb);
3022         else
3023                 hash = (__force u32)flow.ports;
3024         hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
3025         hash ^= (hash >> 16);
3026         hash ^= (hash >> 8);
3027
3028         return hash % count;
3029 }
3030
3031 /*-------------------------- Device entry points ----------------------------*/
3032
3033 static void bond_work_init_all(struct bonding *bond)
3034 {
3035         INIT_DELAYED_WORK(&bond->mcast_work,
3036                           bond_resend_igmp_join_requests_delayed);
3037         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3038         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3039         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3040                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3041         else
3042                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3043         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3044 }
3045
3046 static void bond_work_cancel_all(struct bonding *bond)
3047 {
3048         cancel_delayed_work_sync(&bond->mii_work);
3049         cancel_delayed_work_sync(&bond->arp_work);
3050         cancel_delayed_work_sync(&bond->alb_work);
3051         cancel_delayed_work_sync(&bond->ad_work);
3052         cancel_delayed_work_sync(&bond->mcast_work);
3053 }
3054
3055 static int bond_open(struct net_device *bond_dev)
3056 {
3057         struct bonding *bond = netdev_priv(bond_dev);
3058         struct list_head *iter;
3059         struct slave *slave;
3060
3061         /* reset slave->backup and slave->inactive */
3062         read_lock(&bond->lock);
3063         if (bond_has_slaves(bond)) {
3064                 read_lock(&bond->curr_slave_lock);
3065                 bond_for_each_slave(bond, slave, iter) {
3066                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3067                                 && (slave != bond->curr_active_slave)) {
3068                                 bond_set_slave_inactive_flags(slave);
3069                         } else {
3070                                 bond_set_slave_active_flags(slave);
3071                         }
3072                 }
3073                 read_unlock(&bond->curr_slave_lock);
3074         }
3075         read_unlock(&bond->lock);
3076
3077         bond_work_init_all(bond);
3078
3079         if (bond_is_lb(bond)) {
3080                 /* bond_alb_initialize must be called before the timer
3081                  * is started.
3082                  */
3083                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3084                         return -ENOMEM;
3085                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3086         }
3087
3088         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3089                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3090
3091         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3092                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3093                 if (bond->params.arp_validate)
3094                         bond->recv_probe = bond_arp_rcv;
3095         }
3096
3097         if (bond->params.mode == BOND_MODE_8023AD) {
3098                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3099                 /* register to receive LACPDUs */
3100                 bond->recv_probe = bond_3ad_lacpdu_recv;
3101                 bond_3ad_initiate_agg_selection(bond, 1);
3102         }
3103
3104         return 0;
3105 }
3106
3107 static int bond_close(struct net_device *bond_dev)
3108 {
3109         struct bonding *bond = netdev_priv(bond_dev);
3110
3111         bond_work_cancel_all(bond);
3112         bond->send_peer_notif = 0;
3113         if (bond_is_lb(bond))
3114                 bond_alb_deinitialize(bond);
3115         bond->recv_probe = NULL;
3116
3117         return 0;
3118 }
3119
3120 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3121                                                 struct rtnl_link_stats64 *stats)
3122 {
3123         struct bonding *bond = netdev_priv(bond_dev);
3124         struct rtnl_link_stats64 temp;
3125         struct list_head *iter;
3126         struct slave *slave;
3127
3128         memset(stats, 0, sizeof(*stats));
3129
3130         read_lock_bh(&bond->lock);
3131         bond_for_each_slave(bond, slave, iter) {
3132                 const struct rtnl_link_stats64 *sstats =
3133                         dev_get_stats(slave->dev, &temp);
3134
3135                 stats->rx_packets += sstats->rx_packets;
3136                 stats->rx_bytes += sstats->rx_bytes;
3137                 stats->rx_errors += sstats->rx_errors;
3138                 stats->rx_dropped += sstats->rx_dropped;
3139
3140                 stats->tx_packets += sstats->tx_packets;
3141                 stats->tx_bytes += sstats->tx_bytes;
3142                 stats->tx_errors += sstats->tx_errors;
3143                 stats->tx_dropped += sstats->tx_dropped;
3144
3145                 stats->multicast += sstats->multicast;
3146                 stats->collisions += sstats->collisions;
3147
3148                 stats->rx_length_errors += sstats->rx_length_errors;
3149                 stats->rx_over_errors += sstats->rx_over_errors;
3150                 stats->rx_crc_errors += sstats->rx_crc_errors;
3151                 stats->rx_frame_errors += sstats->rx_frame_errors;
3152                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3153                 stats->rx_missed_errors += sstats->rx_missed_errors;
3154
3155                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3156                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3157                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3158                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3159                 stats->tx_window_errors += sstats->tx_window_errors;
3160         }
3161         read_unlock_bh(&bond->lock);
3162
3163         return stats;
3164 }
3165
3166 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3167 {
3168         struct bonding *bond = netdev_priv(bond_dev);
3169         struct net_device *slave_dev = NULL;
3170         struct ifbond k_binfo;
3171         struct ifbond __user *u_binfo = NULL;
3172         struct ifslave k_sinfo;
3173         struct ifslave __user *u_sinfo = NULL;
3174         struct mii_ioctl_data *mii = NULL;
3175         struct net *net;
3176         int res = 0;
3177
3178         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3179
3180         switch (cmd) {
3181         case SIOCGMIIPHY:
3182                 mii = if_mii(ifr);
3183                 if (!mii)
3184                         return -EINVAL;
3185
3186                 mii->phy_id = 0;
3187                 /* Fall Through */
3188         case SIOCGMIIREG:
3189                 /*
3190                  * We do this again just in case we were called by SIOCGMIIREG
3191                  * instead of SIOCGMIIPHY.
