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 "bonding.h"
  82 #include "bond_3ad.h"
  83 #include "bond_alb.h"
  84
  85 /*---------------------------- Module parameters ----------------------------*/
  86
  87 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  88 #define BOND_LINK_MON_INTERV    0
  89 #define BOND_LINK_ARP_INTERV    0
  90
  91 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  92 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  93 static int num_peer_notif = 1;
  94 static int miimon       = BOND_LINK_MON_INTERV;
  95 static int updelay;
  96 static int downdelay;
  97 static int use_carrier  = 1;
  98 static char *mode;
  99 static char *primary;
 100 static char *primary_reselect;
 101 static char *lacp_rate;
 102 static int min_links;
 103 static char *ad_select;
 104 static char *xmit_hash_policy;
 105 static int arp_interval = BOND_LINK_ARP_INTERV;
 106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 107 static char *arp_validate;
 108 static char *arp_all_targets;
 109 static char *fail_over_mac;
 110 static int all_slaves_active;
 111 static struct bond_params bonding_defaults;
 112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 113
 114 module_param(max_bonds, int, 0);
 115 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 116 module_param(tx_queues, int, 0);
 117 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 118 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 119 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 120                                "failover event (alias of num_unsol_na)");
 121 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 122 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 123                                "failover event (alias of num_grat_arp)");
 124 module_param(miimon, int, 0);
 125 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 126 module_param(updelay, int, 0);
 127 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 128 module_param(downdelay, int, 0);
 129 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 130                             "in milliseconds");
 131 module_param(use_carrier, int, 0);
 132 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 133                               "0 for off, 1 for on (default)");
 134 module_param(mode, charp, 0);
 135 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 136                        "1 for active-backup, 2 for balance-xor, "
 137                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 138                        "6 for balance-alb");
 139 module_param(primary, charp, 0);
 140 MODULE_PARM_DESC(primary, "Primary network device to use");
 141 module_param(primary_reselect, charp, 0);
 142 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 143                                    "once it comes up; "
 144                                    "0 for always (default), "
 145                                    "1 for only if speed of primary is "
 146                                    "better, "
 147                                    "2 for only on active slave "
 148                                    "failure");
 149 module_param(lacp_rate, charp, 0);
 150 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 151                             "0 for slow, 1 for fast");
 152 module_param(ad_select, charp, 0);
 153 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 154                             "0 for stable (default), 1 for bandwidth, "
 155                             "2 for count");
 156 module_param(min_links, int, 0);
 157 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 158
 159 module_param(xmit_hash_policy, charp, 0);
 160 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 161                                    "0 for layer 2 (default), 1 for layer 3+4, "
 162                                    "2 for layer 2+3");
 163 module_param(arp_interval, int, 0);
 164 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 165 module_param_array(arp_ip_target, charp, NULL, 0);
 166 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 167 module_param(arp_validate, charp, 0);
 168 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 169                                "0 for none (default), 1 for active, "
 170                                "2 for backup, 3 for all");
 171 module_param(arp_all_targets, charp, 0);
 172 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 173 module_param(fail_over_mac, charp, 0);
 174 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 175                                 "the same MAC; 0 for none (default), "
 176                                 "1 for active, 2 for follow");
 177 module_param(all_slaves_active, int, 0);
 178 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 179                                      "by setting active flag for all slaves; "
 180                                      "0 for never (default), 1 for always.");
 181 module_param(resend_igmp, int, 0);
 182 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 183                               "link failure");
 184
 185 /*----------------------------- Global variables ----------------------------*/
 186
 187 #ifdef CONFIG_NET_POLL_CONTROLLER
 188 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 189 #endif
 190
 191 int bond_net_id __read_mostly;
 192
 193 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 194 static int arp_ip_count;
 195 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 196 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 197 static int lacp_fast;
 198
 199 const struct bond_parm_tbl bond_lacp_tbl[] = {
 200 {       "slow",         AD_LACP_SLOW},
 201 {       "fast",         AD_LACP_FAST},
 202 {       NULL,           -1},
 203 };
 204
 205 const struct bond_parm_tbl bond_mode_tbl[] = {
 206 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 207 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 208 {       "balance-xor",          BOND_MODE_XOR},
 209 {       "broadcast",            BOND_MODE_BROADCAST},
 210 {       "802.3ad",              BOND_MODE_8023AD},
 211 {       "balance-tlb",          BOND_MODE_TLB},
 212 {       "balance-alb",          BOND_MODE_ALB},
 213 {       NULL,                   -1},
 214 };
 215
 216 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 217 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 218 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 219 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 220 {       NULL,                   -1},
 221 };
 222
 223 const struct bond_parm_tbl arp_all_targets_tbl[] = {
 224 {       "any",                  BOND_ARP_TARGETS_ANY},
 225 {       "all",                  BOND_ARP_TARGETS_ALL},
 226 {       NULL,                   -1},
 227 };
 228
 229 const struct bond_parm_tbl arp_validate_tbl[] = {
 230 {       "none",                 BOND_ARP_VALIDATE_NONE},
 231 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 232 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 233 {       "all",                  BOND_ARP_VALIDATE_ALL},
 234 {       NULL,                   -1},
 235 };
 236
 237 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 238 {       "none",                 BOND_FOM_NONE},
 239 {       "active",               BOND_FOM_ACTIVE},
 240 {       "follow",               BOND_FOM_FOLLOW},
 241 {       NULL,                   -1},
 242 };
 243
 244 const struct bond_parm_tbl pri_reselect_tbl[] = {
 245 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 246 {       "better",               BOND_PRI_RESELECT_BETTER},
 247 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 248 {       NULL,                   -1},
 249 };
 250
 251 struct bond_parm_tbl ad_select_tbl[] = {
 252 {       "stable",       BOND_AD_STABLE},
 253 {       "bandwidth",    BOND_AD_BANDWIDTH},
 254 {       "count",        BOND_AD_COUNT},
 255 {       NULL,           -1},
 256 };
 257
 258 /*-------------------------- Forward declarations ---------------------------*/
 259
 260 static int bond_init(struct net_device *bond_dev);
 261 static void bond_uninit(struct net_device *bond_dev);
 262
 263 /*---------------------------- General routines -----------------------------*/
 264
 265 const char *bond_mode_name(int mode)
 266 {
 267         static const char *names[] = {
 268                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 269                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 270                 [BOND_MODE_XOR] = "load balancing (xor)",
 271                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 272                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 273                 [BOND_MODE_TLB] = "transmit load balancing",
 274                 [BOND_MODE_ALB] = "adaptive load balancing",
 275         };
 276
 277         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 278                 return "unknown";
 279
 280         return names[mode];
 281 }
 282
 283 /*---------------------------------- VLAN -----------------------------------*/
 284
 285 /**
 286  * bond_dev_queue_xmit - Prepare skb for xmit.
 287  *
 288  * @bond: bond device that got this skb for tx.
 289  * @skb: hw accel VLAN tagged skb to transmit
 290  * @slave_dev: slave that is supposed to xmit this skbuff
 291  */
 292 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 293                         struct net_device *slave_dev)
 294 {
 295         skb->dev = slave_dev;
 296
 297         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 298                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 299         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 300
 301         if (unlikely(netpoll_tx_running(bond->dev)))
 302                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 303         else
 304                 dev_queue_xmit(skb);
 305
 306         return 0;
 307 }
 308
 309 /*
 310  * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 311  * We don't protect the slave list iteration with a lock because:
 312  * a. This operation is performed in IOCTL context,
 313  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 314  * c. Holding a lock with BH disabled while directly calling a base driver
 315  *    entry point is generally a BAD idea.
 316  *
 317  * The design of synchronization/protection for this operation in the 8021q
 318  * module is good for one or more VLAN devices over a single physical device
 319  * and cannot be extended for a teaming solution like bonding, so there is a
 320  * potential race condition here where a net device from the vlan group might
 321  * be referenced (either by a base driver or the 8021q code) while it is being
 322  * removed from the system. However, it turns out we're not making matters
 323  * worse, and if it works for regular VLAN usage it will work here too.
 324 */
 325
 326 /**
 327  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 328  * @bond_dev: bonding net device that got called
 329  * @vid: vlan id being added
 330  */
 331 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 332                                 __be16 proto, u16 vid)
 333 {
 334         struct bonding *bond = netdev_priv(bond_dev);
 335         struct slave *slave, *rollback_slave;
 336         struct list_head *iter;
 337         int res;
 338
 339         bond_for_each_slave(bond, slave, iter) {
 340                 res = vlan_vid_add(slave->dev, proto, vid);
 341                 if (res)
 342                         goto unwind;
 343         }
 344
 345         return 0;
 346
 347 unwind:
 348         /* unwind to the slave that failed */
 349         bond_for_each_slave(bond, rollback_slave, iter) {
 350                 if (rollback_slave == slave)
 351                         break;
 352
 353                 vlan_vid_del(rollback_slave->dev, proto, vid);
 354         }
 355
 356         return res;
 357 }
 358
 359 /**
 360  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 361  * @bond_dev: bonding net device that got called
 362  * @vid: vlan id being removed
 363  */
 364 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 365                                  __be16 proto, u16 vid)
 366 {
 367         struct bonding *bond = netdev_priv(bond_dev);
 368         struct list_head *iter;
 369         struct slave *slave;
 370
 371         bond_for_each_slave(bond, slave, iter)
 372                 vlan_vid_del(slave->dev, proto, vid);
 373
 374         if (bond_is_lb(bond))
 375                 bond_alb_clear_vlan(bond, vid);
 376
 377         return 0;
 378 }
 379
 380 /*------------------------------- Link status -------------------------------*/
 381
 382 /*
 383  * Set the carrier state for the master according to the state of its
 384  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 385  * do special 802.3ad magic.
 386  *
 387  * Returns zero if carrier state does not change, nonzero if it does.
 388  */
 389 static int bond_set_carrier(struct bonding *bond)
 390 {
 391         struct list_head *iter;
 392         struct slave *slave;
 393
 394         if (!bond_has_slaves(bond))
 395                 goto down;
 396
 397         if (bond->params.mode == BOND_MODE_8023AD)
 398                 return bond_3ad_set_carrier(bond);
 399
 400         bond_for_each_slave(bond, slave, iter) {
 401                 if (slave->link == BOND_LINK_UP) {
 402                         if (!netif_carrier_ok(bond->dev)) {
 403                                 netif_carrier_on(bond->dev);
 404                                 return 1;
 405                         }
 406                         return 0;
 407                 }
 408         }
 409
 410 down:
 411         if (netif_carrier_ok(bond->dev)) {
 412                 netif_carrier_off(bond->dev);
 413                 return 1;
 414         }
 415         return 0;
 416 }
 417
 418 /*
 419  * Get link speed and duplex from the slave's base driver
 420  * using ethtool. If for some reason the call fails or the
 421  * values are invalid, set speed and duplex to -1,
 422  * and return.
 423  */
 424 static void bond_update_speed_duplex(struct slave *slave)
 425 {
 426         struct net_device *slave_dev = slave->dev;
 427         struct ethtool_cmd ecmd;
 428         u32 slave_speed;
 429         int res;
 430
 431         slave->speed = SPEED_UNKNOWN;
 432         slave->duplex = DUPLEX_UNKNOWN;
 433
 434         res = __ethtool_get_settings(slave_dev, &ecmd);
 435         if (res < 0)
 436                 return;
 437
 438         slave_speed = ethtool_cmd_speed(&ecmd);
 439         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 440                 return;
 441
 442         switch (ecmd.duplex) {
 443         case DUPLEX_FULL:
 444         case DUPLEX_HALF:
 445                 break;
 446         default:
 447                 return;
 448         }
 449
 450         slave->speed = slave_speed;
 451         slave->duplex = ecmd.duplex;
 452
 453         return;
 454 }
 455
 456 /*
 457  * if <dev> supports MII link status reporting, check its link status.
 458  *
 459  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 460  * depending upon the setting of the use_carrier parameter.
 461  *
 462  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 463  * can't tell and just pretend it is), or 0, meaning that the link is
 464  * down.
 465  *
 466  * If reporting is non-zero, instead of faking link up, return -1 if
 467  * both ETHTOOL and MII ioctls fail (meaning the device does not
 468  * support them).  If use_carrier is set, return whatever it says.
 469  * It'd be nice if there was a good way to tell if a driver supports
 470  * netif_carrier, but there really isn't.
 471  */
 472 static int bond_check_dev_link(struct bonding *bond,
 473                                struct net_device *slave_dev, int reporting)
 474 {
 475         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 476         int (*ioctl)(struct net_device *, struct ifreq *, int);
 477         struct ifreq ifr;
 478         struct mii_ioctl_data *mii;
 479
 480         if (!reporting && !netif_running(slave_dev))
 481                 return 0;
 482
 483         if (bond->params.use_carrier)
 484                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 485
 486         /* Try to get link status using Ethtool first. */
 487         if (slave_dev->ethtool_ops->get_link)
 488                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 489                         BMSR_LSTATUS : 0;
 490
 491         /* Ethtool can't be used, fallback to MII ioctls. */
 492         ioctl = slave_ops->ndo_do_ioctl;
 493         if (ioctl) {
 494                 /* TODO: set pointer to correct ioctl on a per team member */
 495                 /*       bases to make this more efficient. that is, once  */
 496                 /*       we determine the correct ioctl, we will always    */
 497                 /*       call it and not the others for that team          */
 498                 /*       member.                                           */
 499
 500                 /*
 501                  * We cannot assume that SIOCGMIIPHY will also read a
 502                  * register; not all network drivers (e.g., e100)
 503                  * support that.
 504                  */
 505
 506                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 507                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 508                 mii = if_mii(&ifr);
 509                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 510                         mii->reg_num = MII_BMSR;
 511                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 512                                 return mii->val_out & BMSR_LSTATUS;
 513                 }
 514         }
 515
 516         /*
 517          * If reporting, report that either there's no dev->do_ioctl,
 518          * or both SIOCGMIIREG and get_link failed (meaning that we
 519          * cannot report link status).  If not reporting, pretend
 520          * we're ok.
 521          */
 522         return reporting ? -1 : BMSR_LSTATUS;
 523 }
 524
 525 /*----------------------------- Multicast list ------------------------------*/
 526
 527 /*
 528  * Push the promiscuity flag down to appropriate slaves
 529  */
 530 static int bond_set_promiscuity(struct bonding *bond, int inc)
 531 {
 532         struct list_head *iter;
 533         int err = 0;
 534
 535         if (USES_PRIMARY(bond->params.mode)) {
 536                 /* write lock already acquired */
 537                 if (bond->curr_active_slave) {
 538                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 539                                                   inc);
 540                 }
 541         } else {
 542                 struct slave *slave;
 543
 544                 bond_for_each_slave(bond, slave, iter) {
 545                         err = dev_set_promiscuity(slave->dev, inc);
 546                         if (err)
 547                                 return err;
 548                 }
 549         }
 550         return err;
 551 }
 552
 553 /*
 554  * Push the allmulti flag down to all slaves
 555  */
 556 static int bond_set_allmulti(struct bonding *bond, int inc)
 557 {
 558         struct list_head *iter;
 559         int err = 0;
 560
 561         if (USES_PRIMARY(bond->params.mode)) {
 562                 /* write lock already acquired */
 563                 if (bond->curr_active_slave) {
 564                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 565                                                inc);
 566                 }
 567         } else {
 568                 struct slave *slave;
 569
 570                 bond_for_each_slave(bond, slave, iter) {
 571                         err = dev_set_allmulti(slave->dev, inc);
 572                         if (err)
 573                                 return err;
 574                 }
 575         }
 576         return err;
 577 }
 578
 579 /*
 580  * Retrieve the list of registered multicast addresses for the bonding
 581  * device and retransmit an IGMP JOIN request to the current active
 582  * slave.
 583  */
 584 static void bond_resend_igmp_join_requests(struct bonding *bond)
 585 {
 586         if (!rtnl_trylock()) {
 587                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 588                 return;
 589         }
 590         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 591         rtnl_unlock();
 592
 593         /* We use curr_slave_lock to protect against concurrent access to
 594          * igmp_retrans from multiple running instances of this function and
 595          * bond_change_active_slave
 596          */
 597         write_lock_bh(&bond->curr_slave_lock);
 598         if (bond->igmp_retrans > 1) {
 599                 bond->igmp_retrans--;
 600                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 601         }
 602         write_unlock_bh(&bond->curr_slave_lock);
 603 }
 604
 605 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 606 {
 607         struct bonding *bond = container_of(work, struct bonding,
 608                                             mcast_work.work);
 609
 610         bond_resend_igmp_join_requests(bond);
 611 }
 612
 613 /* Flush bond's hardware addresses from slave
 614  */
 615 static void bond_hw_addr_flush(struct net_device *bond_dev,
 616                                struct net_device *slave_dev)
 617 {
 618         struct bonding *bond = netdev_priv(bond_dev);
 619
 620         dev_uc_unsync(slave_dev, bond_dev);
 621         dev_mc_unsync(slave_dev, bond_dev);
 622
 623         if (bond->params.mode == BOND_MODE_8023AD) {
 624                 /* del lacpdu mc addr from mc list */
 625                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 626
 627                 dev_mc_del(slave_dev, lacpdu_multicast);
 628         }
 629 }
 630
 631 /*--------------------------- Active slave change ---------------------------*/
 632
 633 /* Update the hardware address list and promisc/allmulti for the new and
 634  * old active slaves (if any).  Modes that are !USES_PRIMARY keep all
 635  * slaves up date at all times; only the USES_PRIMARY modes need to call
 636  * this function to swap these settings during a failover.
 637  */
 638 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 639                               struct slave *old_active)
 640 {
 641         ASSERT_RTNL();
 642
 643         if (old_active) {
 644                 if (bond->dev->flags & IFF_PROMISC)
 645                         dev_set_promiscuity(old_active->dev, -1);
 646
 647                 if (bond->dev->flags & IFF_ALLMULTI)
 648                         dev_set_allmulti(old_active->dev, -1);
 649
 650                 bond_hw_addr_flush(bond->dev, old_active->dev);
 651         }
 652
 653         if (new_active) {
 654                 /* FIXME: Signal errors upstream. */
 655                 if (bond->dev->flags & IFF_PROMISC)
 656                         dev_set_promiscuity(new_active->dev, 1);
 657
 658                 if (bond->dev->flags & IFF_ALLMULTI)
 659                         dev_set_allmulti(new_active->dev, 1);
 660
 661                 netif_addr_lock_bh(bond->dev);
 662                 dev_uc_sync(new_active->dev, bond->dev);
 663                 dev_mc_sync(new_active->dev, bond->dev);
 664                 netif_addr_unlock_bh(bond->dev);
 665         }
 666 }
 667
 668 /**
 669  * bond_set_dev_addr - clone slave's address to bond
 670  * @bond_dev: bond net device
 671  * @slave_dev: slave net device
 672  *
 673  * Should be called with RTNL held.