3192                  */
3193                 mii = if_mii(ifr);
3194                 if (!mii)
3195                         return -EINVAL;
3196
3197
3198                 if (mii->reg_num == 1) {
3199                         mii->val_out = 0;
3200                         read_lock(&bond->lock);
3201                         read_lock(&bond->curr_slave_lock);
3202                         if (netif_carrier_ok(bond->dev))
3203                                 mii->val_out = BMSR_LSTATUS;
3204
3205                         read_unlock(&bond->curr_slave_lock);
3206                         read_unlock(&bond->lock);
3207                 }
3208
3209                 return 0;
3210         case BOND_INFO_QUERY_OLD:
3211         case SIOCBONDINFOQUERY:
3212                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3213
3214                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3215                         return -EFAULT;
3216
3217                 res = bond_info_query(bond_dev, &k_binfo);
3218                 if (res == 0 &&
3219                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3220                         return -EFAULT;
3221
3222                 return res;
3223         case BOND_SLAVE_INFO_QUERY_OLD:
3224         case SIOCBONDSLAVEINFOQUERY:
3225                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3226
3227                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3228                         return -EFAULT;
3229
3230                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3231                 if (res == 0 &&
3232                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3233                         return -EFAULT;
3234
3235                 return res;
3236         default:
3237                 /* Go on */
3238                 break;
3239         }
3240
3241         net = dev_net(bond_dev);
3242
3243         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3244                 return -EPERM;
3245
3246         slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3247
3248         pr_debug("slave_dev=%p:\n", slave_dev);
3249
3250         if (!slave_dev)
3251                 res = -ENODEV;
3252         else {
3253                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3254                 switch (cmd) {
3255                 case BOND_ENSLAVE_OLD:
3256                 case SIOCBONDENSLAVE:
3257                         res = bond_enslave(bond_dev, slave_dev);
3258                         break;
3259                 case BOND_RELEASE_OLD:
3260                 case SIOCBONDRELEASE:
3261                         res = bond_release(bond_dev, slave_dev);
3262                         break;
3263                 case BOND_SETHWADDR_OLD:
3264                 case SIOCBONDSETHWADDR:
3265                         bond_set_dev_addr(bond_dev, slave_dev);
3266                         res = 0;
3267                         break;
3268                 case BOND_CHANGE_ACTIVE_OLD:
3269                 case SIOCBONDCHANGEACTIVE:
3270                         res = bond_option_active_slave_set(bond, slave_dev);
3271                         break;
3272                 default:
3273                         res = -EOPNOTSUPP;
3274                 }
3275
3276                 dev_put(slave_dev);
3277         }
3278
3279         return res;
3280 }
3281
3282 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3283 {
3284         struct bonding *bond = netdev_priv(bond_dev);
3285
3286         if (change & IFF_PROMISC)
3287                 bond_set_promiscuity(bond,
3288                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3289
3290         if (change & IFF_ALLMULTI)
3291                 bond_set_allmulti(bond,
3292                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3293 }
3294
3295 static void bond_set_rx_mode(struct net_device *bond_dev)
3296 {
3297         struct bonding *bond = netdev_priv(bond_dev);
3298         struct list_head *iter;
3299         struct slave *slave;
3300
3301
3302         rcu_read_lock();
3303         if (USES_PRIMARY(bond->params.mode)) {
3304                 slave = rcu_dereference(bond->curr_active_slave);
3305                 if (slave) {
3306                         dev_uc_sync(slave->dev, bond_dev);
3307                         dev_mc_sync(slave->dev, bond_dev);
3308                 }
3309         } else {
3310                 bond_for_each_slave_rcu(bond, slave, iter) {
3311                         dev_uc_sync_multiple(slave->dev, bond_dev);
3312                         dev_mc_sync_multiple(slave->dev, bond_dev);
3313                 }
3314         }
3315         rcu_read_unlock();
3316 }
3317
3318 static int bond_neigh_init(struct neighbour *n)
3319 {
3320         struct bonding *bond = netdev_priv(n->dev);
3321         const struct net_device_ops *slave_ops;
3322         struct neigh_parms parms;
3323         struct slave *slave;
3324         int ret;
3325
3326         slave = bond_first_slave(bond);
3327         if (!slave)
3328                 return 0;
3329         slave_ops = slave->dev->netdev_ops;
3330         if (!slave_ops->ndo_neigh_setup)
3331                 return 0;
3332
3333         parms.neigh_setup = NULL;
3334         parms.neigh_cleanup = NULL;
3335         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3336         if (ret)
3337                 return ret;
3338
3339         /*
3340          * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3341          * after the last slave has been detached.  Assumes that all slaves
3342          * utilize the same neigh_cleanup (true at this writing as only user
3343          * is ipoib).
3344          */
3345         n->parms->neigh_cleanup = parms.neigh_cleanup;
3346
3347         if (!parms.neigh_setup)
3348                 return 0;
3349
3350         return parms.neigh_setup(n);
3351 }
3352
3353 /*
3354  * The bonding ndo_neigh_setup is called at init time beofre any
3355  * slave exists. So we must declare proxy setup function which will
3356  * be used at run time to resolve the actual slave neigh param setup.
3357  *
3358  * It's also called by master devices (such as vlans) to setup their
3359  * underlying devices. In that case - do nothing, we're already set up from
3360  * our init.
3361  */
3362 static int bond_neigh_setup(struct net_device *dev,
3363                             struct neigh_parms *parms)
3364 {
3365         /* modify only our neigh_parms */
3366         if (parms->dev == dev)
3367                 parms->neigh_setup = bond_neigh_init;
3368
3369         return 0;
3370 }
3371
3372 /*
3373  * Change the MTU of all of a master's slaves to match the master
3374  */
3375 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3376 {
3377         struct bonding *bond = netdev_priv(bond_dev);
3378         struct slave *slave, *rollback_slave;
3379         struct list_head *iter;
3380         int res = 0;
3381
3382         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3383                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3384
3385         /* Can't hold bond->lock with bh disabled here since
3386          * some base drivers panic. On the other hand we can't
3387          * hold bond->lock without bh disabled because we'll
3388          * deadlock. The only solution is to rely on the fact
3389          * that we're under rtnl_lock here, and the slaves
3390          * list won't change. This doesn't solve the problem
3391          * of setting the slave's MTU while it is
3392          * transmitting, but the assumption is that the base
3393          * driver can handle that.