 674  */
 675 static void bond_set_dev_addr(struct net_device *bond_dev,
 676                               struct net_device *slave_dev)
 677 {
 678         pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 679                  bond_dev, slave_dev, slave_dev->addr_len);
 680         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 681         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 682         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 683 }
 684
 685 /*
 686  * bond_do_fail_over_mac
 687  *
 688  * Perform special MAC address swapping for fail_over_mac settings
 689  *
 690  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 691  */
 692 static void bond_do_fail_over_mac(struct bonding *bond,
 693                                   struct slave *new_active,
 694                                   struct slave *old_active)
 695         __releases(&bond->curr_slave_lock)
 696         __releases(&bond->lock)
 697         __acquires(&bond->lock)
 698         __acquires(&bond->curr_slave_lock)
 699 {
 700         u8 tmp_mac[ETH_ALEN];
 701         struct sockaddr saddr;
 702         int rv;
 703
 704         switch (bond->params.fail_over_mac) {
 705         case BOND_FOM_ACTIVE:
 706                 if (new_active) {
 707                         write_unlock_bh(&bond->curr_slave_lock);
 708                         read_unlock(&bond->lock);
 709                         bond_set_dev_addr(bond->dev, new_active->dev);
 710                         read_lock(&bond->lock);
 711                         write_lock_bh(&bond->curr_slave_lock);
 712                 }
 713                 break;
 714         case BOND_FOM_FOLLOW:
 715                 /*
 716                  * if new_active && old_active, swap them
 717                  * if just old_active, do nothing (going to no active slave)
 718                  * if just new_active, set new_active to bond's MAC
 719                  */
 720                 if (!new_active)
 721                         return;
 722
 723                 write_unlock_bh(&bond->curr_slave_lock);
 724                 read_unlock(&bond->lock);
 725
 726                 if (old_active) {
 727                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 728                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 729                                ETH_ALEN);
 730                         saddr.sa_family = new_active->dev->type;
 731                 } else {
 732                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 733                         saddr.sa_family = bond->dev->type;
 734                 }
 735
 736                 rv = dev_set_mac_address(new_active->dev, &saddr);
 737                 if (rv) {
 738                         pr_err("%s: Error %d setting MAC of slave %s\n",
 739                                bond->dev->name, -rv, new_active->dev->name);
 740                         goto out;
 741                 }
 742
 743                 if (!old_active)
 744                         goto out;
 745
 746                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 747                 saddr.sa_family = old_active->dev->type;
 748
 749                 rv = dev_set_mac_address(old_active->dev, &saddr);
 750                 if (rv)
 751                         pr_err("%s: Error %d setting MAC of slave %s\n",
 752                                bond->dev->name, -rv, new_active->dev->name);
 753 out:
 754                 read_lock(&bond->lock);
 755                 write_lock_bh(&bond->curr_slave_lock);
 756                 break;
 757         default:
 758                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 759                        bond->dev->name, bond->params.fail_over_mac);
 760                 break;
 761         }
 762
 763 }
 764
 765 static bool bond_should_change_active(struct bonding *bond)
 766 {
 767         struct slave *prim = bond->primary_slave;
 768         struct slave *curr = bond->curr_active_slave;
 769
 770         if (!prim || !curr || curr->link != BOND_LINK_UP)
 771                 return true;
 772         if (bond->force_primary) {
 773                 bond->force_primary = false;
 774                 return true;
 775         }
 776         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 777             (prim->speed < curr->speed ||
 778              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 779                 return false;
 780         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 781                 return false;
 782         return true;
 783 }
 784
 785 /**
 786  * find_best_interface - select the best available slave to be the active one
 787  * @bond: our bonding struct
 788  */
 789 static struct slave *bond_find_best_slave(struct bonding *bond)
 790 {
 791         struct slave *slave, *bestslave = NULL;
 792         struct list_head *iter;
 793         int mintime = bond->params.updelay;
 794
 795         if (bond->primary_slave && bond->primary_slave->link == BOND_LINK_UP &&
 796             bond_should_change_active(bond))
 797                 return bond->primary_slave;
 798
 799         bond_for_each_slave(bond, slave, iter) {
 800                 if (slave->link == BOND_LINK_UP)
 801                         return slave;
 802                 if (slave->link == BOND_LINK_BACK && IS_UP(slave->dev) &&
 803                     slave->delay < mintime) {
 804                         mintime = slave->delay;
 805                         bestslave = slave;
 806                 }
 807         }
 808
 809         return bestslave;
 810 }
 811
 812 static bool bond_should_notify_peers(struct bonding *bond)
 813 {
 814         struct slave *slave = bond->curr_active_slave;
 815
 816         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
 817                  bond->dev->name, slave ? slave->dev->name : "NULL");
 818
 819         if (!slave || !bond->send_peer_notif ||
 820             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 821                 return false;
 822
 823         return true;
 824 }
 825
 826 /**
 827  * change_active_interface - change the active slave into the specified one
 828  * @bond: our bonding struct
 829  * @new: the new slave to make the active one
 830  *
 831  * Set the new slave to the bond's settings and unset them on the old
 832  * curr_active_slave.
 833  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 834  *
 835  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 836  * because it is apparently the best available slave we have, even though its
 837  * updelay hasn't timed out yet.
 838  *
 839  * If new_active is not NULL, caller must hold bond->lock for read and
 840  * curr_slave_lock for write_bh.
 841  */
 842 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 843 {
 844         struct slave *old_active = bond->curr_active_slave;
 845
 846         if (old_active == new_active)
 847                 return;
 848
 849         if (new_active) {
 850                 new_active->jiffies = jiffies;
 851
 852                 if (new_active->link == BOND_LINK_BACK) {
 853                         if (USES_PRIMARY(bond->params.mode)) {
 854                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
 855                                         bond->dev->name, new_active->dev->name,
 856                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
 857                         }
 858
 859                         new_active->delay = 0;
 860                         new_active->link = BOND_LINK_UP;
 861
 862                         if (bond->params.mode == BOND_MODE_8023AD)
 863                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 864
 865                         if (bond_is_lb(bond))
 866                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 867                 } else {
 868                         if (USES_PRIMARY(bond->params.mode)) {
 869                                 pr_info("%s: making interface %s the new active one.\n",
 870                                         bond->dev->name, new_active->dev->name);
 871                         }
 872                 }
 873         }
 874
 875         if (USES_PRIMARY(bond->params.mode))
 876                 bond_hw_addr_swap(bond, new_active, old_active);
 877
 878         if (bond_is_lb(bond)) {
 879                 bond_alb_handle_active_change(bond, new_active);
 880                 if (old_active)
 881                         bond_set_slave_inactive_flags(old_active);
 882                 if (new_active)
 883                         bond_set_slave_active_flags(new_active);
 884         } else {
 885                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 886         }
 887
 888         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
 889                 if (old_active)
 890                         bond_set_slave_inactive_flags(old_active);
 891
 892                 if (new_active) {
 893                         bool should_notify_peers = false;
 894
 895                         bond_set_slave_active_flags(new_active);
 896
 897                         if (bond->params.fail_over_mac)
 898                                 bond_do_fail_over_mac(bond, new_active,
 899                                                       old_active);
 900
 901                         if (netif_running(bond->dev)) {
 902                                 bond->send_peer_notif =
 903                                         bond->params.num_peer_notif;
 904                                 should_notify_peers =
 905                                         bond_should_notify_peers(bond);
 906                         }
 907
 908                         write_unlock_bh(&bond->curr_slave_lock);
 909                         read_unlock(&bond->lock);
 910
 911                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 912                         if (should_notify_peers)
 913                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 914                                                          bond->dev);
 915
 916                         read_lock(&bond->lock);
 917                         write_lock_bh(&bond->curr_slave_lock);
 918                 }
 919         }
 920
 921         /* resend IGMP joins since active slave has changed or
 922          * all were sent on curr_active_slave.
 923          * resend only if bond is brought up with the affected
 924          * bonding modes and the retransmission is enabled */
 925         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 926             ((USES_PRIMARY(bond->params.mode) && new_active) ||
 927              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
 928                 bond->igmp_retrans = bond->params.resend_igmp;
 929                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 930         }
 931 }
 932
 933 /**
 934  * bond_select_active_slave - select a new active slave, if needed
 935  * @bond: our bonding struct
 936  *
 937  * This functions should be called when one of the following occurs:
 938  * - The old curr_active_slave has been released or lost its link.
 939  * - The primary_slave has got its link back.
 940  * - A slave has got its link back and there's no old curr_active_slave.
 941  *
 942  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
 943  */
 944 void bond_select_active_slave(struct bonding *bond)
 945 {
 946         struct slave *best_slave;
 947         int rv;
 948
 949         best_slave = bond_find_best_slave(bond);
 950         if (best_slave != bond->curr_active_slave) {
 951                 bond_change_active_slave(bond, best_slave);
 952                 rv = bond_set_carrier(bond);
 953                 if (!rv)
 954                         return;
 955
 956                 if (netif_carrier_ok(bond->dev)) {
 957                         pr_info("%s: first active interface up!\n",
 958                                 bond->dev->name);
 959                 } else {
 960                         pr_info("%s: now running without any active interface !\n",
 961                                 bond->dev->name);
 962                 }
 963         }
 964 }
 965
 966 /*--------------------------- slave list handling ---------------------------*/
 967
 968 /*
 969  * This function attaches the slave to the end of list.
 970  *
 971  * bond->lock held for writing by caller.
 972  */
 973 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
 974 {
 975         bond->slave_cnt++;
 976 }
 977
 978 /*
 979  * This function detaches the slave from the list.
 980  * WARNING: no check is made to verify if the slave effectively
 981  * belongs to <bond>.
 982  * Nothing is freed on return, structures are just unchained.
 983  * If any slave pointer in bond was pointing to <slave>,
 984  * it should be changed by the calling function.
 985  *
 986  * bond->lock held for writing by caller.
 987  */
 988 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
 989 {
 990         bond->slave_cnt--;
 991 }
 992
 993 #ifdef CONFIG_NET_POLL_CONTROLLER
 994 static inline int slave_enable_netpoll(struct slave *slave)
 995 {
 996         struct netpoll *np;
 997         int err = 0;
 998
 999         np = kzalloc(sizeof(*np), GFP_ATOMIC);
1000         err = -ENOMEM;
1001         if (!np)
1002                 goto out;
1003
1004         err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
1005         if (err) {
1006                 kfree(np);
1007                 goto out;
1008         }
1009         slave->np = np;
1010 out:
1011         return err;
1012 }
1013 static inline void slave_disable_netpoll(struct slave *slave)
1014 {
1015         struct netpoll *np = slave->np;
1016
1017         if (!np)
1018                 return;
1019
1020         slave->np = NULL;
1021         __netpoll_free_async(np);
1022 }
1023 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1024 {
1025         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1026                 return false;
1027         if (!slave_dev->netdev_ops->ndo_poll_controller)
1028                 return false;
1029         return true;
1030 }
1031
1032 static void bond_poll_controller(struct net_device *bond_dev)
1033 {
1034 }
1035
1036 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1037 {
1038         struct bonding *bond = netdev_priv(bond_dev);
1039         struct list_head *iter;
1040         struct slave *slave;
1041
1042         bond_for_each_slave(bond, slave, iter)
1043                 if (IS_UP(slave->dev))
1044                         slave_disable_netpoll(slave);
1045 }
1046
1047 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1048 {
1049         struct bonding *bond = netdev_priv(dev);
1050         struct list_head *iter;
1051         struct slave *slave;
1052         int err = 0;
1053
1054         bond_for_each_slave(bond, slave, iter) {
1055                 err = slave_enable_netpoll(slave);
1056                 if (err) {
1057                         bond_netpoll_cleanup(dev);
1058                         break;
1059                 }
1060         }
1061         return err;
1062 }
1063 #else
1064 static inline int slave_enable_netpoll(struct slave *slave)
1065 {
1066         return 0;
1067 }
1068 static inline void slave_disable_netpoll(struct slave *slave)
1069 {
1070 }
1071 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1072 {
1073 }
1074 #endif
1075
1076 /*---------------------------------- IOCTL ----------------------------------*/
1077
1078 static netdev_features_t bond_fix_features(struct net_device *dev,
1079                                            netdev_features_t features)
1080 {
1081         struct bonding *bond = netdev_priv(dev);
1082         struct list_head *iter;
1083         netdev_features_t mask;
1084         struct slave *slave;
1085
1086         if (!bond_has_slaves(bond)) {
1087                 /* Disable adding VLANs to empty bond. But why? --mq */
1088                 features |= NETIF_F_VLAN_CHALLENGED;
1089                 return features;
1090         }
1091
1092         mask = features;
1093         features &= ~NETIF_F_ONE_FOR_ALL;
1094         features |= NETIF_F_ALL_FOR_ALL;
1095
1096         bond_for_each_slave(bond, slave, iter) {
1097                 features = netdev_increment_features(features,
1098                                                      slave->dev->features,
1099                                                      mask);
1100         }
1101         features = netdev_add_tso_features(features, mask);
1102
1103         return features;
1104 }
1105
1106 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1107                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1108                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1109
1110 static void bond_compute_features(struct bonding *bond)
1111 {
1112         unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1113         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1114         struct net_device *bond_dev = bond->dev;
1115         struct list_head *iter;
1116         struct slave *slave;
1117         unsigned short max_hard_header_len = ETH_HLEN;
1118         unsigned int gso_max_size = GSO_MAX_SIZE;
1119         u16 gso_max_segs = GSO_MAX_SEGS;
1120
1121         if (!bond_has_slaves(bond))
1122                 goto done;
1123
1124         bond_for_each_slave(bond, slave, iter) {
1125                 vlan_features = netdev_increment_features(vlan_features,
1126                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1127
1128                 dst_release_flag &= slave->dev->priv_flags;
1129                 if (slave->dev->hard_header_len > max_hard_header_len)
1130                         max_hard_header_len = slave->dev->hard_header_len;
1131
1132                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1133                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1134         }
1135
1136 done:
1137         bond_dev->vlan_features = vlan_features;
1138         bond_dev->hard_header_len = max_hard_header_len;
1139         bond_dev->gso_max_segs = gso_max_segs;
1140         netif_set_gso_max_size(bond_dev, gso_max_size);
1141
1142         flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1143         bond_dev->priv_flags = flags | dst_release_flag;
1144
1145         netdev_change_features(bond_dev);
1146 }
1147
1148 static void bond_setup_by_slave(struct net_device *bond_dev,
1149                                 struct net_device *slave_dev)
1150 {
1151         bond_dev->header_ops        = slave_dev->header_ops;
1152
1153         bond_dev->type              = slave_dev->type;
1154         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1155         bond_dev->addr_len          = slave_dev->addr_len;
1156
1157         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1158                 slave_dev->addr_len);
1159 }
1160
1161 /* On bonding slaves other than the currently active slave, suppress
1162  * duplicates except for alb non-mcast/bcast.
1163  */
1164 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1165                                             struct slave *slave,
1166                                             struct bonding *bond)
1167 {
1168         if (bond_is_slave_inactive(slave)) {
1169                 if (bond->params.mode == BOND_MODE_ALB &&
1170                     skb->pkt_type != PACKET_BROADCAST &&
1171                     skb->pkt_type != PACKET_MULTICAST)
1172                         return false;
1173                 return true;
1174         }
1175         return false;
1176 }
1177
1178 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1179 {
1180         struct sk_buff *skb = *pskb;
1181         struct slave *slave;
1182         struct bonding *bond;
1183         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1184                           struct slave *);
1185         int ret = RX_HANDLER_ANOTHER;
1186
1187         skb = skb_share_check(skb, GFP_ATOMIC);
1188         if (unlikely(!skb))
1189                 return RX_HANDLER_CONSUMED;
1190
1191         *pskb = skb;
1192
1193         slave = bond_slave_get_rcu(skb->dev);
1194         bond = slave->bond;
1195
1196         if (bond->params.arp_interval)
1197                 slave->dev->last_rx = jiffies;
1198
1199         recv_probe = ACCESS_ONCE(bond->recv_probe);
1200         if (recv_probe) {
1201                 ret = recv_probe(skb, bond, slave);
1202                 if (ret == RX_HANDLER_CONSUMED) {
1203                         consume_skb(skb);
1204                         return ret;
1205                 }
1206         }
1207
1208         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1209                 return RX_HANDLER_EXACT;
1210         }
1211
1212         skb->dev = bond->dev;
1213
1214         if (bond->params.mode == BOND_MODE_ALB &&
1215             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1216             skb->pkt_type == PACKET_HOST) {
1217
1218                 if (unlikely(skb_cow_head(skb,
1219                                           skb->data - skb_mac_header(skb)))) {
1220                         kfree_skb(skb);
1221                         return RX_HANDLER_CONSUMED;
1222                 }
1223                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1224         }
1225
1226         return ret;
1227 }
1228
1229 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1230                                       struct net_device *slave_dev,
1231                                       struct slave *slave)
1232 {
1233         int err;
1234
1235         err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1236         if (err)
1237                 return err;
1238         slave_dev->flags |= IFF_SLAVE;
1239         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1240         return 0;
1241 }
1242
1243 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1244                                   struct net_device *slave_dev)
1245 {
1246         netdev_upper_dev_unlink(slave_dev, bond_dev);
1247         slave_dev->flags &= ~IFF_SLAVE;
1248         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1249 }
1250
1251 /* enslave device <slave> to bond device <master> */
1252 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1253 {
1254         struct bonding *bond = netdev_priv(bond_dev);
1255         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1256         struct slave *new_slave = NULL, *prev_slave;
1257         struct sockaddr addr;
1258         int link_reporting;
1259         int res = 0, i;
1260
1261         if (!bond->params.use_carrier &&
1262             slave_dev->ethtool_ops->get_link == NULL &&
1263             slave_ops->ndo_do_ioctl == NULL) {
1264                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1265                            bond_dev->name, slave_dev->name);
1266         }
1267
1268         /* already enslaved */
1269         if (slave_dev->flags & IFF_SLAVE) {
1270                 pr_debug("Error, Device was already enslaved\n");
1271                 return -EBUSY;
1272         }
1273
1274         /* vlan challenged mutual exclusion */
1275         /* no need to lock since we're protected by rtnl_lock */
1276         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1277                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1278                 if (vlan_uses_dev(bond_dev)) {
1279                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1280                                bond_dev->name, slave_dev->name, bond_dev->name);
1281                         return -EPERM;
1282                 } else {
1283                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1284                                    bond_dev->name, slave_dev->name,
1285                                    slave_dev->name, bond_dev->name);
1286                 }
1287         } else {
1288                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1289         }
1290
1291         /*
1292          * Old ifenslave binaries are no longer supported.  These can
1293          * be identified with moderate accuracy by the state of the slave:
1294          * the current ifenslave will set the interface down prior to
1295          * enslaving it; the old ifenslave will not.