3394          *
3395          * TODO: figure out a way to safely iterate the slaves
3396          * list, but without holding a lock around the actual
3397          * call to the base driver.
3398          */
3399
3400         bond_for_each_slave(bond, slave, iter) {
3401                 pr_debug("s %p c_m %p\n",
3402                          slave,
3403                          slave->dev->netdev_ops->ndo_change_mtu);
3404
3405                 res = dev_set_mtu(slave->dev, new_mtu);
3406
3407                 if (res) {
3408                         /* If we failed to set the slave's mtu to the new value
3409                          * we must abort the operation even in ACTIVE_BACKUP
3410                          * mode, because if we allow the backup slaves to have
3411                          * different mtu values than the active slave we'll
3412                          * need to change their mtu when doing a failover. That
3413                          * means changing their mtu from timer context, which
3414                          * is probably not a good idea.
3415                          */
3416                         pr_debug("err %d %s\n", res, slave->dev->name);
3417                         goto unwind;
3418                 }
3419         }
3420
3421         bond_dev->mtu = new_mtu;
3422
3423         return 0;
3424
3425 unwind:
3426         /* unwind from head to the slave that failed */
3427         bond_for_each_slave(bond, rollback_slave, iter) {
3428                 int tmp_res;
3429
3430                 if (rollback_slave == slave)
3431                         break;
3432
3433                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3434                 if (tmp_res) {
3435                         pr_debug("unwind err %d dev %s\n",
3436                                  tmp_res, rollback_slave->dev->name);
3437                 }
3438         }
3439
3440         return res;
3441 }
3442
3443 /*
3444  * Change HW address
3445  *
3446  * Note that many devices must be down to change the HW address, and
3447  * downing the master releases all slaves.  We can make bonds full of
3448  * bonding devices to test this, however.
3449  */
3450 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3451 {
3452         struct bonding *bond = netdev_priv(bond_dev);
3453         struct slave *slave, *rollback_slave;
3454         struct sockaddr *sa = addr, tmp_sa;
3455         struct list_head *iter;
3456         int res = 0;
3457
3458         if (bond->params.mode == BOND_MODE_ALB)
3459                 return bond_alb_set_mac_address(bond_dev, addr);
3460
3461
3462         pr_debug("bond=%p, name=%s\n",
3463                  bond, bond_dev ? bond_dev->name : "None");
3464
3465         /* If fail_over_mac is enabled, do nothing and return success.
3466          * Returning an error causes ifenslave to fail.
3467          */
3468         if (bond->params.fail_over_mac)
3469                 return 0;
3470
3471         if (!is_valid_ether_addr(sa->sa_data))
3472                 return -EADDRNOTAVAIL;
3473
3474         /* Can't hold bond->lock with bh disabled here since
3475          * some base drivers panic. On the other hand we can't
3476          * hold bond->lock without bh disabled because we'll
3477          * deadlock. The only solution is to rely on the fact
3478          * that we're under rtnl_lock here, and the slaves
3479          * list won't change. This doesn't solve the problem
3480          * of setting the slave's hw address while it is
3481          * transmitting, but the assumption is that the base
3482          * driver can handle that.
3483          *
3484          * TODO: figure out a way to safely iterate the slaves
3485          * list, but without holding a lock around the actual
3486          * call to the base driver.
3487          */
3488
3489         bond_for_each_slave(bond, slave, iter) {
3490                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3491                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3492
3493                 if (slave_ops->ndo_set_mac_address == NULL) {
3494                         res = -EOPNOTSUPP;
3495                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3496                         goto unwind;
3497                 }
3498
3499                 res = dev_set_mac_address(slave->dev, addr);
3500                 if (res) {
3501                         /* TODO: consider downing the slave
3502                          * and retry ?
3503                          * User should expect communications
3504                          * breakage anyway until ARP finish
3505                          * updating, so...
3506                          */
3507                         pr_debug("err %d %s\n", res, slave->dev->name);
3508                         goto unwind;
3509                 }
3510         }
3511
3512         /* success */
3513         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3514         return 0;
3515
3516 unwind:
3517         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3518         tmp_sa.sa_family = bond_dev->type;
3519
3520         /* unwind from head to the slave that failed */
3521         bond_for_each_slave(bond, rollback_slave, iter) {
3522                 int tmp_res;
3523
3524                 if (rollback_slave == slave)
3525                         break;
3526
3527                 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3528                 if (tmp_res) {
3529                         pr_debug("unwind err %d dev %s\n",
3530                                  tmp_res, rollback_slave->dev->name);
3531                 }
3532         }
3533
3534         return res;
3535 }
3536
3537 /**
3538  * bond_xmit_slave_id - transmit skb through slave with slave_id
3539  * @bond: bonding device that is transmitting
3540  * @skb: buffer to transmit
3541  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3542  *
3543  * This function tries to transmit through slave with slave_id but in case
3544  * it fails, it tries to find the first available slave for transmission.
3545  * The skb is consumed in all cases, thus the function is void.
3546  */
3547 void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3548 {
3549         struct list_head *iter;
3550         struct slave *slave;
3551         int i = slave_id;
3552
3553         /* Here we start from the slave with slave_id */
3554         bond_for_each_slave_rcu(bond, slave, iter) {
3555                 if (--i < 0) {
3556                         if (slave_can_tx(slave)) {
3557                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3558                                 return;
3559                         }
3560                 }
3561         }
3562
3563         /* Here we start from the first slave up to slave_id */
3564         i = slave_id;
3565         bond_for_each_slave_rcu(bond, slave, iter) {
3566                 if (--i < 0)
3567                         break;
3568                 if (slave_can_tx(slave)) {
3569                         bond_dev_queue_xmit(bond, skb, slave->dev);
3570                         return;
3571                 }
3572         }
3573         /* no slave that can tx has been found */
3574         kfree_skb(skb);
3575 }
3576
3577 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3578 {
3579         struct bonding *bond = netdev_priv(bond_dev);
3580         struct iphdr *iph = ip_hdr(skb);
3581         struct slave *slave;
3582
3583         /*
3584          * Start with the curr_active_slave that joined the bond as the
3585          * default for sending IGMP traffic.  For failover purposes one
3586          * needs to maintain some consistency for the interface that will
3587          * send the join/membership reports.  The curr_active_slave found
3588          * will send all of this type of traffic.
3589          */
3590         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3591                 slave = rcu_dereference(bond->curr_active_slave);
3592                 if (slave && slave_can_tx(slave))
3593                         bond_dev_queue_xmit(bond, skb, slave->dev);
3594                 else
3595                         bond_xmit_slave_id(bond, skb, 0);
3596         } else {
3597                 bond_xmit_slave_id(bond, skb,
3598                                    bond->rr_tx_counter++ % bond->slave_cnt);
3599         }
3600
3601         return NETDEV_TX_OK;
3602 }
3603
3604 /*
3605  * in active-backup mode, we know that bond->curr_active_slave is always valid if
3606  * the bond has a usable interface.
3607  */
3608 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3609 {
3610         struct bonding *bond = netdev_priv(bond_dev);
3611         struct slave *slave;
3612
3613         slave = rcu_dereference(bond->curr_active_slave);
3614         if (slave)
3615                 bond_dev_queue_xmit(bond, skb, slave->dev);
3616         else
3617                 kfree_skb(skb);
3618
3619         return NETDEV_TX_OK;
3620 }
3621
3622 /* In bond_xmit_xor() , we determine the output device by using a pre-
3623  * determined xmit_hash_policy(), If the selected device is not enabled,
3624  * find the next active slave.
3625  */
3626 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3627 {
3628         struct bonding *bond = netdev_priv(bond_dev);
3629
3630         bond_xmit_slave_id(bond, skb, bond_xmit_hash(bond, skb, bond->slave_cnt));
3631
3632         return NETDEV_TX_OK;
3633 }
3634
3635 /* in broadcast mode, we send everything to all usable interfaces. */
3636 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3637 {
3638         struct bonding *bond = netdev_priv(bond_dev);
3639         struct slave *slave = NULL;
3640         struct list_head *iter;
3641
3642         bond_for_each_slave_rcu(bond, slave, iter) {
3643                 if (bond_is_last_slave(bond, slave))
3644                         break;
3645                 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3646                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3647
3648                         if (!skb2) {
3649                                 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3650                                        bond_dev->name);
3651                                 continue;
3652                         }
3653                         /* bond_dev_queue_xmit always returns 0 */
3654                         bond_dev_queue_xmit(bond, skb2, slave->dev);
3655                 }
3656         }
3657         if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3658                 bond_dev_queue_xmit(bond, skb, slave->dev);
3659         else
3660                 kfree_skb(skb);
3661
3662         return NETDEV_TX_OK;
3663 }
3664
3665 /*------------------------- Device initialization ---------------------------*/
3666
3667 /*
3668  * Lookup the slave that corresponds to a qid
3669  */
3670 static inline int bond_slave_override(struct bonding *bond,
3671                                       struct sk_buff *skb)
3672 {
3673         struct slave *slave = NULL;
3674         struct slave *check_slave;
3675         struct list_head *iter;
3676         int res = 1;
3677
3678         if (!skb->queue_mapping)
3679                 return 1;
3680
3681         /* Find out if any slaves have the same mapping as this skb. */
3682         bond_for_each_slave_rcu(bond, check_slave, iter) {
3683                 if (check_slave->queue_id == skb->queue_mapping) {
3684                         slave = check_slave;
3685                         break;
3686                 }
3687         }
3688
3689         /* If the slave isn't UP, use default transmit policy. */
3690         if (slave && slave->queue_id && IS_UP(slave->dev) &&
3691             (slave->link == BOND_LINK_UP)) {
3692                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3693         }
3694
3695         return res;
3696 }
3697
3698
3699 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
3700 {
3701         /*
3702          * This helper function exists to help dev_pick_tx get the correct
3703          * destination queue.  Using a helper function skips a call to
3704          * skb_tx_hash and will put the skbs in the queue we expect on their
3705          * way down to the bonding driver.
3706          */
3707         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3708
3709         /*
3710          * Save the original txq to restore before passing to the driver
3711          */
3712         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3713
3714         if (unlikely(txq >= dev->real_num_tx_queues)) {
3715                 do {
3716                         txq -= dev->real_num_tx_queues;
3717                 } while (txq >= dev->real_num_tx_queues);
3718         }
3719         return txq;
3720 }
3721
3722 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3723 {
3724         struct bonding *bond = netdev_priv(dev);
3725
3726         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
3727                 if (!bond_slave_override(bond, skb))
3728                         return NETDEV_TX_OK;
3729         }
3730
3731         switch (bond->params.mode) {
3732         case BOND_MODE_ROUNDROBIN:
3733                 return bond_xmit_roundrobin(skb, dev);
3734         case BOND_MODE_ACTIVEBACKUP:
3735                 return bond_xmit_activebackup(skb, dev);
3736         case BOND_MODE_XOR:
3737                 return bond_xmit_xor(skb, dev);
3738         case BOND_MODE_BROADCAST:
3739                 return bond_xmit_broadcast(skb, dev);
3740         case BOND_MODE_8023AD:
3741                 return bond_3ad_xmit_xor(skb, dev);
3742         case BOND_MODE_ALB:
3743         case BOND_MODE_TLB:
3744                 return bond_alb_xmit(skb, dev);
3745         default:
3746                 /* Should never happen, mode already checked */
3747                 pr_err("%s: Error: Unknown bonding mode %d\n",
3748                        dev->name, bond->params.mode);
3749                 WARN_ON_ONCE(1);
3750                 kfree_skb(skb);
3751                 return NETDEV_TX_OK;
3752         }
3753 }
3754
3755 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3756 {
3757         struct bonding *bond = netdev_priv(dev);
3758         netdev_tx_t ret = NETDEV_TX_OK;
3759
3760         /*
3761          * If we risk deadlock from transmitting this in the
3762          * netpoll path, tell netpoll to queue the frame for later tx
3763          */
3764         if (is_netpoll_tx_blocked(dev))
3765                 return NETDEV_TX_BUSY;
3766
3767         rcu_read_lock();
3768         if (bond_has_slaves(bond))
3769                 ret = __bond_start_xmit(skb, dev);
3770         else
3771                 kfree_skb(skb);
3772         rcu_read_unlock();
3773
3774         return ret;
3775 }
3776
3777 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3778                                      struct ethtool_cmd *ecmd)
3779 {
3780         struct bonding *bond = netdev_priv(bond_dev);
3781         unsigned long speed = 0;
3782         struct list_head *iter;
3783         struct slave *slave;
3784
3785         ecmd->duplex = DUPLEX_UNKNOWN;
3786         ecmd->port = PORT_OTHER;
3787
3788         /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3789          * do not need to check mode.  Though link speed might not represent
3790          * the true receive or transmit bandwidth (not all modes are symmetric)
3791          * this is an accurate maximum.