1296          */
1297         if ((slave_dev->flags & IFF_UP)) {
1298                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1299                        slave_dev->name);
1300                 res = -EPERM;
1301                 goto err_undo_flags;
1302         }
1303
1304         /* set bonding device ether type by slave - bonding netdevices are
1305          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1306          * there is a need to override some of the type dependent attribs/funcs.
1307          *
1308          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1309          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1310          */
1311         if (!bond_has_slaves(bond)) {
1312                 if (bond_dev->type != slave_dev->type) {
1313                         pr_debug("%s: change device type from %d to %d\n",
1314                                  bond_dev->name,
1315                                  bond_dev->type, slave_dev->type);
1316
1317                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1318                                                        bond_dev);
1319                         res = notifier_to_errno(res);
1320                         if (res) {
1321                                 pr_err("%s: refused to change device type\n",
1322                                        bond_dev->name);
1323                                 res = -EBUSY;
1324                                 goto err_undo_flags;
1325                         }
1326
1327                         /* Flush unicast and multicast addresses */
1328                         dev_uc_flush(bond_dev);
1329                         dev_mc_flush(bond_dev);
1330
1331                         if (slave_dev->type != ARPHRD_ETHER)
1332                                 bond_setup_by_slave(bond_dev, slave_dev);
1333                         else {
1334                                 ether_setup(bond_dev);
1335                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1336                         }
1337
1338                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1339                                                  bond_dev);
1340                 }
1341         } else if (bond_dev->type != slave_dev->type) {
1342                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1343                        slave_dev->name,
1344                        slave_dev->type, bond_dev->type);
1345                 res = -EINVAL;
1346                 goto err_undo_flags;
1347         }
1348
1349         if (slave_ops->ndo_set_mac_address == NULL) {
1350                 if (!bond_has_slaves(bond)) {
1351                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1352                                    bond_dev->name);
1353                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1354                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1355                         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",
1356                                bond_dev->name);
1357                         res = -EOPNOTSUPP;
1358                         goto err_undo_flags;
1359                 }
1360         }
1361
1362         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1363
1364         /* If this is the first slave, then we need to set the master's hardware
1365          * address to be the same as the slave's. */
1366         if (!bond_has_slaves(bond) &&
1367             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1368                 bond_set_dev_addr(bond->dev, slave_dev);
1369
1370         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1371         if (!new_slave) {
1372                 res = -ENOMEM;
1373                 goto err_undo_flags;
1374         }
1375         /*
1376          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1377          * is set via sysfs or module option if desired.
1378          */
1379         new_slave->queue_id = 0;
1380
1381         /* Save slave's original mtu and then set it to match the bond */
1382         new_slave->original_mtu = slave_dev->mtu;
1383         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1384         if (res) {
1385                 pr_debug("Error %d calling dev_set_mtu\n", res);
1386                 goto err_free;
1387         }
1388
1389         /*
1390          * Save slave's original ("permanent") mac address for modes
1391          * that need it, and for restoring it upon release, and then
1392          * set it to the master's address
1393          */
1394         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1395
1396         if (!bond->params.fail_over_mac) {
1397                 /*
1398                  * Set slave to master's mac address.  The application already
1399                  * set the master's mac address to that of the first slave
1400                  */
1401                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1402                 addr.sa_family = slave_dev->type;
1403                 res = dev_set_mac_address(slave_dev, &addr);
1404                 if (res) {
1405                         pr_debug("Error %d calling set_mac_address\n", res);
1406                         goto err_restore_mtu;
1407                 }
1408         }
1409
1410         /* open the slave since the application closed it */
1411         res = dev_open(slave_dev);
1412         if (res) {
1413                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1414                 goto err_restore_mac;
1415         }
1416
1417         new_slave->bond = bond;
1418         new_slave->dev = slave_dev;
1419         slave_dev->priv_flags |= IFF_BONDING;
1420
1421         if (bond_is_lb(bond)) {
1422                 /* bond_alb_init_slave() must be called before all other stages since
1423                  * it might fail and we do not want to have to undo everything
1424                  */
1425                 res = bond_alb_init_slave(bond, new_slave);
1426                 if (res)
1427                         goto err_close;
1428         }
1429
1430         /* If the mode USES_PRIMARY, then the following is handled by
1431          * bond_change_active_slave().
1432          */
1433         if (!USES_PRIMARY(bond->params.mode)) {
1434                 /* set promiscuity level to new slave */
1435                 if (bond_dev->flags & IFF_PROMISC) {
1436                         res = dev_set_promiscuity(slave_dev, 1);
1437                         if (res)
1438                                 goto err_close;
1439                 }
1440
1441                 /* set allmulti level to new slave */
1442                 if (bond_dev->flags & IFF_ALLMULTI) {
1443                         res = dev_set_allmulti(slave_dev, 1);
1444                         if (res)
1445                                 goto err_close;
1446                 }
1447
1448                 netif_addr_lock_bh(bond_dev);
1449
1450                 dev_mc_sync_multiple(slave_dev, bond_dev);
1451                 dev_uc_sync_multiple(slave_dev, bond_dev);
1452
1453                 netif_addr_unlock_bh(bond_dev);
1454         }
1455
1456         if (bond->params.mode == BOND_MODE_8023AD) {
1457                 /* add lacpdu mc addr to mc list */
1458                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1459
1460                 dev_mc_add(slave_dev, lacpdu_multicast);
1461         }
1462
1463         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1464         if (res) {
1465                 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1466                        bond_dev->name, slave_dev->name);
1467                 goto err_close;
1468         }
1469
1470         write_lock_bh(&bond->lock);
1471
1472         prev_slave = bond_last_slave(bond);
1473         bond_attach_slave(bond, new_slave);
1474
1475         new_slave->delay = 0;
1476         new_slave->link_failure_count = 0;
1477
1478         write_unlock_bh(&bond->lock);
1479
1480         bond_compute_features(bond);
1481
1482         bond_update_speed_duplex(new_slave);
1483
1484         read_lock(&bond->lock);
1485
1486         new_slave->last_arp_rx = jiffies -
1487                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1488         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1489                 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
1490
1491         if (bond->params.miimon && !bond->params.use_carrier) {
1492                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1493
1494                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1495                         /*
1496                          * miimon is set but a bonded network driver
1497                          * does not support ETHTOOL/MII and
1498                          * arp_interval is not set.  Note: if
1499                          * use_carrier is enabled, we will never go
1500                          * here (because netif_carrier is always
1501                          * supported); thus, we don't need to change
1502                          * the messages for netif_carrier.
1503                          */
1504                         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",
1505                                bond_dev->name, slave_dev->name);
1506                 } else if (link_reporting == -1) {
1507                         /* unable get link status using mii/ethtool */
1508                         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",
1509                                    bond_dev->name, slave_dev->name);
1510                 }
1511         }
1512
1513         /* check for initial state */
1514         if (bond->params.miimon) {
1515                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1516                         if (bond->params.updelay) {
1517                                 new_slave->link = BOND_LINK_BACK;
1518                                 new_slave->delay = bond->params.updelay;
1519                         } else {
1520                                 new_slave->link = BOND_LINK_UP;
1521                         }
1522                 } else {
1523                         new_slave->link = BOND_LINK_DOWN;
1524                 }
1525         } else if (bond->params.arp_interval) {
1526                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1527                         BOND_LINK_UP : BOND_LINK_DOWN);
1528         } else {
1529                 new_slave->link = BOND_LINK_UP;
1530         }
1531
1532         if (new_slave->link != BOND_LINK_DOWN)
1533                 new_slave->jiffies = jiffies;
1534         pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1535                 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1536                         (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1537
1538         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1539                 /* if there is a primary slave, remember it */
1540                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1541                         bond->primary_slave = new_slave;
1542                         bond->force_primary = true;
1543                 }
1544         }
1545
1546         write_lock_bh(&bond->curr_slave_lock);
1547
1548         switch (bond->params.mode) {
1549         case BOND_MODE_ACTIVEBACKUP:
1550                 bond_set_slave_inactive_flags(new_slave);
1551                 bond_select_active_slave(bond);
1552                 break;
1553         case BOND_MODE_8023AD:
1554                 /* in 802.3ad mode, the internal mechanism
1555                  * will activate the slaves in the selected
1556                  * aggregator
1557                  */
1558                 bond_set_slave_inactive_flags(new_slave);
1559                 /* if this is the first slave */
1560                 if (!prev_slave) {
1561                         SLAVE_AD_INFO(new_slave).id = 1;
1562                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1563                          * can be called only after the mac address of the bond is set
1564                          */
1565                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1566                 } else {
1567                         SLAVE_AD_INFO(new_slave).id =
1568                                 SLAVE_AD_INFO(prev_slave).id + 1;
1569                 }
1570
1571                 bond_3ad_bind_slave(new_slave);
1572                 break;
1573         case BOND_MODE_TLB:
1574         case BOND_MODE_ALB:
1575                 bond_set_active_slave(new_slave);
1576                 bond_set_slave_inactive_flags(new_slave);
1577                 bond_select_active_slave(bond);
1578                 break;
1579         default:
1580                 pr_debug("This slave is always active in trunk mode\n");
1581
1582                 /* always active in trunk mode */
1583                 bond_set_active_slave(new_slave);
1584
1585                 /* In trunking mode there is little meaning to curr_active_slave
1586                  * anyway (it holds no special properties of the bond device),
1587                  * so we can change it without calling change_active_interface()
1588                  */
1589                 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1590                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1591
1592                 break;
1593         } /* switch(bond_mode) */
1594
1595         write_unlock_bh(&bond->curr_slave_lock);
1596
1597         bond_set_carrier(bond);
1598
1599 #ifdef CONFIG_NET_POLL_CONTROLLER
1600         slave_dev->npinfo = bond->dev->npinfo;
1601         if (slave_dev->npinfo) {
1602                 if (slave_enable_netpoll(new_slave)) {
1603                         read_unlock(&bond->lock);
1604                         pr_info("Error, %s: master_dev is using netpoll, "
1605                                  "but new slave device does not support netpoll.\n",
1606                                  bond_dev->name);
1607                         res = -EBUSY;
1608                         goto err_detach;
1609                 }
1610         }
1611 #endif
1612
1613         read_unlock(&bond->lock);
1614
1615         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1616                                          new_slave);
1617         if (res) {
1618                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1619                 goto err_detach;
1620         }
1621
1622         res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1623         if (res) {
1624                 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1625                 goto err_unregister;
1626         }
1627
1628
1629         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1630                 bond_dev->name, slave_dev->name,
1631                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1632                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1633
1634         /* enslave is successful */
1635         return 0;
1636
1637 /* Undo stages on error */
1638 err_unregister:
1639         netdev_rx_handler_unregister(slave_dev);
1640
1641 err_detach:
1642         if (!USES_PRIMARY(bond->params.mode))
1643                 bond_hw_addr_flush(bond_dev, slave_dev);
1644
1645         vlan_vids_del_by_dev(slave_dev, bond_dev);
1646         write_lock_bh(&bond->lock);
1647         bond_detach_slave(bond, new_slave);
1648         if (bond->primary_slave == new_slave)
1649                 bond->primary_slave = NULL;
1650         if (bond->curr_active_slave == new_slave) {
1651                 bond_change_active_slave(bond, NULL);
1652                 write_unlock_bh(&bond->lock);
1653                 read_lock(&bond->lock);
1654                 write_lock_bh(&bond->curr_slave_lock);
1655                 bond_select_active_slave(bond);
1656                 write_unlock_bh(&bond->curr_slave_lock);
1657                 read_unlock(&bond->lock);
1658         } else {
1659                 write_unlock_bh(&bond->lock);
1660         }
1661         slave_disable_netpoll(new_slave);
1662
1663 err_close:
1664         slave_dev->priv_flags &= ~IFF_BONDING;
1665         dev_close(slave_dev);
1666
1667 err_restore_mac:
1668         if (!bond->params.fail_over_mac) {
1669                 /* XXX TODO - fom follow mode needs to change master's
1670                  * MAC if this slave's MAC is in use by the bond, or at
1671                  * least print a warning.
1672                  */
1673                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1674                 addr.sa_family = slave_dev->type;
1675                 dev_set_mac_address(slave_dev, &addr);
1676         }
1677
1678 err_restore_mtu:
1679         dev_set_mtu(slave_dev, new_slave->original_mtu);
1680
1681 err_free:
1682         kfree(new_slave);
1683
1684 err_undo_flags:
1685         bond_compute_features(bond);
1686         /* Enslave of first slave has failed and we need to fix master's mac */
1687         if (!bond_has_slaves(bond) &&
1688             ether_addr_equal(bond_dev->dev_addr, slave_dev->dev_addr))
1689                 eth_hw_addr_random(bond_dev);
1690
1691         return res;
1692 }
1693
1694 /*
1695  * Try to release the slave device <slave> from the bond device <master>
1696  * It is legal to access curr_active_slave without a lock because all the function
1697  * is write-locked. If "all" is true it means that the function is being called
1698  * while destroying a bond interface and all slaves are being released.
1699  *
1700  * The rules for slave state should be:
1701  *   for Active/Backup:
1702  *     Active stays on all backups go down
1703  *   for Bonded connections:
1704  *     The first up interface should be left on and all others downed.
1705  */
1706 static int __bond_release_one(struct net_device *bond_dev,
1707                               struct net_device *slave_dev,
1708                               bool all)
1709 {
1710         struct bonding *bond = netdev_priv(bond_dev);
1711         struct slave *slave, *oldcurrent;
1712         struct sockaddr addr;
1713         int old_flags = bond_dev->flags;
1714         netdev_features_t old_features = bond_dev->features;
1715
1716         /* slave is not a slave or master is not master of this slave */
1717         if (!(slave_dev->flags & IFF_SLAVE) ||
1718             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1719                 pr_err("%s: Error: cannot release %s.\n",
1720                        bond_dev->name, slave_dev->name);
1721                 return -EINVAL;
1722         }
1723
1724         block_netpoll_tx();
1725         write_lock_bh(&bond->lock);
1726
1727         slave = bond_get_slave_by_dev(bond, slave_dev);
1728         if (!slave) {
1729                 /* not a slave of this bond */
1730                 pr_info("%s: %s not enslaved\n",
1731                         bond_dev->name, slave_dev->name);
1732                 write_unlock_bh(&bond->lock);
1733                 unblock_netpoll_tx();
1734                 return -EINVAL;
1735         }
1736
1737         write_unlock_bh(&bond->lock);
1738
1739         bond_upper_dev_unlink(bond_dev, slave_dev);
1740         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1741          * for this slave anymore.
1742          */
1743         netdev_rx_handler_unregister(slave_dev);
1744         write_lock_bh(&bond->lock);
1745
1746         /* Inform AD package of unbinding of slave. */
1747         if (bond->params.mode == BOND_MODE_8023AD) {
1748                 /* must be called before the slave is
1749                  * detached from the list
1750                  */
1751                 bond_3ad_unbind_slave(slave);
1752         }
1753
1754         pr_info("%s: releasing %s interface %s\n",
1755                 bond_dev->name,
1756                 bond_is_active_slave(slave) ? "active" : "backup",
1757                 slave_dev->name);
1758
1759         oldcurrent = bond->curr_active_slave;
1760
1761         bond->current_arp_slave = NULL;
1762
1763         /* release the slave from its bond */
1764         bond_detach_slave(bond, slave);
1765
1766         if (!all && !bond->params.fail_over_mac) {
1767                 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1768                     bond_has_slaves(bond))
1769                         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",
1770                                    bond_dev->name, slave_dev->name,
1771                                    slave->perm_hwaddr,
1772                                    bond_dev->name, slave_dev->name);
1773         }
1774
1775         if (bond->primary_slave == slave)
1776                 bond->primary_slave = NULL;
1777
1778         if (oldcurrent == slave)
1779                 bond_change_active_slave(bond, NULL);
1780
1781         if (bond_is_lb(bond)) {
1782                 /* Must be called only after the slave has been
1783                  * detached from the list and the curr_active_slave
1784                  * has been cleared (if our_slave == old_current),
1785                  * but before a new active slave is selected.
1786                  */
1787                 write_unlock_bh(&bond->lock);
1788                 bond_alb_deinit_slave(bond, slave);
1789                 write_lock_bh(&bond->lock);
1790         }
1791
1792         if (all) {
1793                 rcu_assign_pointer(bond->curr_active_slave, NULL);
1794         } else if (oldcurrent == slave) {
1795                 /*
1796                  * Note that we hold RTNL over this sequence, so there
1797                  * is no concern that another slave add/remove event
1798                  * will interfere.