3792          */
3793         read_lock(&bond->lock);
3794         bond_for_each_slave(bond, slave, iter) {
3795                 if (SLAVE_IS_OK(slave)) {
3796                         if (slave->speed != SPEED_UNKNOWN)
3797                                 speed += slave->speed;
3798                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
3799                             slave->duplex != DUPLEX_UNKNOWN)
3800                                 ecmd->duplex = slave->duplex;
3801                 }
3802         }
3803         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3804         read_unlock(&bond->lock);
3805
3806         return 0;
3807 }
3808
3809 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3810                                      struct ethtool_drvinfo *drvinfo)
3811 {
3812         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3813         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3814         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3815                  BOND_ABI_VERSION);
3816 }
3817
3818 static const struct ethtool_ops bond_ethtool_ops = {
3819         .get_drvinfo            = bond_ethtool_get_drvinfo,
3820         .get_settings           = bond_ethtool_get_settings,
3821         .get_link               = ethtool_op_get_link,
3822 };
3823
3824 static const struct net_device_ops bond_netdev_ops = {
3825         .ndo_init               = bond_init,
3826         .ndo_uninit             = bond_uninit,
3827         .ndo_open               = bond_open,
3828         .ndo_stop               = bond_close,
3829         .ndo_start_xmit         = bond_start_xmit,
3830         .ndo_select_queue       = bond_select_queue,
3831         .ndo_get_stats64        = bond_get_stats,
3832         .ndo_do_ioctl           = bond_do_ioctl,
3833         .ndo_change_rx_flags    = bond_change_rx_flags,
3834         .ndo_set_rx_mode        = bond_set_rx_mode,
3835         .ndo_change_mtu         = bond_change_mtu,
3836         .ndo_set_mac_address    = bond_set_mac_address,
3837         .ndo_neigh_setup        = bond_neigh_setup,
3838         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
3839         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
3840 #ifdef CONFIG_NET_POLL_CONTROLLER
3841         .ndo_netpoll_setup      = bond_netpoll_setup,
3842         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
3843         .ndo_poll_controller    = bond_poll_controller,
3844 #endif
3845         .ndo_add_slave          = bond_enslave,
3846         .ndo_del_slave          = bond_release,
3847         .ndo_fix_features       = bond_fix_features,
3848 };
3849
3850 static const struct device_type bond_type = {
3851         .name = "bond",
3852 };
3853
3854 static void bond_destructor(struct net_device *bond_dev)
3855 {
3856         struct bonding *bond = netdev_priv(bond_dev);
3857         if (bond->wq)
3858                 destroy_workqueue(bond->wq);
3859         free_netdev(bond_dev);
3860 }
3861
3862 void bond_setup(struct net_device *bond_dev)
3863 {
3864         struct bonding *bond = netdev_priv(bond_dev);
3865
3866         /* initialize rwlocks */
3867         rwlock_init(&bond->lock);
3868         rwlock_init(&bond->curr_slave_lock);
3869         bond->params = bonding_defaults;
3870
3871         /* Initialize pointers */
3872         bond->dev = bond_dev;
3873
3874         /* Initialize the device entry points */
3875         ether_setup(bond_dev);
3876         bond_dev->netdev_ops = &bond_netdev_ops;
3877         bond_dev->ethtool_ops = &bond_ethtool_ops;
3878
3879         bond_dev->destructor = bond_destructor;
3880
3881         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
3882
3883         /* Initialize the device options */
3884         bond_dev->tx_queue_len = 0;
3885         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
3886         bond_dev->priv_flags |= IFF_BONDING;
3887         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
3888
3889         /* At first, we block adding VLANs. That's the only way to
3890          * prevent problems that occur when adding VLANs over an
3891          * empty bond. The block will be removed once non-challenged
3892          * slaves are enslaved.
3893          */
3894         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
3895
3896         /* don't acquire bond device's netif_tx_lock when
3897          * transmitting */
3898         bond_dev->features |= NETIF_F_LLTX;
3899
3900         /* By default, we declare the bond to be fully
3901          * VLAN hardware accelerated capable. Special
3902          * care is taken in the various xmit functions
3903          * when there are slaves that are not hw accel
3904          * capable
3905          */
3906
3907         bond_dev->hw_features = BOND_VLAN_FEATURES |
3908                                 NETIF_F_HW_VLAN_CTAG_TX |
3909                                 NETIF_F_HW_VLAN_CTAG_RX |
3910                                 NETIF_F_HW_VLAN_CTAG_FILTER;
3911
3912         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
3913         bond_dev->features |= bond_dev->hw_features;
3914 }
3915
3916 /*
3917 * Destroy a bonding device.
3918 * Must be under rtnl_lock when this function is called.
3919 */
3920 static void bond_uninit(struct net_device *bond_dev)
3921 {
3922         struct bonding *bond = netdev_priv(bond_dev);
3923         struct list_head *iter;
3924         struct slave *slave;
3925
3926         bond_netpoll_cleanup(bond_dev);
3927
3928         /* Release the bonded slaves */
3929         bond_for_each_slave(bond, slave, iter)
3930                 __bond_release_one(bond_dev, slave->dev, true);
3931         pr_info("%s: released all slaves\n", bond_dev->name);
3932
3933         list_del(&bond->bond_list);
3934
3935         bond_debug_unregister(bond);
3936 }
3937
3938 /*------------------------- Module initialization ---------------------------*/
3939
3940 /*
3941  * Convert string input module parms.  Accept either the
3942  * number of the mode or its string name.  A bit complicated because
3943  * some mode names are substrings of other names, and calls from sysfs
3944  * may have whitespace in the name (trailing newlines, for example).
3945  */
3946 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
3947 {
3948         int modeint = -1, i, rv;
3949         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
3950
3951         for (p = (char *)buf; *p; p++)
3952                 if (!(isdigit(*p) || isspace(*p)))
3953                         break;
3954
3955         if (*p)
3956                 rv = sscanf(buf, "%20s", modestr);
3957         else
3958                 rv = sscanf(buf, "%d", &modeint);
3959
3960         if (!rv)
3961                 return -1;
3962
3963         for (i = 0; tbl[i].modename; i++) {
3964                 if (modeint == tbl[i].mode)
3965                         return tbl[i].mode;
3966                 if (strcmp(modestr, tbl[i].modename) == 0)
3967                         return tbl[i].mode;
3968         }
3969
3970         return -1;
3971 }
3972
3973 static int bond_check_params(struct bond_params *params)
3974 {
3975         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
3976         int arp_all_targets_value;
3977
3978         /*
3979          * Convert string parameters.
3980          */
3981         if (mode) {
3982                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
3983                 if (bond_mode == -1) {
3984                         pr_err("Error: Invalid bonding mode \"%s\"\n",
3985                                mode == NULL ? "NULL" : mode);
3986                         return -EINVAL;
3987                 }
3988         }
3989
3990         if (xmit_hash_policy) {
3991                 if ((bond_mode != BOND_MODE_XOR) &&
3992                     (bond_mode != BOND_MODE_8023AD)) {
3993                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
3994                                bond_mode_name(bond_mode));
3995                 } else {
3996                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
3997                                                         xmit_hashtype_tbl);
3998                         if (xmit_hashtype == -1) {
3999                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4000                                        xmit_hash_policy == NULL ? "NULL" :
4001                                        xmit_hash_policy);
4002                                 return -EINVAL;
4003                         }
4004                 }
4005         }
4006
4007         if (lacp_rate) {
4008                 if (bond_mode != BOND_MODE_8023AD) {
4009                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4010                                 bond_mode_name(bond_mode));
4011                 } else {
4012                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4013                         if (lacp_fast == -1) {
4014                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4015                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4016                                 return -EINVAL;
4017                         }
4018                 }
4019         }
4020
4021         if (ad_select) {
4022                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4023                 if (params->ad_select == -1) {
4024                         pr_err("Error: Invalid ad_select \"%s\"\n",
4025                                ad_select == NULL ? "NULL" : ad_select);
4026                         return -EINVAL;
4027                 }
4028
4029                 if (bond_mode != BOND_MODE_8023AD) {
4030                         pr_warning("ad_select param only affects 802.3ad mode\n");
4031                 }
4032         } else {
4033                 params->ad_select = BOND_AD_STABLE;
4034         }
4035
4036         if (max_bonds < 0) {
4037                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4038                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4039                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4040         }
4041
4042         if (miimon < 0) {
4043                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4044                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4045                 miimon = BOND_LINK_MON_INTERV;
4046         }
4047
4048         if (updelay < 0) {
4049                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4050                            updelay, INT_MAX);
4051                 updelay = 0;
4052         }
4053
4054         if (downdelay < 0) {
4055                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4056                            downdelay, INT_MAX);
4057                 downdelay = 0;
4058         }
4059
4060         if ((use_carrier != 0) && (use_carrier != 1)) {
4061                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4062                            use_carrier);
4063                 use_carrier = 1;
4064         }
4065
4066         if (num_peer_notif < 0 || num_peer_notif > 255) {
4067                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4068                            num_peer_notif);
4069                 num_peer_notif = 1;
4070         }
4071
4072         /* reset values for 802.3ad */
4073         if (bond_mode == BOND_MODE_8023AD) {
4074                 if (!miimon) {
4075                         pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4076                         pr_warning("Forcing miimon to 100msec\n");
4077                         miimon = 100;
4078                 }
4079         }
4080
4081         if (tx_queues < 1 || tx_queues > 255) {
4082                 pr_warning("Warning: tx_queues (%d) should be between "
4083                            "1 and 255, resetting to %d\n",
4084                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4085                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4086         }
4087
4088         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4089                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4090                            "not of valid value (0/1), so it was set to "
4091                            "0\n", all_slaves_active);
4092                 all_slaves_active = 0;
4093         }
4094
4095         if (resend_igmp < 0 || resend_igmp > 255) {
4096                 pr_warning("Warning: resend_igmp (%d) should be between "
4097                            "0 and 255, resetting to %d\n",
4098                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4099                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4100         }
4101
4102         /* reset values for TLB/ALB */
4103         if ((bond_mode == BOND_MODE_TLB) ||
4104             (bond_mode == BOND_MODE_ALB)) {
4105                 if (!miimon) {
4106                         pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure and link speed which are essential for TLB/ALB load balancing\n");
4107                         pr_warning("Forcing miimon to 100msec\n");
4108                         miimon = 100;
4109                 }
4110         }
4111
4112         if (bond_mode == BOND_MODE_ALB) {
4113                 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",
4114                           updelay);
4115         }
4116
4117         if (!miimon) {
4118                 if (updelay || downdelay) {
4119                         /* just warn the user the up/down delay will have
4120                          * no effect since miimon is zero...