1799                  */
1800                 write_unlock_bh(&bond->lock);
1801                 read_lock(&bond->lock);
1802                 write_lock_bh(&bond->curr_slave_lock);
1803
1804                 bond_select_active_slave(bond);
1805
1806                 write_unlock_bh(&bond->curr_slave_lock);
1807                 read_unlock(&bond->lock);
1808                 write_lock_bh(&bond->lock);
1809         }
1810
1811         if (!bond_has_slaves(bond)) {
1812                 bond_set_carrier(bond);
1813                 eth_hw_addr_random(bond_dev);
1814
1815                 if (vlan_uses_dev(bond_dev)) {
1816                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1817                                    bond_dev->name, bond_dev->name);
1818                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1819                                    bond_dev->name);
1820                 }
1821         }
1822
1823         write_unlock_bh(&bond->lock);
1824         unblock_netpoll_tx();
1825         synchronize_rcu();
1826
1827         if (!bond_has_slaves(bond)) {
1828                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1829                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1830         }
1831
1832         bond_compute_features(bond);
1833         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1834             (old_features & NETIF_F_VLAN_CHALLENGED))
1835                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1836                         bond_dev->name, slave_dev->name, bond_dev->name);
1837
1838         /* must do this from outside any spinlocks */
1839         vlan_vids_del_by_dev(slave_dev, bond_dev);
1840
1841         /* If the mode USES_PRIMARY, then this cases was handled above by
1842          * bond_change_active_slave(..., NULL)
1843          */
1844         if (!USES_PRIMARY(bond->params.mode)) {
1845                 /* unset promiscuity level from slave
1846                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1847                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1848                  * value of that flag before that change, as that was the value
1849                  * when this slave was attached, so we cache at the start of the
1850                  * function and use it here. Same goes for ALLMULTI below
1851                  */
1852                 if (old_flags & IFF_PROMISC)
1853                         dev_set_promiscuity(slave_dev, -1);
1854
1855                 /* unset allmulti level from slave */
1856                 if (old_flags & IFF_ALLMULTI)
1857                         dev_set_allmulti(slave_dev, -1);
1858
1859                 bond_hw_addr_flush(bond_dev, slave_dev);
1860         }
1861
1862         slave_disable_netpoll(slave);
1863
1864         /* close slave before restoring its mac address */
1865         dev_close(slave_dev);
1866
1867         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1868                 /* restore original ("permanent") mac address */
1869                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1870                 addr.sa_family = slave_dev->type;
1871                 dev_set_mac_address(slave_dev, &addr);
1872         }
1873
1874         dev_set_mtu(slave_dev, slave->original_mtu);
1875
1876         slave_dev->priv_flags &= ~IFF_BONDING;
1877
1878         kfree(slave);
1879
1880         return 0;  /* deletion OK */
1881 }
1882
1883 /* A wrapper used because of ndo_del_link */
1884 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1885 {
1886         return __bond_release_one(bond_dev, slave_dev, false);
1887 }
1888
1889 /*
1890 * First release a slave and then destroy the bond if no more slaves are left.
1891 * Must be under rtnl_lock when this function is called.
1892 */
1893 static int  bond_release_and_destroy(struct net_device *bond_dev,
1894                                      struct net_device *slave_dev)
1895 {
1896         struct bonding *bond = netdev_priv(bond_dev);
1897         int ret;
1898
1899         ret = bond_release(bond_dev, slave_dev);
1900         if (ret == 0 && !bond_has_slaves(bond)) {
1901                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1902                 pr_info("%s: destroying bond %s.\n",
1903                         bond_dev->name, bond_dev->name);
1904                 unregister_netdevice(bond_dev);
1905         }
1906         return ret;
1907 }
1908
1909 /*
1910  * This function changes the active slave to slave <slave_dev>.
1911  * It returns -EINVAL in the following cases.
1912  *  - <slave_dev> is not found in the list.
1913  *  - There is not active slave now.
1914  *  - <slave_dev> is already active.
1915  *  - The link state of <slave_dev> is not BOND_LINK_UP.
1916  *  - <slave_dev> is not running.
1917  * In these cases, this function does nothing.
1918  * In the other cases, current_slave pointer is changed and 0 is returned.
1919  */
1920 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
1921 {
1922         struct bonding *bond = netdev_priv(bond_dev);
1923         struct slave *old_active = NULL;
1924         struct slave *new_active = NULL;
1925         int res = 0;
1926
1927         if (!USES_PRIMARY(bond->params.mode))
1928                 return -EINVAL;
1929
1930         /* Verify that bond_dev is indeed the master of slave_dev */
1931         if (!(slave_dev->flags & IFF_SLAVE) ||
1932             !netdev_has_upper_dev(slave_dev, bond_dev))
1933                 return -EINVAL;
1934
1935         read_lock(&bond->lock);
1936
1937         old_active = bond->curr_active_slave;
1938         new_active = bond_get_slave_by_dev(bond, slave_dev);
1939         /*
1940          * Changing to the current active: do nothing; return success.
1941          */
1942         if (new_active && new_active == old_active) {
1943                 read_unlock(&bond->lock);
1944                 return 0;
1945         }
1946
1947         if (new_active &&
1948             old_active &&
1949             new_active->link == BOND_LINK_UP &&
1950             IS_UP(new_active->dev)) {
1951                 block_netpoll_tx();
1952                 write_lock_bh(&bond->curr_slave_lock);
1953                 bond_change_active_slave(bond, new_active);
1954                 write_unlock_bh(&bond->curr_slave_lock);
1955                 unblock_netpoll_tx();
1956         } else
1957                 res = -EINVAL;
1958
1959         read_unlock(&bond->lock);
1960
1961         return res;
1962 }
1963
1964 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1965 {
1966         struct bonding *bond = netdev_priv(bond_dev);
1967
1968         info->bond_mode = bond->params.mode;
1969         info->miimon = bond->params.miimon;
1970
1971         read_lock(&bond->lock);
1972         info->num_slaves = bond->slave_cnt;
1973         read_unlock(&bond->lock);
1974
1975         return 0;
1976 }
1977
1978 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1979 {
1980         struct bonding *bond = netdev_priv(bond_dev);
1981         struct list_head *iter;
1982         int i = 0, res = -ENODEV;
1983         struct slave *slave;
1984
1985         read_lock(&bond->lock);
1986         bond_for_each_slave(bond, slave, iter) {
1987                 if (i++ == (int)info->slave_id) {
1988                         res = 0;
1989                         strcpy(info->slave_name, slave->dev->name);
1990                         info->link = slave->link;
1991                         info->state = bond_slave_state(slave);
1992                         info->link_failure_count = slave->link_failure_count;
1993                         break;
1994                 }
1995         }
1996         read_unlock(&bond->lock);
1997
1998         return res;
1999 }
2000
2001 /*-------------------------------- Monitoring -------------------------------*/
2002
2003
2004 static int bond_miimon_inspect(struct bonding *bond)
2005 {
2006         int link_state, commit = 0;
2007         struct list_head *iter;
2008         struct slave *slave;
2009         bool ignore_updelay;
2010
2011         ignore_updelay = !bond->curr_active_slave ? true : false;
2012
2013         bond_for_each_slave(bond, slave, iter) {
2014                 slave->new_link = BOND_LINK_NOCHANGE;
2015
2016                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2017
2018                 switch (slave->link) {
2019                 case BOND_LINK_UP:
2020                         if (link_state)
2021                                 continue;
2022
2023                         slave->link = BOND_LINK_FAIL;
2024                         slave->delay = bond->params.downdelay;
2025                         if (slave->delay) {
2026                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2027                                         bond->dev->name,
2028                                         (bond->params.mode ==
2029                                          BOND_MODE_ACTIVEBACKUP) ?
2030                                         (bond_is_active_slave(slave) ?
2031                                          "active " : "backup ") : "",
2032                                         slave->dev->name,
2033                                         bond->params.downdelay * bond->params.miimon);
2034                         }
2035                         /*FALLTHRU*/
2036                 case BOND_LINK_FAIL:
2037                         if (link_state) {
2038                                 /*
2039                                  * recovered before downdelay expired
2040                                  */
2041                                 slave->link = BOND_LINK_UP;
2042                                 slave->jiffies = jiffies;
2043                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2044                                         bond->dev->name,
2045                                         (bond->params.downdelay - slave->delay) *
2046                                         bond->params.miimon,
2047                                         slave->dev->name);
2048                                 continue;
2049                         }
2050
2051                         if (slave->delay <= 0) {
2052                                 slave->new_link = BOND_LINK_DOWN;
2053                                 commit++;
2054                                 continue;
2055                         }
2056
2057                         slave->delay--;
2058                         break;
2059
2060                 case BOND_LINK_DOWN:
2061                         if (!link_state)
2062                                 continue;
2063
2064                         slave->link = BOND_LINK_BACK;
2065                         slave->delay = bond->params.updelay;
2066
2067                         if (slave->delay) {
2068                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2069                                         bond->dev->name, slave->dev->name,
2070                                         ignore_updelay ? 0 :
2071                                         bond->params.updelay *
2072                                         bond->params.miimon);
2073                         }
2074                         /*FALLTHRU*/
2075                 case BOND_LINK_BACK:
2076                         if (!link_state) {
2077                                 slave->link = BOND_LINK_DOWN;
2078                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2079                                         bond->dev->name,
2080                                         (bond->params.updelay - slave->delay) *
2081                                         bond->params.miimon,
2082                                         slave->dev->name);
2083
2084                                 continue;
2085                         }
2086
2087                         if (ignore_updelay)
2088                                 slave->delay = 0;
2089
2090                         if (slave->delay <= 0) {
2091                                 slave->new_link = BOND_LINK_UP;
2092                                 commit++;
2093                                 ignore_updelay = false;
2094                                 continue;
2095                         }
2096
2097                         slave->delay--;
2098                         break;
2099                 }
2100         }
2101
2102         return commit;
2103 }
2104
2105 static void bond_miimon_commit(struct bonding *bond)
2106 {
2107         struct list_head *iter;
2108         struct slave *slave;
2109
2110         bond_for_each_slave(bond, slave, iter) {
2111                 switch (slave->new_link) {
2112                 case BOND_LINK_NOCHANGE:
2113                         continue;
2114
2115                 case BOND_LINK_UP:
2116                         slave->link = BOND_LINK_UP;
2117                         slave->jiffies = jiffies;
2118
2119                         if (bond->params.mode == BOND_MODE_8023AD) {
2120                                 /* prevent it from being the active one */
2121                                 bond_set_backup_slave(slave);
2122                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2123                                 /* make it immediately active */
2124                                 bond_set_active_slave(slave);
2125                         } else if (slave != bond->primary_slave) {
2126                                 /* prevent it from being the active one */
2127                                 bond_set_backup_slave(slave);
2128                         }
2129
2130                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2131                                 bond->dev->name, slave->dev->name,
2132                                 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2133                                 slave->duplex ? "full" : "half");
2134
2135                         /* notify ad that the link status has changed */
2136                         if (bond->params.mode == BOND_MODE_8023AD)
2137                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2138
2139                         if (bond_is_lb(bond))
2140                                 bond_alb_handle_link_change(bond, slave,
2141                                                             BOND_LINK_UP);
2142
2143                         if (!bond->curr_active_slave ||
2144                             (slave == bond->primary_slave))
2145                                 goto do_failover;
2146
2147                         continue;
2148
2149                 case BOND_LINK_DOWN:
2150                         if (slave->link_failure_count < UINT_MAX)
2151                                 slave->link_failure_count++;
2152
2153                         slave->link = BOND_LINK_DOWN;
2154
2155                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2156                             bond->params.mode == BOND_MODE_8023AD)
2157                                 bond_set_slave_inactive_flags(slave);
2158
2159                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2160                                 bond->dev->name, slave->dev->name);
2161
2162                         if (bond->params.mode == BOND_MODE_8023AD)
2163                                 bond_3ad_handle_link_change(slave,
2164                                                             BOND_LINK_DOWN);
2165
2166                         if (bond_is_lb(bond))
2167                                 bond_alb_handle_link_change(bond, slave,
2168                                                             BOND_LINK_DOWN);
2169
2170                         if (slave == bond->curr_active_slave)
2171                                 goto do_failover;
2172
2173                         continue;
2174
2175                 default:
2176                         pr_err("%s: invalid new link %d on slave %s\n",
2177                                bond->dev->name, slave->new_link,
2178                                slave->dev->name);
2179                         slave->new_link = BOND_LINK_NOCHANGE;
2180
2181                         continue;
2182                 }
2183
2184 do_failover:
2185                 ASSERT_RTNL();
2186                 block_netpoll_tx();
2187                 write_lock_bh(&bond->curr_slave_lock);
2188                 bond_select_active_slave(bond);
2189                 write_unlock_bh(&bond->curr_slave_lock);
2190                 unblock_netpoll_tx();
2191         }
2192
2193         bond_set_carrier(bond);
2194 }
2195
2196 /*
2197  * bond_mii_monitor
2198  *
2199  * Really a wrapper that splits the mii monitor into two phases: an
2200  * inspection, then (if inspection indicates something needs to be done)
2201  * an acquisition of appropriate locks followed by a commit phase to
2202  * implement whatever link state changes are indicated.
2203  */
2204 void bond_mii_monitor(struct work_struct *work)
2205 {
2206         struct bonding *bond = container_of(work, struct bonding,
2207                                             mii_work.work);
2208         bool should_notify_peers = false;
2209         unsigned long delay;
2210
2211         read_lock(&bond->lock);
2212
2213         delay = msecs_to_jiffies(bond->params.miimon);
2214
2215         if (!bond_has_slaves(bond))
2216                 goto re_arm;
2217
2218         should_notify_peers = bond_should_notify_peers(bond);
2219
2220         if (bond_miimon_inspect(bond)) {
2221                 read_unlock(&bond->lock);
2222
2223                 /* Race avoidance with bond_close cancel of workqueue */
2224                 if (!rtnl_trylock()) {
2225                         read_lock(&bond->lock);
2226                         delay = 1;
2227                         should_notify_peers = false;
2228                         goto re_arm;
2229                 }
2230
2231                 read_lock(&bond->lock);
2232
2233                 bond_miimon_commit(bond);
2234
2235                 read_unlock(&bond->lock);
2236                 rtnl_unlock();  /* might sleep, hold no other locks */
2237                 read_lock(&bond->lock);
2238         }
2239
2240 re_arm:
2241         if (bond->params.miimon)
2242                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2243
2244         read_unlock(&bond->lock);
2245
2246         if (should_notify_peers) {
2247                 if (!rtnl_trylock())
2248                         return;
2249                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2250                 rtnl_unlock();
2251         }
2252 }
2253
2254 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2255 {
2256         struct net_device *upper;
2257         struct list_head *iter;
2258         bool ret = false;
2259
2260         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2261                 return true;
2262
2263         rcu_read_lock();
2264         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2265                 if (ip == bond_confirm_addr(upper, 0, ip)) {
2266                         ret = true;
2267                         break;
2268                 }
2269         }
2270         rcu_read_unlock();
2271
2272         return ret;
2273 }
2274
2275 /*
2276  * We go to the (large) trouble of VLAN tagging ARP frames because
2277  * switches in VLAN mode (especially if ports are configured as
2278  * "native" to a VLAN) might not pass non-tagged frames.
2279  */
2280 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2281 {
2282         struct sk_buff *skb;
2283
2284         pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2285                  slave_dev->name, &dest_ip, &src_ip, vlan_id);
2286
2287         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2288                          NULL, slave_dev->dev_addr, NULL);
2289
2290         if (!skb) {
2291                 pr_err("ARP packet allocation failed\n");
2292                 return;
2293         }
2294         if (vlan_id) {
2295                 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2296                 if (!skb) {
2297                         pr_err("failed to insert VLAN tag\n");
2298                         return;
2299                 }
2300         }
2301         arp_xmit(skb);
2302 }
2303
2304
2305 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2306 {
2307         struct net_device *upper, *vlan_upper;
2308         struct list_head *iter, *vlan_iter;
2309         struct rtable *rt;
2310         __be32 *targets = bond->params.arp_targets, addr;
2311         int i, vlan_id;
2312
2313         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2314                 pr_debug("basa: target %pI4\n", &targets[i]);
2315
2316                 /* Find out through which dev should the packet go */
2317                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2318                                      RTO_ONLINK, 0);
2319                 if (IS_ERR(rt)) {
2320                         pr_debug("%s: no route to arp_ip_target %pI4\n",
2321                                  bond->dev->name, &targets[i]);
2322                         continue;
2323                 }
2324
2325                 vlan_id = 0;
2326
2327                 /* bond device itself */
2328                 if (rt->dst.dev == bond->dev)
2329                         goto found;
2330
2331                 rcu_read_lock();
2332                 /* first we search only for vlan devices. for every vlan
2333                  * found we verify its upper dev list, searching for the
2334                  * rt->dst.dev. If found we save the tag of the vlan and
2335                  * proceed to send the packet.
2336                  *
2337                  * TODO: QinQ?
2338                  */
2339                 netdev_for_each_all_upper_dev_rcu(bond->dev, vlan_upper,
2340                                                   vlan_iter) {
2341                         if (!is_vlan_dev(vlan_upper))
2342                                 continue;
2343                         netdev_for_each_all_upper_dev_rcu(vlan_upper, upper,
2344                                                           iter) {
2345                                 if (upper == rt->dst.dev) {
2346                                         vlan_id = vlan_dev_vlan_id(vlan_upper);
2347                                         rcu_read_unlock();
2348                                         goto found;
2349                                 }
2350                         }
2351                 }
2352
2353                 /* if the device we're looking for is not on top of any of
2354                  * our upper vlans, then just search for any dev that
2355                  * matches, and in case it's a vlan - save the id
2356                  */
2357                 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2358                         if (upper == rt->dst.dev) {
2359                                 /* if it's a vlan - get its VID */
2360                                 if (is_vlan_dev(upper))
2361                                         vlan_id = vlan_dev_vlan_id(upper);
2362
2363                                 rcu_read_unlock();
2364                                 goto found;
2365                         }
2366                 }
2367                 rcu_read_unlock();
2368
2369                 /* Not our device - skip */
2370                 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2371                          bond->dev->name, &targets[i],
2372                          rt->dst.dev ? rt->dst.dev->name : "NULL");
2373
2374                 ip_rt_put(rt);
2375                 continue;
2376
2377 found:
2378                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2379                 ip_rt_put(rt);
2380                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2381                               addr, vlan_id);
2382         }
2383 }
2384
2385 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2386 {
2387         int i;
2388
2389         if (!sip || !bond_has_this_ip(bond, tip)) {
2390                 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2391                 return;
2392         }
2393
2394         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2395         if (i == -1) {
2396                 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2397                 return;
2398         }
2399         slave->last_arp_rx = jiffies;
2400         slave->target_last_arp_rx[i] = jiffies;
2401 }
2402
2403 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2404                  struct slave *slave)
2405 {
2406         struct arphdr *arp = (struct arphdr *)skb->data;
2407         unsigned char *arp_ptr;
2408         __be32 sip, tip;
2409         int alen;
2410
2411         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2412                 return RX_HANDLER_ANOTHER;
2413
2414         read_lock(&bond->lock);
2415
2416         if (!slave_do_arp_validate(bond, slave))
2417                 goto out_unlock;
2418
2419         alen = arp_hdr_len(bond->dev);
2420
2421         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2422                  bond->dev->name, skb->dev->name);
2423
2424         if (alen > skb_headlen(skb)) {
2425                 arp = kmalloc(alen, GFP_ATOMIC);
2426                 if (!arp)
2427                         goto out_unlock;
2428                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2429                         goto out_unlock;
2430         }
2431
2432         if (arp->ar_hln != bond->dev->addr_len ||
2433             skb->pkt_type == PACKET_OTHERHOST ||
2434             skb->pkt_type == PACKET_LOOPBACK ||
2435             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2436             arp->ar_pro != htons(ETH_P_IP) ||
2437             arp->ar_pln != 4)
2438                 goto out_unlock;
2439
2440         arp_ptr = (unsigned char *)(arp + 1);
2441         arp_ptr += bond->dev->addr_len;
2442         memcpy(&sip, arp_ptr, 4);
2443         arp_ptr += 4 + bond->dev->addr_len;
2444         memcpy(&tip, arp_ptr, 4);
2445
2446         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2447                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2448                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2449                  &sip, &tip);
2450
2451         /*
2452          * Backup slaves won't see the ARP reply, but do come through
2453          * here for each ARP probe (so we swap the sip/tip to validate
2454          * the probe).  In a "redundant switch, common router" type of
2455          * configuration, the ARP probe will (hopefully) travel from
2456          * the active, through one switch, the router, then the other
2457          * switch before reaching the backup.
2458          *
2459          * We 'trust' the arp requests if there is an active slave and
2460          * it received valid arp reply(s) after it became active. This
2461          * is done to avoid endless looping when we can't reach the
2462          * arp_ip_target and fool ourselves with our own arp requests.
2463          */
2464         if (bond_is_active_slave(slave))
2465                 bond_validate_arp(bond, slave, sip, tip);
2466         else if (bond->curr_active_slave &&
2467                  time_after(slave_last_rx(bond, bond->curr_active_slave),
2468                             bond->curr_active_slave->jiffies))
2469                 bond_validate_arp(bond, slave, tip, sip);
2470
2471 out_unlock:
2472         read_unlock(&bond->lock);
2473         if (arp != (struct arphdr *)skb->data)
2474                 kfree(arp);
2475         return RX_HANDLER_ANOTHER;
2476 }
2477
2478 /* function to verify if we're in the arp_interval timeslice, returns true if
2479  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2480  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2481  */
2482 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2483                                   int mod)
2484 {
2485         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2486
2487         return time_in_range(jiffies,
2488                              last_act - delta_in_ticks,
2489                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2490 }
2491
2492 /*
2493  * this function is called regularly to monitor each slave's link
2494  * ensuring that traffic is being sent and received when arp monitoring
2495  * is used in load-balancing mode. if the adapter has been dormant, then an
2496  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2497  * arp monitoring in active backup mode.
2498  */
2499 void bond_loadbalance_arp_mon(struct work_struct *work)
2500 {
2501         struct bonding *bond = container_of(work, struct bonding,
2502                                             arp_work.work);
2503         struct slave *slave, *oldcurrent;
2504         struct list_head *iter;
2505         int do_failover = 0;
2506
2507         read_lock(&bond->lock);
2508
2509         if (!bond_has_slaves(bond))
2510                 goto re_arm;
2511
2512         oldcurrent = bond->curr_active_slave;
2513         /* see if any of the previous devices are up now (i.e. they have
2514          * xmt and rcv traffic). the curr_active_slave does not come into
2515          * the picture unless it is null. also, slave->jiffies is not needed
2516          * here because we send an arp on each slave and give a slave as
2517          * long as it needs to get the tx/rx within the delta.
2518          * TODO: what about up/down delay in arp mode? it wasn't here before
2519          *       so it can wait
2520          */
2521         bond_for_each_slave(bond, slave, iter) {
2522                 unsigned long trans_start = dev_trans_start(slave->dev);
2523
2524                 if (slave->link != BOND_LINK_UP) {
2525                         if (bond_time_in_interval(bond, trans_start, 1) &&
2526                             bond_time_in_interval(bond, slave->dev->last_rx, 1)) {
2527
2528                                 slave->link  = BOND_LINK_UP;
2529                                 bond_set_active_slave(slave);
2530
2531                                 /* primary_slave has no meaning in round-robin
2532                                  * mode. the window of a slave being up and
2533                                  * curr_active_slave being null after enslaving
2534                                  * is closed.
2535                                  */
2536                                 if (!oldcurrent) {
2537                                         pr_info("%s: link status definitely up for interface %s, ",
2538                                                 bond->dev->name,
2539                                                 slave->dev->name);
2540                                         do_failover = 1;
2541                                 } else {
2542                                         pr_info("%s: interface %s is now up\n",
2543                                                 bond->dev->name,
2544                                                 slave->dev->name);
2545                                 }
2546                         }
2547                 } else {
2548                         /* slave->link == BOND_LINK_UP */
2549
2550                         /* not all switches will respond to an arp request
2551                          * when the source ip is 0, so don't take the link down
2552                          * if we don't know our ip yet
2553                          */
2554                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2555                             !bond_time_in_interval(bond, slave->dev->last_rx, 2)) {
2556
2557                                 slave->link  = BOND_LINK_DOWN;
2558                                 bond_set_backup_slave(slave);
2559
2560                                 if (slave->link_failure_count < UINT_MAX)
2561                                         slave->link_failure_count++;
2562
2563                                 pr_info("%s: interface %s is now down.\n",
2564                                         bond->dev->name,
2565                                         slave->dev->name);
2566
2567                                 if (slave == oldcurrent)
2568                                         do_failover = 1;
2569                         }
2570                 }
2571
2572                 /* note: if switch is in round-robin mode, all links
2573                  * must tx arp to ensure all links rx an arp - otherwise
2574                  * links may oscillate or not come up at all; if switch is
2575                  * in something like xor mode, there is nothing we can
2576                  * do - all replies will be rx'ed on same link causing slaves
2577                  * to be unstable during low/no traffic periods
2578                  */
2579                 if (IS_UP(slave->dev))
2580                         bond_arp_send_all(bond, slave);
2581         }
2582
2583         if (do_failover) {
2584                 block_netpoll_tx();
2585                 write_lock_bh(&bond->curr_slave_lock);
2586
2587                 bond_select_active_slave(bond);
2588
2589                 write_unlock_bh(&bond->curr_slave_lock);
2590                 unblock_netpoll_tx();
2591         }
2592
2593 re_arm:
2594         if (bond->params.arp_interval)
2595                 queue_delayed_work(bond->wq, &bond->arp_work,
2596                                    msecs_to_jiffies(bond->params.arp_interval));
2597
2598         read_unlock(&bond->lock);
2599 }
2600
2601 /*
2602  * Called to inspect slaves for active-backup mode ARP monitor link state
2603  * changes.  Sets new_link in slaves to specify what action should take
2604  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2605  * to link states must be committed.
2606  *
2607  * Called with bond->lock held for read.
2608  */
2609 static int bond_ab_arp_inspect(struct bonding *bond)
2610 {
2611         unsigned long trans_start, last_rx;
2612         struct list_head *iter;
2613         struct slave *slave;
2614         int commit = 0;
2615
2616         bond_for_each_slave(bond, slave, iter) {
2617                 slave->new_link = BOND_LINK_NOCHANGE;
2618                 last_rx = slave_last_rx(bond, slave);
2619
2620                 if (slave->link != BOND_LINK_UP) {
2621                         if (bond_time_in_interval(bond, last_rx, 1)) {
2622                                 slave->new_link = BOND_LINK_UP;
2623                                 commit++;
2624                         }
2625                         continue;
2626                 }
2627
2628                 /*
2629                  * Give slaves 2*delta after being enslaved or made
2630                  * active.  This avoids bouncing, as the last receive
2631                  * times need a full ARP monitor cycle to be updated.
2632                  */
2633                 if (bond_time_in_interval(bond, slave->jiffies, 2))
2634                         continue;
2635
2636                 /*
2637                  * Backup slave is down if:
2638                  * - No current_arp_slave AND
2639                  * - more than 3*delta since last receive AND
2640                  * - the bond has an IP address
2641                  *
2642                  * Note: a non-null current_arp_slave indicates
2643                  * the curr_active_slave went down and we are
2644                  * searching for a new one; under this condition
2645                  * we only take the curr_active_slave down - this
2646                  * gives each slave a chance to tx/rx traffic
2647                  * before being taken out
2648                  */
2649                 if (!bond_is_active_slave(slave) &&
2650                     !bond->current_arp_slave &&
2651                     !bond_time_in_interval(bond, last_rx, 3)) {
2652                         slave->new_link = BOND_LINK_DOWN;
2653                         commit++;
2654                 }
2655
2656                 /*
2657                  * Active slave is down if:
2658                  * - more than 2*delta since transmitting OR
2659                  * - (more than 2*delta since receive AND
2660                  *    the bond has an IP address)
2661                  */
2662                 trans_start = dev_trans_start(slave->dev);
2663                 if (bond_is_active_slave(slave) &&
2664                     (!bond_time_in_interval(bond, trans_start, 2) ||
2665                      !bond_time_in_interval(bond, last_rx, 2))) {
2666                         slave->new_link = BOND_LINK_DOWN;
2667                         commit++;
2668                 }
2669         }
2670
2671         return commit;
2672 }
2673
2674 /*
2675  * Called to commit link state changes noted by inspection step of
2676  * active-backup mode ARP monitor.
2677  *
2678  * Called with RTNL and bond->lock for read.
2679  */
2680 static void bond_ab_arp_commit(struct bonding *bond)
2681 {
2682         unsigned long trans_start;
2683         struct list_head *iter;
2684         struct slave *slave;
2685
2686         bond_for_each_slave(bond, slave, iter) {
2687                 switch (slave->new_link) {
2688                 case BOND_LINK_NOCHANGE:
2689                         continue;
2690
2691                 case BOND_LINK_UP:
2692                         trans_start = dev_trans_start(slave->dev);
2693                         if (bond->curr_active_slave != slave ||
2694                             (!bond->curr_active_slave &&
2695                              bond_time_in_interval(bond, trans_start, 1))) {
2696                                 slave->link = BOND_LINK_UP;
2697                                 if (bond->current_arp_slave) {
2698                                         bond_set_slave_inactive_flags(
2699                                                 bond->current_arp_slave);
2700                                         bond->current_arp_slave = NULL;
2701                                 }
2702
2703                                 pr_info("%s: link status definitely up for interface %s.\n",
2704                                         bond->dev->name, slave->dev->name);
2705
2706                                 if (!bond->curr_active_slave ||
2707                                     (slave == bond->primary_slave))
2708                                         goto do_failover;
2709
2710                         }
2711
2712                         continue;
2713
2714                 case BOND_LINK_DOWN:
2715                         if (slave->link_failure_count < UINT_MAX)
2716                                 slave->link_failure_count++;
2717
2718                         slave->link = BOND_LINK_DOWN;
2719                         bond_set_slave_inactive_flags(slave);
2720
2721                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2722                                 bond->dev->name, slave->dev->name);
2723
2724                         if (slave == bond->curr_active_slave) {
2725                                 bond->current_arp_slave = NULL;
2726                                 goto do_failover;
2727                         }
2728
2729                         continue;
2730
2731                 default:
2732                         pr_err("%s: impossible: new_link %d on slave %s\n",
2733                                bond->dev->name, slave->new_link,
2734                                slave->dev->name);
2735                         continue;
2736                 }
2737
2738 do_failover:
2739                 ASSERT_RTNL();
2740                 block_netpoll_tx();
2741                 write_lock_bh(&bond->curr_slave_lock);
2742                 bond_select_active_slave(bond);
2743                 write_unlock_bh(&bond->curr_slave_lock);
2744                 unblock_netpoll_tx();
2745         }
2746
2747         bond_set_carrier(bond);
2748 }
2749
2750 /*
2751  * Send ARP probes for active-backup mode ARP monitor.
2752  *
2753  * Called with bond->lock held for read.
2754  */
2755 static void bond_ab_arp_probe(struct bonding *bond)
2756 {
2757         struct slave *slave, *before = NULL, *new_slave = NULL;
2758         struct list_head *iter;
2759         bool found = false;
2760
2761         read_lock(&bond->curr_slave_lock);
2762
2763         if (bond->current_arp_slave && bond->curr_active_slave)
2764                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2765                         bond->current_arp_slave->dev->name,
2766                         bond->curr_active_slave->dev->name);
2767
2768         if (bond->curr_active_slave) {
2769                 bond_arp_send_all(bond, bond->curr_active_slave);
2770                 read_unlock(&bond->curr_slave_lock);
2771                 return;
2772         }
2773
2774         read_unlock(&bond->curr_slave_lock);
2775
2776         /* if we don't have a curr_active_slave, search for the next available
2777          * backup slave from the current_arp_slave and make it the candidate
2778          * for becoming the curr_active_slave
2779          */
2780
2781         if (!bond->current_arp_slave) {
2782                 bond->current_arp_slave = bond_first_slave(bond);
2783                 if (!bond->current_arp_slave)
2784                         return;
2785         }
2786
2787         bond_set_slave_inactive_flags(bond->current_arp_slave);
2788
2789         bond_for_each_slave(bond, slave, iter) {
2790                 if (!found && !before && IS_UP(slave->dev))
2791                         before = slave;
2792
2793                 if (found && !new_slave && IS_UP(slave->dev))
2794                         new_slave = slave;
2795                 /* if the link state is up at this point, we
2796                  * mark it down - this can happen if we have
2797                  * simultaneous link failures and
2798                  * reselect_active_interface doesn't make this
2799                  * one the current slave so it is still marked
2800                  * up when it is actually down
2801                  */
2802                 if (!IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
2803                         slave->link = BOND_LINK_DOWN;
2804                         if (slave->link_failure_count < UINT_MAX)
2805                                 slave->link_failure_count++;
2806
2807                         bond_set_slave_inactive_flags(slave);
2808
2809                         pr_info("%s: backup interface %s is now down.\n",
2810                                 bond->dev->name, slave->dev->name);
2811                 }
2812                 if (slave == bond->current_arp_slave)
2813                         found = true;
2814         }
2815
2816         if (!new_slave && before)
2817                 new_slave = before;
2818
2819         if (!new_slave)
2820                 return;
2821
2822         new_slave->link = BOND_LINK_BACK;
2823         bond_set_slave_active_flags(new_slave);
2824         bond_arp_send_all(bond, new_slave);
2825         new_slave->jiffies = jiffies;
2826         bond->current_arp_slave = new_slave;
2827
2828 }
2829
2830 void bond_activebackup_arp_mon(struct work_struct *work)
2831 {
2832         struct bonding *bond = container_of(work, struct bonding,
2833                                             arp_work.work);
2834         bool should_notify_peers = false;
2835         int delta_in_ticks;
2836
2837         read_lock(&bond->lock);
2838
2839         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2840
2841         if (!bond_has_slaves(bond))
2842                 goto re_arm;
2843
2844         should_notify_peers = bond_should_notify_peers(bond);
2845
2846         if (bond_ab_arp_inspect(bond)) {
2847                 read_unlock(&bond->lock);
2848
2849                 /* Race avoidance with bond_close flush of workqueue */
2850                 if (!rtnl_trylock()) {
2851                         read_lock(&bond->lock);
2852                         delta_in_ticks = 1;
2853                         should_notify_peers = false;
2854                         goto re_arm;
2855                 }
2856
2857                 read_lock(&bond->lock);
2858
2859                 bond_ab_arp_commit(bond);
2860
2861                 read_unlock(&bond->lock);
2862                 rtnl_unlock();
2863                 read_lock(&bond->lock);
2864         }
2865
2866         bond_ab_arp_probe(bond);
2867
2868 re_arm:
2869         if (bond->params.arp_interval)
2870                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2871
2872         read_unlock(&bond->lock);
2873
2874         if (should_notify_peers) {
2875                 if (!rtnl_trylock())
2876                         return;
2877                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2878                 rtnl_unlock();
2879         }
2880 }
2881
2882 /*-------------------------- netdev event handling --------------------------*/
2883
2884 /*
2885  * Change device name
2886  */
2887 static int bond_event_changename(struct bonding *bond)
2888 {
2889         bond_remove_proc_entry(bond);
2890         bond_create_proc_entry(bond);
2891
2892         bond_debug_reregister(bond);
2893
2894         return NOTIFY_DONE;
2895 }
2896
2897 static int bond_master_netdev_event(unsigned long event,
2898                                     struct net_device *bond_dev)
2899 {
2900         struct bonding *event_bond = netdev_priv(bond_dev);
2901
2902         switch (event) {
2903         case NETDEV_CHANGENAME:
2904                 return bond_event_changename(event_bond);
2905         case NETDEV_UNREGISTER:
2906                 bond_remove_proc_entry(event_bond);
2907                 break;
2908         case NETDEV_REGISTER:
2909                 bond_create_proc_entry(event_bond);
2910                 break;
2911         case NETDEV_NOTIFY_PEERS:
2912                 if (event_bond->send_peer_notif)
2913                         event_bond->send_peer_notif--;
2914                 break;
2915         default:
2916                 break;
2917         }
2918
2919         return NOTIFY_DONE;
2920 }
2921
2922 static int bond_slave_netdev_event(unsigned long event,
2923                                    struct net_device *slave_dev)
2924 {
2925         struct slave *slave = bond_slave_get_rtnl(slave_dev);
2926         struct bonding *bond;
2927         struct net_device *bond_dev;
2928         u32 old_speed;
2929         u8 old_duplex;
2930
2931         /* A netdev event can be generated while enslaving a device
2932          * before netdev_rx_handler_register is called in which case
2933          * slave will be NULL
2934          */
2935         if (!slave)
2936                 return NOTIFY_DONE;
2937         bond_dev = slave->bond->dev;
2938         bond = slave->bond;
2939
2940         switch (event) {
2941         case NETDEV_UNREGISTER:
2942                 if (bond_dev->type != ARPHRD_ETHER)
2943                         bond_release_and_destroy(bond_dev, slave_dev);
2944                 else
2945                         bond_release(bond_dev, slave_dev);
2946                 break;
2947         case NETDEV_UP:
2948         case NETDEV_CHANGE:
2949                 old_speed = slave->speed;
2950                 old_duplex = slave->duplex;
2951
2952                 bond_update_speed_duplex(slave);
2953
2954                 if (bond->params.mode == BOND_MODE_8023AD) {
2955                         if (old_speed != slave->speed)
2956                                 bond_3ad_adapter_speed_changed(slave);
2957                         if (old_duplex != slave->duplex)
2958                                 bond_3ad_adapter_duplex_changed(slave);
2959                 }
2960                 break;
2961         case NETDEV_DOWN:
2962                 /*
2963                  * ... Or is it this?