4121                          */
4122                         pr_warning("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",
4123                                    updelay, downdelay);
4124                 }
4125         } else {
4126                 /* don't allow arp monitoring */
4127                 if (arp_interval) {
4128                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4129                                    miimon, arp_interval);
4130                         arp_interval = 0;
4131                 }
4132
4133                 if ((updelay % miimon) != 0) {
4134                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4135                                    updelay, miimon,
4136                                    (updelay / miimon) * miimon);
4137                 }
4138
4139                 updelay /= miimon;
4140
4141                 if ((downdelay % miimon) != 0) {
4142                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4143                                    downdelay, miimon,
4144                                    (downdelay / miimon) * miimon);
4145                 }
4146
4147                 downdelay /= miimon;
4148         }
4149
4150         if (arp_interval < 0) {
4151                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4152                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4153                 arp_interval = BOND_LINK_ARP_INTERV;
4154         }
4155
4156         for (arp_ip_count = 0, i = 0;
4157              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4158                 /* not complete check, but should be good enough to
4159                    catch mistakes */
4160                 __be32 ip = in_aton(arp_ip_target[i]);
4161                 if (!isdigit(arp_ip_target[i][0]) || ip == 0 ||
4162                     ip == htonl(INADDR_BROADCAST)) {
4163                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4164                                    arp_ip_target[i]);
4165                         arp_interval = 0;
4166                 } else {
4167                         if (bond_get_targets_ip(arp_target, ip) == -1)
4168                                 arp_target[arp_ip_count++] = ip;
4169                         else
4170                                 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4171                                            &ip);
4172                 }
4173         }
4174
4175         if (arp_interval && !arp_ip_count) {
4176                 /* don't allow arping if no arp_ip_target given... */
4177                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4178                            arp_interval);
4179                 arp_interval = 0;
4180         }
4181
4182         if (arp_validate) {
4183                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4184                         pr_err("arp_validate only supported in active-backup mode\n");
4185                         return -EINVAL;
4186                 }
4187                 if (!arp_interval) {
4188                         pr_err("arp_validate requires arp_interval\n");
4189                         return -EINVAL;
4190                 }
4191
4192                 arp_validate_value = bond_parse_parm(arp_validate,
4193                                                      arp_validate_tbl);
4194                 if (arp_validate_value == -1) {
4195                         pr_err("Error: invalid arp_validate \"%s\"\n",
4196                                arp_validate == NULL ? "NULL" : arp_validate);
4197                         return -EINVAL;
4198                 }
4199         } else
4200                 arp_validate_value = 0;
4201
4202         arp_all_targets_value = 0;
4203         if (arp_all_targets) {
4204                 arp_all_targets_value = bond_parse_parm(arp_all_targets,
4205                                                         arp_all_targets_tbl);
4206
4207                 if (arp_all_targets_value == -1) {
4208                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4209                                arp_all_targets);
4210                         arp_all_targets_value = 0;
4211                 }
4212         }
4213
4214         if (miimon) {
4215                 pr_info("MII link monitoring set to %d ms\n", miimon);
4216         } else if (arp_interval) {
4217                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4218                         arp_interval,
4219                         arp_validate_tbl[arp_validate_value].modename,
4220                         arp_ip_count);
4221
4222                 for (i = 0; i < arp_ip_count; i++)
4223                         pr_info(" %s", arp_ip_target[i]);
4224
4225                 pr_info("\n");
4226
4227         } else if (max_bonds) {
4228                 /* miimon and arp_interval not set, we need one so things
4229                  * work as expected, see bonding.txt for details
4230                  */
4231                 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");
4232         }
4233
4234         if (primary && !USES_PRIMARY(bond_mode)) {
4235                 /* currently, using a primary only makes sense
4236                  * in active backup, TLB or ALB modes
4237                  */
4238                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4239                            primary, bond_mode_name(bond_mode));
4240                 primary = NULL;
4241         }
4242
4243         if (primary && primary_reselect) {
4244                 primary_reselect_value = bond_parse_parm(primary_reselect,
4245                                                          pri_reselect_tbl);
4246                 if (primary_reselect_value == -1) {
4247                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4248                                primary_reselect ==
4249                                         NULL ? "NULL" : primary_reselect);
4250                         return -EINVAL;
4251                 }
4252         } else {
4253                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4254         }
4255
4256         if (fail_over_mac) {
4257                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4258                                                       fail_over_mac_tbl);
4259                 if (fail_over_mac_value == -1) {
4260                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4261                                arp_validate == NULL ? "NULL" : arp_validate);
4262                         return -EINVAL;
4263                 }
4264
4265                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4266                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4267         } else {
4268                 fail_over_mac_value = BOND_FOM_NONE;
4269         }
4270
4271         /* fill params struct with the proper values */
4272         params->mode = bond_mode;
4273         params->xmit_policy = xmit_hashtype;
4274         params->miimon = miimon;
4275         params->num_peer_notif = num_peer_notif;
4276         params->arp_interval = arp_interval;
4277         params->arp_validate = arp_validate_value;
4278         params->arp_all_targets = arp_all_targets_value;