2964                  */
2965                 break;
2966         case NETDEV_CHANGEMTU:
2967                 /*
2968                  * TODO: Should slaves be allowed to
2969                  * independently alter their MTU?  For
2970                  * an active-backup bond, slaves need
2971                  * not be the same type of device, so
2972                  * MTUs may vary.  For other modes,
2973                  * slaves arguably should have the
2974                  * same MTUs. To do this, we'd need to
2975                  * take over the slave's change_mtu
2976                  * function for the duration of their
2977                  * servitude.
2978                  */
2979                 break;
2980         case NETDEV_CHANGENAME:
2981                 /*
2982                  * TODO: handle changing the primary's name
2983                  */
2984                 break;
2985         case NETDEV_FEAT_CHANGE:
2986                 bond_compute_features(bond);
2987                 break;
2988         case NETDEV_RESEND_IGMP:
2989                 /* Propagate to master device */
2990                 call_netdevice_notifiers(event, slave->bond->dev);
2991                 break;
2992         default:
2993                 break;
2994         }
2995
2996         return NOTIFY_DONE;
2997 }
2998
2999 /*
3000  * bond_netdev_event: handle netdev notifier chain events.
3001  *
3002  * This function receives events for the netdev chain.  The caller (an
3003  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3004  * locks for us to safely manipulate the slave devices (RTNL lock,
3005  * dev_probe_lock).
3006  */
3007 static int bond_netdev_event(struct notifier_block *this,
3008                              unsigned long event, void *ptr)
3009 {
3010         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3011
3012         pr_debug("event_dev: %s, event: %lx\n",
3013                  event_dev ? event_dev->name : "None",
3014                  event);
3015
3016         if (!(event_dev->priv_flags & IFF_BONDING))
3017                 return NOTIFY_DONE;
3018
3019         if (event_dev->flags & IFF_MASTER) {
3020                 pr_debug("IFF_MASTER\n");
3021                 return bond_master_netdev_event(event, event_dev);
3022         }
3023
3024         if (event_dev->flags & IFF_SLAVE) {
3025                 pr_debug("IFF_SLAVE\n");
3026                 return bond_slave_netdev_event(event, event_dev);
3027         }
3028
3029         return NOTIFY_DONE;
3030 }
3031
3032 static struct notifier_block bond_netdev_notifier = {
3033         .notifier_call = bond_netdev_event,
3034 };
3035
3036 /*---------------------------- Hashing Policies -----------------------------*/
3037
3038 /*
3039  * Hash for the output device based upon layer 2 data
3040  */
3041 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3042 {
3043         struct ethhdr *data = (struct ethhdr *)skb->data;
3044
3045         if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3046                 return (data->h_dest[5] ^ data->h_source[5]) % count;
3047
3048         return 0;
3049 }
3050
3051 /*
3052  * Hash for the output device based upon layer 2 and layer 3 data. If
3053  * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3054  */
3055 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3056 {
3057         const struct ethhdr *data;
3058         const struct iphdr *iph;
3059         const struct ipv6hdr *ipv6h;
3060         u32 v6hash;
3061         const __be32 *s, *d;
3062
3063         if (skb->protocol == htons(ETH_P_IP) &&
3064             pskb_network_may_pull(skb, sizeof(*iph))) {
3065                 iph = ip_hdr(skb);
3066                 data = (struct ethhdr *)skb->data;
3067                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3068                         (data->h_dest[5] ^ data->h_source[5])) % count;
3069         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3070                    pskb_network_may_pull(skb, sizeof(*ipv6h))) {
3071                 ipv6h = ipv6_hdr(skb);
3072                 data = (struct ethhdr *)skb->data;
3073                 s = &ipv6h->saddr.s6_addr32[0];
3074                 d = &ipv6h->daddr.s6_addr32[0];
3075                 v6hash = (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3076                 v6hash ^= (v6hash >> 24) ^ (v6hash >> 16) ^ (v6hash >> 8);
3077                 return (v6hash ^ data->h_dest[5] ^ data->h_source[5]) % count;
3078         }
3079
3080         return bond_xmit_hash_policy_l2(skb, count);
3081 }
3082
3083 /*
3084  * Hash for the output device based upon layer 3 and layer 4 data. If
3085  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3086  * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3087  */
3088 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3089 {
3090         u32 layer4_xor = 0;
3091         const struct iphdr *iph;
3092         const struct ipv6hdr *ipv6h;
3093         const __be32 *s, *d;
3094         const __be16 *l4 = NULL;
3095         __be16 _l4[2];
3096         int noff = skb_network_offset(skb);
3097         int poff;
3098
3099         if (skb->protocol == htons(ETH_P_IP) &&
3100             pskb_may_pull(skb, noff + sizeof(*iph))) {
3101                 iph = ip_hdr(skb);
3102                 poff = proto_ports_offset(iph->protocol);
3103
3104                 if (!ip_is_fragment(iph) && poff >= 0) {
3105                         l4 = skb_header_pointer(skb, noff + (iph->ihl << 2) + poff,
3106                                                 sizeof(_l4), &_l4);
3107                         if (l4)
3108                                 layer4_xor = ntohs(l4[0] ^ l4[1]);
3109                 }
3110                 return (layer4_xor ^
3111                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3112         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3113                    pskb_may_pull(skb, noff + sizeof(*ipv6h))) {
3114                 ipv6h = ipv6_hdr(skb);
3115                 poff = proto_ports_offset(ipv6h->nexthdr);
3116                 if (poff >= 0) {
3117                         l4 = skb_header_pointer(skb, noff + sizeof(*ipv6h) + poff,
3118                                                 sizeof(_l4), &_l4);
3119                         if (l4)
3120                                 layer4_xor = ntohs(l4[0] ^ l4[1]);
3121                 }
3122                 s = &ipv6h->saddr.s6_addr32[0];
3123                 d = &ipv6h->daddr.s6_addr32[0];
3124                 layer4_xor ^= (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3125                 layer4_xor ^= (layer4_xor >> 24) ^ (layer4_xor >> 16) ^
3126                                (layer4_xor >> 8);
3127                 return layer4_xor % count;
3128         }
3129
3130         return bond_xmit_hash_policy_l2(skb, count);
3131 }
3132
3133 /*-------------------------- Device entry points ----------------------------*/
3134
3135 static void bond_work_init_all(struct bonding *bond)
3136 {
3137         INIT_DELAYED_WORK(&bond->mcast_work,
3138                           bond_resend_igmp_join_requests_delayed);
3139         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3140         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3141         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3142                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3143         else
3144                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3145         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3146 }
3147
3148 static void bond_work_cancel_all(struct bonding *bond)
3149 {
3150         cancel_delayed_work_sync(&bond->mii_work);
3151         cancel_delayed_work_sync(&bond->arp_work);
3152         cancel_delayed_work_sync(&bond->alb_work);
3153         cancel_delayed_work_sync(&bond->ad_work);
3154         cancel_delayed_work_sync(&bond->mcast_work);
3155 }
3156
3157 static int bond_open(struct net_device *bond_dev)
3158 {
3159         struct bonding *bond = netdev_priv(bond_dev);
3160         struct list_head *iter;
3161         struct slave *slave;
3162
3163         /* reset slave->backup and slave->inactive */
3164         read_lock(&bond->lock);
3165         if (bond_has_slaves(bond)) {
3166                 read_lock(&bond->curr_slave_lock);
3167                 bond_for_each_slave(bond, slave, iter) {
3168                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3169                                 && (slave != bond->curr_active_slave)) {
3170                                 bond_set_slave_inactive_flags(slave);
3171                         } else {
3172                                 bond_set_slave_active_flags(slave);
3173                         }
3174                 }
3175                 read_unlock(&bond->curr_slave_lock);
3176         }
3177         read_unlock(&bond->lock);
3178
3179         bond_work_init_all(bond);
3180
3181         if (bond_is_lb(bond)) {
3182                 /* bond_alb_initialize must be called before the timer
3183                  * is started.
3184                  */
3185                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3186                         return -ENOMEM;
3187                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3188         }
3189
3190         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3191                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3192
3193         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3194                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3195                 if (bond->params.arp_validate)
3196                         bond->recv_probe = bond_arp_rcv;
3197         }
3198
3199         if (bond->params.mode == BOND_MODE_8023AD) {
3200                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3201                 /* register to receive LACPDUs */
3202                 bond->recv_probe = bond_3ad_lacpdu_recv;
3203                 bond_3ad_initiate_agg_selection(bond, 1);
3204         }
3205
3206         return 0;
3207 }
3208
3209 static int bond_close(struct net_device *bond_dev)
3210 {
3211         struct bonding *bond = netdev_priv(bond_dev);
3212
3213         bond_work_cancel_all(bond);
3214         bond->send_peer_notif = 0;
3215         if (bond_is_lb(bond))
3216                 bond_alb_deinitialize(bond);
3217         bond->recv_probe = NULL;
3218
3219         return 0;
3220 }
3221
3222 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3223                                                 struct rtnl_link_stats64 *stats)
3224 {
3225         struct bonding *bond = netdev_priv(bond_dev);
3226         struct rtnl_link_stats64 temp;
3227         struct list_head *iter;
3228         struct slave *slave;
3229
3230         memset(stats, 0, sizeof(*stats));
3231
3232         read_lock_bh(&bond->lock);
3233         bond_for_each_slave(bond, slave, iter) {
3234                 const struct rtnl_link_stats64 *sstats =
3235                         dev_get_stats(slave->dev, &temp);
3236
3237                 stats->rx_packets += sstats->rx_packets;
3238                 stats->rx_bytes += sstats->rx_bytes;
3239                 stats->rx_errors += sstats->rx_errors;
3240                 stats->rx_dropped += sstats->rx_dropped;
3241
3242                 stats->tx_packets += sstats->tx_packets;
3243                 stats->tx_bytes += sstats->tx_bytes;
3244                 stats->tx_errors += sstats->tx_errors;
3245                 stats->tx_dropped += sstats->tx_dropped;
3246
3247                 stats->multicast += sstats->multicast;
3248                 stats->collisions += sstats->collisions;
3249
3250                 stats->rx_length_errors += sstats->rx_length_errors;
3251                 stats->rx_over_errors += sstats->rx_over_errors;
3252                 stats->rx_crc_errors += sstats->rx_crc_errors;
3253                 stats->rx_frame_errors += sstats->rx_frame_errors;
3254                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3255                 stats->rx_missed_errors += sstats->rx_missed_errors;
3256
3257                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3258                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3259                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3260                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3261                 stats->tx_window_errors += sstats->tx_window_errors;
3262         }
3263         read_unlock_bh(&bond->lock);
3264
3265         return stats;
3266 }
3267
3268 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3269 {
3270         struct net_device *slave_dev = NULL;
3271         struct ifbond k_binfo;
3272         struct ifbond __user *u_binfo = NULL;
3273         struct ifslave k_sinfo;
3274         struct ifslave __user *u_sinfo = NULL;
3275         struct mii_ioctl_data *mii = NULL;
3276         struct net *net;
3277         int res = 0;
3278
3279         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3280
3281         switch (cmd) {
3282         case SIOCGMIIPHY:
3283                 mii = if_mii(ifr);
3284                 if (!mii)
3285                         return -EINVAL;
3286
3287                 mii->phy_id = 0;
3288                 /* Fall Through */
3289         case SIOCGMIIREG:
3290                 /*
3291                  * We do this again just in case we were called by SIOCGMIIREG
3292                  * instead of SIOCGMIIPHY.
3293                  */
3294                 mii = if_mii(ifr);
3295                 if (!mii)
3296                         return -EINVAL;
3297
3298
3299                 if (mii->reg_num == 1) {
3300                         struct bonding *bond = netdev_priv(bond_dev);
3301                         mii->val_out = 0;
3302                         read_lock(&bond->lock);
3303                         read_lock(&bond->curr_slave_lock);
3304                         if (netif_carrier_ok(bond->dev))
3305                                 mii->val_out = BMSR_LSTATUS;
3306
3307                         read_unlock(&bond->curr_slave_lock);
3308                         read_unlock(&bond->lock);
3309                 }
3310
3311                 return 0;
3312         case BOND_INFO_QUERY_OLD:
3313         case SIOCBONDINFOQUERY:
3314                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3315
3316                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3317                         return -EFAULT;
3318
3319                 res = bond_info_query(bond_dev, &k_binfo);
3320                 if (res == 0 &&
3321                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3322                         return -EFAULT;
3323
3324                 return res;
3325         case BOND_SLAVE_INFO_QUERY_OLD:
3326         case SIOCBONDSLAVEINFOQUERY:
3327                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3328
3329                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3330                         return -EFAULT;
3331
3332                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3333                 if (res == 0 &&
3334                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3335                         return -EFAULT;
3336
3337                 return res;
3338         default:
3339                 /* Go on */
3340                 break;
3341         }
3342
3343         net = dev_net(bond_dev);
3344
3345         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3346                 return -EPERM;
3347
3348         slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3349
3350         pr_debug("slave_dev=%p:\n", slave_dev);
3351
3352         if (!slave_dev)
3353                 res = -ENODEV;
3354         else {
3355                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3356                 switch (cmd) {
3357                 case BOND_ENSLAVE_OLD:
3358                 case SIOCBONDENSLAVE:
3359                         res = bond_enslave(bond_dev, slave_dev);
3360                         break;
3361                 case BOND_RELEASE_OLD:
3362                 case SIOCBONDRELEASE:
3363                         res = bond_release(bond_dev, slave_dev);
3364                         break;
3365                 case BOND_SETHWADDR_OLD:
3366                 case SIOCBONDSETHWADDR:
3367                         bond_set_dev_addr(bond_dev, slave_dev);
3368                         res = 0;
3369                         break;
3370                 case BOND_CHANGE_ACTIVE_OLD:
3371                 case SIOCBONDCHANGEACTIVE:
3372                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3373                         break;
3374                 default:
3375                         res = -EOPNOTSUPP;
3376                 }
3377
3378                 dev_put(slave_dev);
3379         }
3380
3381         return res;
3382 }
3383
3384 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3385 {
3386         struct bonding *bond = netdev_priv(bond_dev);
3387
3388         if (change & IFF_PROMISC)
3389                 bond_set_promiscuity(bond,
3390                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3391
3392         if (change & IFF_ALLMULTI)
3393                 bond_set_allmulti(bond,
3394                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3395 }
3396
3397 static void bond_set_rx_mode(struct net_device *bond_dev)
3398 {
3399         struct bonding *bond = netdev_priv(bond_dev);
3400         struct list_head *iter;
3401         struct slave *slave;
3402
3403
3404         rcu_read_lock();
3405         if (USES_PRIMARY(bond->params.mode)) {
3406                 slave = rcu_dereference(bond->curr_active_slave);
3407                 if (slave) {
3408                         dev_uc_sync(slave->dev, bond_dev);
3409                         dev_mc_sync(slave->dev, bond_dev);
3410                 }
3411         } else {
3412                 bond_for_each_slave_rcu(bond, slave, iter) {
3413                         dev_uc_sync_multiple(slave->dev, bond_dev);
3414                         dev_mc_sync_multiple(slave->dev, bond_dev);
3415                 }
3416         }
3417         rcu_read_unlock();
3418 }
3419
3420 static int bond_neigh_init(struct neighbour *n)
3421 {
3422         struct bonding *bond = netdev_priv(n->dev);
3423         const struct net_device_ops *slave_ops;
3424         struct neigh_parms parms;
3425         struct slave *slave;
3426         int ret;
3427
3428         slave = bond_first_slave(bond);
3429         if (!slave)
3430                 return 0;
3431         slave_ops = slave->dev->netdev_ops;
3432         if (!slave_ops->ndo_neigh_setup)
3433                 return 0;
3434
3435         parms.neigh_setup = NULL;
3436         parms.neigh_cleanup = NULL;
3437         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3438         if (ret)
3439                 return ret;
3440
3441         /*
3442          * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3443          * after the last slave has been detached.  Assumes that all slaves
3444          * utilize the same neigh_cleanup (true at this writing as only user
3445          * is ipoib).
3446          */
3447         n->parms->neigh_cleanup = parms.neigh_cleanup;
3448
3449         if (!parms.neigh_setup)
3450                 return 0;
3451
3452         return parms.neigh_setup(n);
3453 }
3454
3455 /*
3456  * The bonding ndo_neigh_setup is called at init time beofre any
3457  * slave exists. So we must declare proxy setup function which will
3458  * be used at run time to resolve the actual slave neigh param setup.
3459  *
3460  * It's also called by master devices (such as vlans) to setup their
3461  * underlying devices. In that case - do nothing, we're already set up from
3462  * our init.
3463  */
3464 static int bond_neigh_setup(struct net_device *dev,
3465                             struct neigh_parms *parms)
3466 {
3467         /* modify only our neigh_parms */
3468         if (parms->dev == dev)
3469                 parms->neigh_setup = bond_neigh_init;
3470
3471         return 0;
3472 }
3473
3474 /*
3475  * Change the MTU of all of a master's slaves to match the master
3476  */
3477 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3478 {
3479         struct bonding *bond = netdev_priv(bond_dev);
3480         struct slave *slave, *rollback_slave;
3481         struct list_head *iter;
3482         int res = 0;
3483
3484         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3485                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3486
3487         /* Can't hold bond->lock with bh disabled here since
3488          * some base drivers panic. On the other hand we can't
3489          * hold bond->lock without bh disabled because we'll
3490          * deadlock. The only solution is to rely on the fact
3491          * that we're under rtnl_lock here, and the slaves
3492          * list won't change. This doesn't solve the problem
3493          * of setting the slave's MTU while it is
3494          * transmitting, but the assumption is that the base
3495          * driver can handle that.
3496          *
3497          * TODO: figure out a way to safely iterate the slaves
3498          * list, but without holding a lock around the actual
3499          * call to the base driver.
3500          */
3501
3502         bond_for_each_slave(bond, slave, iter) {
3503                 pr_debug("s %p c_m %p\n",
3504                          slave,
3505                          slave->dev->netdev_ops->ndo_change_mtu);
3506
3507                 res = dev_set_mtu(slave->dev, new_mtu);
3508
3509                 if (res) {
3510                         /* If we failed to set the slave's mtu to the new value
3511                          * we must abort the operation even in ACTIVE_BACKUP
3512                          * mode, because if we allow the backup slaves to have
3513                          * different mtu values than the active slave we'll
3514                          * need to change their mtu when doing a failover. That
3515                          * means changing their mtu from timer context, which
3516                          * is probably not a good idea.
3517                          */
3518                         pr_debug("err %d %s\n", res, slave->dev->name);
3519                         goto unwind;
3520                 }
3521         }
3522
3523         bond_dev->mtu = new_mtu;
3524
3525         return 0;
3526
3527 unwind:
3528         /* unwind from head to the slave that failed */
3529         bond_for_each_slave(bond, rollback_slave, iter) {
3530                 int tmp_res;
3531
3532                 if (rollback_slave == slave)
3533                         break;
3534
3535                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3536                 if (tmp_res) {
3537                         pr_debug("unwind err %d dev %s\n",
3538                                  tmp_res, rollback_slave->dev->name);
3539                 }
3540         }
3541
3542         return res;
3543 }
3544
3545 /*
3546  * Change HW address
3547  *
3548  * Note that many devices must be down to change the HW address, and
3549  * downing the master releases all slaves.  We can make bonds full of
3550  * bonding devices to test this, however.
3551  */
3552 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3553 {
3554         struct bonding *bond = netdev_priv(bond_dev);
3555         struct slave *slave, *rollback_slave;
3556         struct sockaddr *sa = addr, tmp_sa;
3557         struct list_head *iter;
3558         int res = 0;
3559
3560         if (bond->params.mode == BOND_MODE_ALB)
3561                 return bond_alb_set_mac_address(bond_dev, addr);
3562
3563
3564         pr_debug("bond=%p, name=%s\n",
3565                  bond, bond_dev ? bond_dev->name : "None");
3566
3567         /* If fail_over_mac is enabled, do nothing and return success.
3568          * Returning an error causes ifenslave to fail.
3569          */
3570         if (bond->params.fail_over_mac)
3571                 return 0;
3572
3573         if (!is_valid_ether_addr(sa->sa_data))
3574                 return -EADDRNOTAVAIL;
3575
3576         /* Can't hold bond->lock with bh disabled here since
3577          * some base drivers panic. On the other hand we can't
3578          * hold bond->lock without bh disabled because we'll
3579          * deadlock. The only solution is to rely on the fact
3580          * that we're under rtnl_lock here, and the slaves
3581          * list won't change. This doesn't solve the problem
3582          * of setting the slave's hw address while it is
3583          * transmitting, but the assumption is that the base
3584          * driver can handle that.
3585          *
3586          * TODO: figure out a way to safely iterate the slaves
3587          * list, but without holding a lock around the actual
3588          * call to the base driver.
3589          */
3590
3591         bond_for_each_slave(bond, slave, iter) {
3592                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3593                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3594
3595                 if (slave_ops->ndo_set_mac_address == NULL) {
3596                         res = -EOPNOTSUPP;
3597                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3598                         goto unwind;
3599                 }
3600
3601                 res = dev_set_mac_address(slave->dev, addr);
3602                 if (res) {
3603                         /* TODO: consider downing the slave
3604                          * and retry ?
3605                          * User should expect communications
3606                          * breakage anyway until ARP finish
3607                          * updating, so...
3608                          */
3609                         pr_debug("err %d %s\n", res, slave->dev->name);
3610                         goto unwind;
3611                 }
3612         }
3613
3614         /* success */
3615         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3616         return 0;
3617
3618 unwind:
3619         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3620         tmp_sa.sa_family = bond_dev->type;
3621
3622         /* unwind from head to the slave that failed */
3623         bond_for_each_slave(bond, rollback_slave, iter) {
3624                 int tmp_res;
3625
3626                 if (rollback_slave == slave)
3627                         break;
3628
3629                 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3630                 if (tmp_res) {
3631                         pr_debug("unwind err %d dev %s\n",
3632                                  tmp_res, rollback_slave->dev->name);
3633                 }
3634         }
3635
3636         return res;
3637 }
3638
3639 /**
3640  * bond_xmit_slave_id - transmit skb through slave with slave_id
3641  * @bond: bonding device that is transmitting
3642  * @skb: buffer to transmit
3643  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3644  *
3645  * This function tries to transmit through slave with slave_id but in case
3646  * it fails, it tries to find the first available slave for transmission.
3647  * The skb is consumed in all cases, thus the function is void.
3648  */
3649 void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3650 {
3651         struct list_head *iter;
3652         struct slave *slave;
3653         int i = slave_id;
3654
3655         /* Here we start from the slave with slave_id */
3656         bond_for_each_slave_rcu(bond, slave, iter) {
3657                 if (--i < 0) {
3658                         if (slave_can_tx(slave)) {
3659                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3660                                 return;
3661                         }
3662                 }
3663         }
3664
3665         /* Here we start from the first slave up to slave_id */
3666         i = slave_id;
3667         bond_for_each_slave_rcu(bond, slave, iter) {
3668                 if (--i < 0)
3669                         break;
3670                 if (slave_can_tx(slave)) {
3671                         bond_dev_queue_xmit(bond, skb, slave->dev);
3672                         return;
3673                 }
3674         }
3675         /* no slave that can tx has been found */
3676         kfree_skb(skb);
3677 }
3678
3679 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3680 {
3681         struct bonding *bond = netdev_priv(bond_dev);
3682         struct iphdr *iph = ip_hdr(skb);
3683         struct slave *slave;
3684
3685         /*
3686          * Start with the curr_active_slave that joined the bond as the
3687          * default for sending IGMP traffic.  For failover purposes one
3688          * needs to maintain some consistency for the interface that will
3689          * send the join/membership reports.  The curr_active_slave found
3690          * will send all of this type of traffic.
3691          */
3692         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3693                 slave = rcu_dereference(bond->curr_active_slave);
3694                 if (slave && slave_can_tx(slave))
3695                         bond_dev_queue_xmit(bond, skb, slave->dev);
3696                 else
3697                         bond_xmit_slave_id(bond, skb, 0);
3698         } else {
3699                 bond_xmit_slave_id(bond, skb,
3700                                    bond->rr_tx_counter++ % bond->slave_cnt);
3701         }
3702
3703         return NETDEV_TX_OK;
3704 }
3705
3706 /*
3707  * in active-backup mode, we know that bond->curr_active_slave is always valid if
3708  * the bond has a usable interface.
3709  */
3710 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3711 {
3712         struct bonding *bond = netdev_priv(bond_dev);
3713         struct slave *slave;
3714
3715         slave = rcu_dereference(bond->curr_active_slave);
3716         if (slave)
3717                 bond_dev_queue_xmit(bond, skb, slave->dev);
3718         else
3719                 kfree_skb(skb);
3720
3721         return NETDEV_TX_OK;
3722 }
3723
3724 /*
3725  * In bond_xmit_xor() , we determine the output device by using a pre-
3726  * determined xmit_hash_policy(), If the selected device is not enabled,
3727  * find the next active slave.
3728  */
3729 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3730 {
3731         struct bonding *bond = netdev_priv(bond_dev);
3732
3733         bond_xmit_slave_id(bond, skb,
3734                            bond->xmit_hash_policy(skb, bond->slave_cnt));
3735
3736         return NETDEV_TX_OK;
3737 }
3738
3739 /* in broadcast mode, we send everything to all usable interfaces. */
3740 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3741 {
3742         struct bonding *bond = netdev_priv(bond_dev);
3743         struct slave *slave = NULL;
3744         struct list_head *iter;
3745
3746         bond_for_each_slave_rcu(bond, slave, iter) {
3747                 if (bond_is_last_slave(bond, slave))
3748                         break;
3749                 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3750                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3751
3752                         if (!skb2) {
3753                                 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3754                                        bond_dev->name);
3755                                 continue;
3756                         }
3757                         /* bond_dev_queue_xmit always returns 0 */
3758                         bond_dev_queue_xmit(bond, skb2, slave->dev);
3759                 }
3760         }
3761         if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3762                 bond_dev_queue_xmit(bond, skb, slave->dev);
3763         else
3764                 kfree_skb(skb);
3765
3766         return NETDEV_TX_OK;
3767 }
3768
3769 /*------------------------- Device initialization ---------------------------*/
3770
3771 static void bond_set_xmit_hash_policy(struct bonding *bond)
3772 {
3773         switch (bond->params.xmit_policy) {
3774         case BOND_XMIT_POLICY_LAYER23:
3775                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
3776                 break;
3777         case BOND_XMIT_POLICY_LAYER34:
3778                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
3779                 break;
3780         case BOND_XMIT_POLICY_LAYER2:
3781         default:
3782                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
3783                 break;
3784         }
3785 }
3786
3787 /*
3788  * Lookup the slave that corresponds to a qid
3789  */
3790 static inline int bond_slave_override(struct bonding *bond,
3791                                       struct sk_buff *skb)
3792 {
3793         struct slave *slave = NULL;
3794         struct slave *check_slave;
3795         struct list_head *iter;
3796         int res = 1;
3797
3798         if (!skb->queue_mapping)
3799                 return 1;
3800
3801         /* Find out if any slaves have the same mapping as this skb. */
3802         bond_for_each_slave_rcu(bond, check_slave, iter) {
3803                 if (check_slave->queue_id == skb->queue_mapping) {
3804                         slave = check_slave;
3805                         break;
3806                 }
3807         }
3808
3809         /* If the slave isn't UP, use default transmit policy. */
3810         if (slave && slave->queue_id && IS_UP(slave->dev) &&
3811             (slave->link == BOND_LINK_UP)) {
3812                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3813         }
3814
3815         return res;
3816 }
3817
3818
3819 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
3820 {
3821         /*
3822          * This helper function exists to help dev_pick_tx get the correct
3823          * destination queue.  Using a helper function skips a call to
3824          * skb_tx_hash and will put the skbs in the queue we expect on their
3825          * way down to the bonding driver.
3826          */
3827         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3828
3829         /*
3830          * Save the original txq to restore before passing to the driver
3831          */
3832         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3833
3834         if (unlikely(txq >= dev->real_num_tx_queues)) {
3835                 do {
3836                         txq -= dev->real_num_tx_queues;
3837                 } while (txq >= dev->real_num_tx_queues);
3838         }
3839         return txq;
3840 }
3841
3842 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3843 {
3844         struct bonding *bond = netdev_priv(dev);
3845
3846         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
3847                 if (!bond_slave_override(bond, skb))
3848                         return NETDEV_TX_OK;
3849         }
3850
3851         switch (bond->params.mode) {
3852         case BOND_MODE_ROUNDROBIN:
3853                 return bond_xmit_roundrobin(skb, dev);
3854         case BOND_MODE_ACTIVEBACKUP:
3855                 return bond_xmit_activebackup(skb, dev);
3856         case BOND_MODE_XOR:
3857                 return bond_xmit_xor(skb, dev);
3858         case BOND_MODE_BROADCAST:
3859                 return bond_xmit_broadcast(skb, dev);
3860         case BOND_MODE_8023AD:
3861                 return bond_3ad_xmit_xor(skb, dev);
3862         case BOND_MODE_ALB:
3863         case BOND_MODE_TLB:
3864                 return bond_alb_xmit(skb, dev);
3865         default:
3866                 /* Should never happen, mode already checked */
3867                 pr_err("%s: Error: Unknown bonding mode %d\n",
3868                        dev->name, bond->params.mode);
3869                 WARN_ON_ONCE(1);
3870                 kfree_skb(skb);
3871                 return NETDEV_TX_OK;
3872         }
3873 }
3874
3875 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3876 {
3877         struct bonding *bond = netdev_priv(dev);
3878         netdev_tx_t ret = NETDEV_TX_OK;
3879
3880         /*
3881          * If we risk deadlock from transmitting this in the
3882          * netpoll path, tell netpoll to queue the frame for later tx
3883          */
3884         if (is_netpoll_tx_blocked(dev))
3885                 return NETDEV_TX_BUSY;
3886
3887         rcu_read_lock();
3888         if (bond_has_slaves(bond))
3889                 ret = __bond_start_xmit(skb, dev);
3890         else
3891                 kfree_skb(skb);
3892         rcu_read_unlock();
3893
3894         return ret;
3895 }
3896
3897 /*
3898  * set bond mode specific net device operations
3899  */
3900 void bond_set_mode_ops(struct bonding *bond, int mode)
3901 {
3902         struct net_device *bond_dev = bond->dev;
3903
3904         switch (mode) {
3905         case BOND_MODE_ROUNDROBIN:
3906                 break;
3907         case BOND_MODE_ACTIVEBACKUP:
3908                 break;
3909         case BOND_MODE_XOR:
3910                 bond_set_xmit_hash_policy(bond);
3911                 break;
3912         case BOND_MODE_BROADCAST:
3913                 break;
3914         case BOND_MODE_8023AD:
3915                 bond_set_xmit_hash_policy(bond);
3916                 break;
3917         case BOND_MODE_ALB:
3918                 /* FALLTHRU */
3919         case BOND_MODE_TLB:
3920                 break;
3921         default:
3922                 /* Should never happen, mode already checked */
3923                 pr_err("%s: Error: Unknown bonding mode %d\n",
3924                        bond_dev->name, mode);
3925                 break;
3926         }
3927 }
3928
3929 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3930                                      struct ethtool_cmd *ecmd)
3931 {
3932         struct bonding *bond = netdev_priv(bond_dev);
3933         unsigned long speed = 0;
3934         struct list_head *iter;
3935         struct slave *slave;
3936
3937         ecmd->duplex = DUPLEX_UNKNOWN;
3938         ecmd->port = PORT_OTHER;
3939
3940         /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3941          * do not need to check mode.  Though link speed might not represent
3942          * the true receive or transmit bandwidth (not all modes are symmetric)
3943          * this is an accurate maximum.
3944          */
3945         read_lock(&bond->lock);
3946         bond_for_each_slave(bond, slave, iter) {
3947                 if (SLAVE_IS_OK(slave)) {
3948                         if (slave->speed != SPEED_UNKNOWN)
3949                                 speed += slave->speed;
3950                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
3951                             slave->duplex != DUPLEX_UNKNOWN)
3952                                 ecmd->duplex = slave->duplex;
3953                 }
3954         }
3955         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3956         read_unlock(&bond->lock);
3957
3958         return 0;
3959 }
3960
3961 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3962                                      struct ethtool_drvinfo *drvinfo)
3963 {
3964         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3965         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3966         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3967                  BOND_ABI_VERSION);
3968 }
3969
3970 static const struct ethtool_ops bond_ethtool_ops = {
3971         .get_drvinfo            = bond_ethtool_get_drvinfo,
3972         .get_settings           = bond_ethtool_get_settings,
3973         .get_link               = ethtool_op_get_link,
3974 };
3975
3976 static const struct net_device_ops bond_netdev_ops = {
3977         .ndo_init               = bond_init,
3978         .ndo_uninit             = bond_uninit,
3979         .ndo_open               = bond_open,
3980         .ndo_stop               = bond_close,
3981         .ndo_start_xmit         = bond_start_xmit,
3982         .ndo_select_queue       = bond_select_queue,
3983         .ndo_get_stats64        = bond_get_stats,
3984         .ndo_do_ioctl           = bond_do_ioctl,
3985         .ndo_change_rx_flags    = bond_change_rx_flags,
3986         .ndo_set_rx_mode        = bond_set_rx_mode,
3987         .ndo_change_mtu         = bond_change_mtu,
3988         .ndo_set_mac_address    = bond_set_mac_address,
3989         .ndo_neigh_setup        = bond_neigh_setup,
3990         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
3991         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
3992 #ifdef CONFIG_NET_POLL_CONTROLLER
3993         .ndo_netpoll_setup      = bond_netpoll_setup,
3994         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
3995         .ndo_poll_controller    = bond_poll_controller,
3996 #endif
3997         .ndo_add_slave          = bond_enslave,
3998         .ndo_del_slave          = bond_release,
3999         .ndo_fix_features       = bond_fix_features,
4000 };
4001
4002 static const struct device_type bond_type = {
4003         .name = "bond",
4004 };
4005
4006 static void bond_destructor(struct net_device *bond_dev)
4007 {
4008         struct bonding *bond = netdev_priv(bond_dev);
4009         if (bond->wq)
4010                 destroy_workqueue(bond->wq);
4011         free_netdev(bond_dev);
4012 }
4013
4014 static void bond_setup(struct net_device *bond_dev)
4015 {
4016         struct bonding *bond = netdev_priv(bond_dev);
4017
4018         /* initialize rwlocks */
4019         rwlock_init(&bond->lock);
4020         rwlock_init(&bond->curr_slave_lock);
4021         bond->params = bonding_defaults;
4022
4023         /* Initialize pointers */
4024         bond->dev = bond_dev;
4025
4026         /* Initialize the device entry points */
4027         ether_setup(bond_dev);
4028         bond_dev->netdev_ops = &bond_netdev_ops;
4029         bond_dev->ethtool_ops = &bond_ethtool_ops;
4030         bond_set_mode_ops(bond, bond->params.mode);
4031
4032         bond_dev->destructor = bond_destructor;
4033
4034         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4035
4036         /* Initialize the device options */
4037         bond_dev->tx_queue_len = 0;
4038         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4039         bond_dev->priv_flags |= IFF_BONDING;
4040         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4041
4042         /* At first, we block adding VLANs. That's the only way to
4043          * prevent problems that occur when adding VLANs over an
4044          * empty bond. The block will be removed once non-challenged
4045          * slaves are enslaved.
4046          */
4047         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4048
4049         /* don't acquire bond device's netif_tx_lock when
4050          * transmitting */
4051         bond_dev->features |= NETIF_F_LLTX;
4052
4053         /* By default, we declare the bond to be fully
4054          * VLAN hardware accelerated capable. Special
4055          * care is taken in the various xmit functions
4056          * when there are slaves that are not hw accel
4057          * capable
4058          */
4059
4060         bond_dev->hw_features = BOND_VLAN_FEATURES |
4061                                 NETIF_F_HW_VLAN_CTAG_TX |
4062                                 NETIF_F_HW_VLAN_CTAG_RX |
4063                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4064
4065         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4066         bond_dev->features |= bond_dev->hw_features;
4067 }
4068
4069 /*
4070 * Destroy a bonding device.