4279         params->updelay = updelay;
4280         params->downdelay = downdelay;
4281         params->use_carrier = use_carrier;
4282         params->lacp_fast = lacp_fast;
4283         params->primary[0] = 0;
4284         params->primary_reselect = primary_reselect_value;
4285         params->fail_over_mac = fail_over_mac_value;
4286         params->tx_queues = tx_queues;
4287         params->all_slaves_active = all_slaves_active;
4288         params->resend_igmp = resend_igmp;
4289         params->min_links = min_links;
4290         params->lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4291
4292         if (primary) {
4293                 strncpy(params->primary, primary, IFNAMSIZ);
4294                 params->primary[IFNAMSIZ - 1] = 0;
4295         }
4296
4297         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4298
4299         return 0;
4300 }
4301
4302 static struct lock_class_key bonding_netdev_xmit_lock_key;
4303 static struct lock_class_key bonding_netdev_addr_lock_key;
4304 static struct lock_class_key bonding_tx_busylock_key;
4305
4306 static void bond_set_lockdep_class_one(struct net_device *dev,
4307                                        struct netdev_queue *txq,
4308                                        void *_unused)
4309 {
4310         lockdep_set_class(&txq->_xmit_lock,
4311                           &bonding_netdev_xmit_lock_key);
4312 }
4313
4314 static void bond_set_lockdep_class(struct net_device *dev)
4315 {
4316         lockdep_set_class(&dev->addr_list_lock,
4317                           &bonding_netdev_addr_lock_key);
4318         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4319         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4320 }
4321
4322 /*
4323  * Called from registration process
4324  */
4325 static int bond_init(struct net_device *bond_dev)
4326 {
4327         struct bonding *bond = netdev_priv(bond_dev);
4328         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4329         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4330
4331         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4332
4333         /*
4334          * Initialize locks that may be required during
4335          * en/deslave operations.  All of the bond_open work
4336          * (of which this is part) should really be moved to
4337          * a phase prior to dev_open
4338          */
4339         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4340         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4341
4342         bond->wq = create_singlethread_workqueue(bond_dev->name);
4343         if (!bond->wq)
4344                 return -ENOMEM;
4345
4346         bond_set_lockdep_class(bond_dev);
4347
4348         list_add_tail(&bond->bond_list, &bn->dev_list);
4349
4350         bond_prepare_sysfs_group(bond);
4351
4352         bond_debug_register(bond);
4353
4354         /* Ensure valid dev_addr */
4355         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4356             bond_dev->addr_assign_type == NET_ADDR_PERM)
4357                 eth_hw_addr_random(bond_dev);
4358
4359         return 0;
4360 }
4361
4362 unsigned int bond_get_num_tx_queues(void)
4363 {
4364         return tx_queues;
4365 }
4366
4367 /* Create a new bond based on the specified name and bonding parameters.
4368  * If name is NULL, obtain a suitable "bond%d" name for us.
4369  * Caller must NOT hold rtnl_lock; we need to release it here before we
4370  * set up our sysfs entries.
4371  */
4372 int bond_create(struct net *net, const char *name)
4373 {
4374         struct net_device *bond_dev;
4375         int res;
4376
4377         rtnl_lock();
4378
4379         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4380                                    name ? name : "bond%d",
4381                                    bond_setup, tx_queues);
4382         if (!bond_dev) {
4383                 pr_err("%s: eek! can't alloc netdev!\n", name);
4384                 rtnl_unlock();
4385                 return -ENOMEM;
4386         }
4387
4388         dev_net_set(bond_dev, net);
4389         bond_dev->rtnl_link_ops = &bond_link_ops;
4390
4391         res = register_netdevice(bond_dev);
4392
4393         netif_carrier_off(bond_dev);
4394
4395         rtnl_unlock();
4396         if (res < 0)
4397                 bond_destructor(bond_dev);
4398         return res;
4399 }
4400
4401 static int __net_init bond_net_init(struct net *net)
4402 {
4403         struct bond_net *bn = net_generic(net, bond_net_id);
4404
4405         bn->net = net;
4406         INIT_LIST_HEAD(&bn->dev_list);
4407
4408         bond_create_proc_dir(bn);
4409         bond_create_sysfs(bn);
4410
4411         return 0;
4412 }
4413
4414 static void __net_exit bond_net_exit(struct net *net)
4415 {
4416         struct bond_net *bn = net_generic(net, bond_net_id);
4417         struct bonding *bond, *tmp_bond;
4418         LIST_HEAD(list);
4419
4420         bond_destroy_sysfs(bn);
4421         bond_destroy_proc_dir(bn);
4422
4423         /* Kill off any bonds created after unregistering bond rtnl ops */
4424         rtnl_lock();
4425         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4426                 unregister_netdevice_queue(bond->dev, &list);
4427         unregister_netdevice_many(&list);
4428         rtnl_unlock();
4429 }
4430
4431 static struct pernet_operations bond_net_ops = {
4432         .init = bond_net_init,
4433         .exit = bond_net_exit,
4434         .id   = &bond_net_id,
4435         .size = sizeof(struct bond_net),
4436 };
4437
4438 static int __init bonding_init(void)
4439 {
4440         int i;
4441         int res;
4442
4443         pr_info("%s", bond_version);
4444
4445         res = bond_check_params(&bonding_defaults);
4446         if (res)
4447                 goto out;
4448
4449         res = register_pernet_subsys(&bond_net_ops);
4450         if (res)
4451                 goto out;
4452
4453         res = bond_netlink_init();
4454         if (res)
4455                 goto err_link;
4456
4457         bond_create_debugfs();
4458
4459         for (i = 0; i < max_bonds; i++) {
4460                 res = bond_create(&init_net, NULL);
4461                 if (res)
4462                         goto err;
4463         }
4464
4465         register_netdevice_notifier(&bond_netdev_notifier);
4466 out:
4467         return res;
4468 err:
4469         bond_netlink_fini();
4470 err_link:
4471         unregister_pernet_subsys(&bond_net_ops);
4472         goto out;
4473
4474 }
4475
4476 static void __exit bonding_exit(void)
4477 {
4478         unregister_netdevice_notifier(&bond_netdev_notifier);
4479
4480         bond_destroy_debugfs();
4481
4482         bond_netlink_fini();
4483         unregister_pernet_subsys(&bond_net_ops);
4484
4485 #ifdef CONFIG_NET_POLL_CONTROLLER
4486         /*
4487          * Make sure we don't have an imbalance on our netpoll blocking
4488          */
4489         WARN_ON(atomic_read(&netpoll_block_tx));
4490 #endif
4491 }
4492
4493 module_init(bonding_init);
4494 module_exit(bonding_exit);
4495 MODULE_LICENSE("GPL");
4496 MODULE_VERSION(DRV_VERSION);
4497 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4498 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");