4071 * Must be under rtnl_lock when this function is called.
4072 */
4073 static void bond_uninit(struct net_device *bond_dev)
4074 {
4075         struct bonding *bond = netdev_priv(bond_dev);
4076         struct list_head *iter;
4077         struct slave *slave;
4078
4079         bond_netpoll_cleanup(bond_dev);
4080
4081         /* Release the bonded slaves */
4082         bond_for_each_slave(bond, slave, iter)
4083                 __bond_release_one(bond_dev, slave->dev, true);
4084         pr_info("%s: released all slaves\n", bond_dev->name);
4085
4086         list_del(&bond->bond_list);
4087
4088         bond_debug_unregister(bond);
4089 }
4090
4091 /*------------------------- Module initialization ---------------------------*/
4092
4093 /*
4094  * Convert string input module parms.  Accept either the
4095  * number of the mode or its string name.  A bit complicated because
4096  * some mode names are substrings of other names, and calls from sysfs
4097  * may have whitespace in the name (trailing newlines, for example).
4098  */
4099 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4100 {
4101         int modeint = -1, i, rv;
4102         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4103
4104         for (p = (char *)buf; *p; p++)
4105                 if (!(isdigit(*p) || isspace(*p)))
4106                         break;
4107
4108         if (*p)
4109                 rv = sscanf(buf, "%20s", modestr);
4110         else
4111                 rv = sscanf(buf, "%d", &modeint);
4112
4113         if (!rv)
4114                 return -1;
4115
4116         for (i = 0; tbl[i].modename; i++) {
4117                 if (modeint == tbl[i].mode)
4118                         return tbl[i].mode;
4119                 if (strcmp(modestr, tbl[i].modename) == 0)
4120                         return tbl[i].mode;
4121         }
4122
4123         return -1;
4124 }
4125
4126 static int bond_check_params(struct bond_params *params)
4127 {
4128         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4129         int arp_all_targets_value;
4130
4131         /*
4132          * Convert string parameters.
4133          */
4134         if (mode) {
4135                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4136                 if (bond_mode == -1) {
4137                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4138                                mode == NULL ? "NULL" : mode);
4139                         return -EINVAL;
4140                 }
4141         }
4142
4143         if (xmit_hash_policy) {
4144                 if ((bond_mode != BOND_MODE_XOR) &&
4145                     (bond_mode != BOND_MODE_8023AD)) {
4146                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4147                                bond_mode_name(bond_mode));
4148                 } else {
4149                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4150                                                         xmit_hashtype_tbl);
4151                         if (xmit_hashtype == -1) {
4152                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4153                                        xmit_hash_policy == NULL ? "NULL" :
4154                                        xmit_hash_policy);
4155                                 return -EINVAL;
4156                         }
4157                 }
4158         }
4159
4160         if (lacp_rate) {
4161                 if (bond_mode != BOND_MODE_8023AD) {
4162                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4163                                 bond_mode_name(bond_mode));
4164                 } else {
4165                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4166                         if (lacp_fast == -1) {
4167                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4168                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4169                                 return -EINVAL;
4170                         }
4171                 }
4172         }
4173
4174         if (ad_select) {
4175                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4176                 if (params->ad_select == -1) {
4177                         pr_err("Error: Invalid ad_select \"%s\"\n",
4178                                ad_select == NULL ? "NULL" : ad_select);
4179                         return -EINVAL;
4180                 }
4181
4182                 if (bond_mode != BOND_MODE_8023AD) {
4183                         pr_warning("ad_select param only affects 802.3ad mode\n");
4184                 }
4185         } else {
4186                 params->ad_select = BOND_AD_STABLE;
4187         }
4188
4189         if (max_bonds < 0) {
4190                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4191                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4192                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4193         }
4194
4195         if (miimon < 0) {
4196                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4197                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4198                 miimon = BOND_LINK_MON_INTERV;
4199         }
4200
4201         if (updelay < 0) {
4202                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4203                            updelay, INT_MAX);
4204                 updelay = 0;
4205         }
4206
4207         if (downdelay < 0) {
4208                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4209                            downdelay, INT_MAX);
4210                 downdelay = 0;
4211         }
4212
4213         if ((use_carrier != 0) && (use_carrier != 1)) {
4214                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4215                            use_carrier);
4216                 use_carrier = 1;
4217         }
4218
4219         if (num_peer_notif < 0 || num_peer_notif > 255) {
4220                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4221                            num_peer_notif);
4222                 num_peer_notif = 1;
4223         }
4224
4225         /* reset values for 802.3ad */
4226         if (bond_mode == BOND_MODE_8023AD) {
4227                 if (!miimon) {
4228                         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");
4229                         pr_warning("Forcing miimon to 100msec\n");
4230                         miimon = 100;
4231                 }
4232         }
4233
4234         if (tx_queues < 1 || tx_queues > 255) {
4235                 pr_warning("Warning: tx_queues (%d) should be between "
4236                            "1 and 255, resetting to %d\n",
4237                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4238                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4239         }
4240
4241         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4242                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4243                            "not of valid value (0/1), so it was set to "
4244                            "0\n", all_slaves_active);
4245                 all_slaves_active = 0;
4246         }
4247
4248         if (resend_igmp < 0 || resend_igmp > 255) {
4249                 pr_warning("Warning: resend_igmp (%d) should be between "
4250                            "0 and 255, resetting to %d\n",
4251                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4252                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4253         }
4254
4255         /* reset values for TLB/ALB */
4256         if ((bond_mode == BOND_MODE_TLB) ||
4257             (bond_mode == BOND_MODE_ALB)) {
4258                 if (!miimon) {
4259                         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");
4260                         pr_warning("Forcing miimon to 100msec\n");
4261                         miimon = 100;
4262                 }
4263         }
4264
4265         if (bond_mode == BOND_MODE_ALB) {
4266                 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",
4267                           updelay);
4268         }
4269
4270         if (!miimon) {
4271                 if (updelay || downdelay) {
4272                         /* just warn the user the up/down delay will have
4273                          * no effect since miimon is zero...
4274                          */
4275                         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",
4276                                    updelay, downdelay);
4277                 }
4278         } else {
4279                 /* don't allow arp monitoring */
4280                 if (arp_interval) {
4281                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4282                                    miimon, arp_interval);
4283                         arp_interval = 0;
4284                 }
4285
4286                 if ((updelay % miimon) != 0) {
4287                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4288                                    updelay, miimon,
4289                                    (updelay / miimon) * miimon);
4290                 }
4291
4292                 updelay /= miimon;
4293
4294                 if ((downdelay % miimon) != 0) {
4295                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4296                                    downdelay, miimon,
4297                                    (downdelay / miimon) * miimon);
4298                 }
4299
4300                 downdelay /= miimon;
4301         }
4302
4303         if (arp_interval < 0) {
4304                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4305                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4306                 arp_interval = BOND_LINK_ARP_INTERV;
4307         }
4308
4309         for (arp_ip_count = 0, i = 0;
4310              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4311                 /* not complete check, but should be good enough to
4312                    catch mistakes */
4313                 __be32 ip = in_aton(arp_ip_target[i]);
4314                 if (!isdigit(arp_ip_target[i][0]) || ip == 0 ||
4315                     ip == htonl(INADDR_BROADCAST)) {
4316                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4317                                    arp_ip_target[i]);
4318                         arp_interval = 0;
4319                 } else {
4320                         if (bond_get_targets_ip(arp_target, ip) == -1)
4321                                 arp_target[arp_ip_count++] = ip;
4322                         else
4323                                 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4324                                            &ip);
4325                 }
4326         }
4327
4328         if (arp_interval && !arp_ip_count) {
4329                 /* don't allow arping if no arp_ip_target given... */
4330                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4331                            arp_interval);
4332                 arp_interval = 0;
4333         }
4334
4335         if (arp_validate) {
4336                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4337                         pr_err("arp_validate only supported in active-backup mode\n");
4338                         return -EINVAL;
4339                 }
4340                 if (!arp_interval) {
4341                         pr_err("arp_validate requires arp_interval\n");
4342                         return -EINVAL;
4343                 }
4344
4345                 arp_validate_value = bond_parse_parm(arp_validate,
4346                                                      arp_validate_tbl);
4347                 if (arp_validate_value == -1) {
4348                         pr_err("Error: invalid arp_validate \"%s\"\n",
4349                                arp_validate == NULL ? "NULL" : arp_validate);
4350                         return -EINVAL;
4351                 }
4352         } else
4353                 arp_validate_value = 0;
4354
4355         arp_all_targets_value = 0;
4356         if (arp_all_targets) {
4357                 arp_all_targets_value = bond_parse_parm(arp_all_targets,
4358                                                         arp_all_targets_tbl);
4359
4360                 if (arp_all_targets_value == -1) {
4361                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4362                                arp_all_targets);
4363                         arp_all_targets_value = 0;
4364                 }
4365         }
4366
4367         if (miimon) {
4368                 pr_info("MII link monitoring set to %d ms\n", miimon);
4369         } else if (arp_interval) {
4370                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4371                         arp_interval,
4372                         arp_validate_tbl[arp_validate_value].modename,
4373                         arp_ip_count);
4374
4375                 for (i = 0; i < arp_ip_count; i++)
4376                         pr_info(" %s", arp_ip_target[i]);
4377
4378                 pr_info("\n");
4379
4380         } else if (max_bonds) {
4381                 /* miimon and arp_interval not set, we need one so things
4382                  * work as expected, see bonding.txt for details
4383                  */
4384                 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");
4385         }
4386
4387         if (primary && !USES_PRIMARY(bond_mode)) {
4388                 /* currently, using a primary only makes sense
4389                  * in active backup, TLB or ALB modes
4390                  */
4391                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4392                            primary, bond_mode_name(bond_mode));
4393                 primary = NULL;
4394         }
4395
4396         if (primary && primary_reselect) {
4397                 primary_reselect_value = bond_parse_parm(primary_reselect,
4398                                                          pri_reselect_tbl);
4399                 if (primary_reselect_value == -1) {
4400                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4401                                primary_reselect ==
4402                                         NULL ? "NULL" : primary_reselect);
4403                         return -EINVAL;
4404                 }
4405         } else {
4406                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4407         }
4408
4409         if (fail_over_mac) {
4410                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4411                                                       fail_over_mac_tbl);
4412                 if (fail_over_mac_value == -1) {
4413                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4414                                arp_validate == NULL ? "NULL" : arp_validate);
4415                         return -EINVAL;
4416                 }
4417
4418                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4419                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4420         } else {
4421                 fail_over_mac_value = BOND_FOM_NONE;
4422         }
4423
4424         /* fill params struct with the proper values */
4425         params->mode = bond_mode;
4426         params->xmit_policy = xmit_hashtype;
4427         params->miimon = miimon;
4428         params->num_peer_notif = num_peer_notif;
4429         params->arp_interval = arp_interval;
4430         params->arp_validate = arp_validate_value;
4431         params->arp_all_targets = arp_all_targets_value;
4432         params->updelay = updelay;
4433         params->downdelay = downdelay;
4434         params->use_carrier = use_carrier;
4435         params->lacp_fast = lacp_fast;
4436         params->primary[0] = 0;
4437         params->primary_reselect = primary_reselect_value;
4438         params->fail_over_mac = fail_over_mac_value;
4439         params->tx_queues = tx_queues;
4440         params->all_slaves_active = all_slaves_active;
4441         params->resend_igmp = resend_igmp;
4442         params->min_links = min_links;
4443         params->lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4444
4445         if (primary) {
4446                 strncpy(params->primary, primary, IFNAMSIZ);
4447                 params->primary[IFNAMSIZ - 1] = 0;
4448         }
4449
4450         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4451
4452         return 0;
4453 }
4454
4455 static struct lock_class_key bonding_netdev_xmit_lock_key;
4456 static struct lock_class_key bonding_netdev_addr_lock_key;
4457 static struct lock_class_key bonding_tx_busylock_key;
4458
4459 static void bond_set_lockdep_class_one(struct net_device *dev,
4460                                        struct netdev_queue *txq,
4461                                        void *_unused)
4462 {
4463         lockdep_set_class(&txq->_xmit_lock,
4464                           &bonding_netdev_xmit_lock_key);
4465 }
4466
4467 static void bond_set_lockdep_class(struct net_device *dev)
4468 {
4469         lockdep_set_class(&dev->addr_list_lock,
4470                           &bonding_netdev_addr_lock_key);
4471         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4472         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4473 }
4474
4475 /*
4476  * Called from registration process
4477  */
4478 static int bond_init(struct net_device *bond_dev)
4479 {
4480         struct bonding *bond = netdev_priv(bond_dev);
4481         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4482         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4483
4484         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4485
4486         /*
4487          * Initialize locks that may be required during
4488          * en/deslave operations.  All of the bond_open work
4489          * (of which this is part) should really be moved to
4490          * a phase prior to dev_open
4491          */
4492         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4493         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4494
4495         bond->wq = create_singlethread_workqueue(bond_dev->name);
4496         if (!bond->wq)
4497                 return -ENOMEM;
4498
4499         bond_set_lockdep_class(bond_dev);
4500
4501         list_add_tail(&bond->bond_list, &bn->dev_list);
4502
4503         bond_prepare_sysfs_group(bond);
4504
4505         bond_debug_register(bond);
4506
4507         /* Ensure valid dev_addr */
4508         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4509             bond_dev->addr_assign_type == NET_ADDR_PERM)
4510                 eth_hw_addr_random(bond_dev);
4511
4512         return 0;
4513 }
4514
4515 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4516 {
4517         if (tb[IFLA_ADDRESS]) {
4518                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4519                         return -EINVAL;
4520                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4521                         return -EADDRNOTAVAIL;
4522         }
4523         return 0;
4524 }
4525
4526 static unsigned int bond_get_num_tx_queues(void)
4527 {
4528         return tx_queues;
4529 }
4530
4531 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4532         .kind                   = "bond",
4533         .priv_size              = sizeof(struct bonding),
4534         .setup                  = bond_setup,
4535         .validate               = bond_validate,
4536         .get_num_tx_queues      = bond_get_num_tx_queues,
4537         .get_num_rx_queues      = bond_get_num_tx_queues, /* Use the same number
4538                                                              as for TX queues */
4539 };
4540
4541 /* Create a new bond based on the specified name and bonding parameters.
4542  * If name is NULL, obtain a suitable "bond%d" name for us.
4543  * Caller must NOT hold rtnl_lock; we need to release it here before we
4544  * set up our sysfs entries.
4545  */
4546 int bond_create(struct net *net, const char *name)
4547 {
4548         struct net_device *bond_dev;
4549         int res;
4550
4551         rtnl_lock();
4552
4553         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4554                                    name ? name : "bond%d",
4555                                    bond_setup, tx_queues);
4556         if (!bond_dev) {
4557                 pr_err("%s: eek! can't alloc netdev!\n", name);
4558                 rtnl_unlock();
4559                 return -ENOMEM;
4560         }
4561
4562         dev_net_set(bond_dev, net);
4563         bond_dev->rtnl_link_ops = &bond_link_ops;
4564
4565         res = register_netdevice(bond_dev);
4566
4567         netif_carrier_off(bond_dev);
4568
4569         rtnl_unlock();
4570         if (res < 0)
4571                 bond_destructor(bond_dev);
4572         return res;
4573 }
4574
4575 static int __net_init bond_net_init(struct net *net)
4576 {
4577         struct bond_net *bn = net_generic(net, bond_net_id);
4578
4579         bn->net = net;
4580         INIT_LIST_HEAD(&bn->dev_list);
4581
4582         bond_create_proc_dir(bn);
4583         bond_create_sysfs(bn);
4584
4585         return 0;
4586 }
4587
4588 static void __net_exit bond_net_exit(struct net *net)
4589 {
4590         struct bond_net *bn = net_generic(net, bond_net_id);
4591         struct bonding *bond, *tmp_bond;
4592         LIST_HEAD(list);
4593
4594         bond_destroy_sysfs(bn);
4595         bond_destroy_proc_dir(bn);
4596
4597         /* Kill off any bonds created after unregistering bond rtnl ops */
4598         rtnl_lock();
4599         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4600                 unregister_netdevice_queue(bond->dev, &list);
4601         unregister_netdevice_many(&list);
4602         rtnl_unlock();
4603 }
4604
4605 static struct pernet_operations bond_net_ops = {
4606         .init = bond_net_init,
4607         .exit = bond_net_exit,
4608         .id   = &bond_net_id,
4609         .size = sizeof(struct bond_net),
4610 };
4611
4612 static int __init bonding_init(void)
4613 {
4614         int i;
4615         int res;
4616
4617         pr_info("%s", bond_version);
4618
4619         res = bond_check_params(&bonding_defaults);
4620         if (res)
4621                 goto out;
4622
4623         res = register_pernet_subsys(&bond_net_ops);
4624         if (res)
4625                 goto out;
4626
4627         res = rtnl_link_register(&bond_link_ops);
4628         if (res)
4629                 goto err_link;
4630
4631         bond_create_debugfs();
4632
4633         for (i = 0; i < max_bonds; i++) {
4634                 res = bond_create(&init_net, NULL);
4635                 if (res)
4636                         goto err;
4637         }
4638
4639         register_netdevice_notifier(&bond_netdev_notifier);
4640 out:
4641         return res;
4642 err:
4643         rtnl_link_unregister(&bond_link_ops);
4644 err_link:
4645         unregister_pernet_subsys(&bond_net_ops);
4646         goto out;
4647
4648 }
4649
4650 static void __exit bonding_exit(void)
4651 {
4652         unregister_netdevice_notifier(&bond_netdev_notifier);
4653
4654         bond_destroy_debugfs();
4655
4656         rtnl_link_unregister(&bond_link_ops);
4657         unregister_pernet_subsys(&bond_net_ops);
4658
4659 #ifdef CONFIG_NET_POLL_CONTROLLER
4660         /*
4661          * Make sure we don't have an imbalance on our netpoll blocking
4662          */
4663         WARN_ON(atomic_read(&netpoll_block_tx));
4664 #endif
4665 }
4666
4667 module_init(bonding_init);
4668 module_exit(bonding_exit);
4669 MODULE_LICENSE("GPL");
4670 MODULE_VERSION(DRV_VERSION);
4671 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4672 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4673 MODULE_ALIAS_RTNL_LINK("bond");